How to Simulate Consciousness

In the previous chapters, I described the theoretical basis for the invention discussed in this book, the prototype consciousness simulator I have been designing and building in a specific object-oriented programming environment to help work out and test many of the ideas used in the invention, and how it is possible to use such an invention to simulate or mimic some animal and human conscious behaviors.

One purpose of this chapter is to help all readers to integrate these ideas by describing how to reduce the invention to practice; in other words, by describing how to build a computer simulation system capable of simulating rational self-consciousness, including simulated volition, or "free will" as it is also called.

Another purpose of this chapter is to make the description of the simulation system intelligible to someone skilled in the art of computer programming, and do so in terms general enough to be applicable to any object-oriented programming environment.

That being said, let us review what current state of the art computer systems are and see how they differ from what I am proposing in my description.

First, it must be stated clearly and unequivocally that a computer system can never be alive or conscious in the exact same sense as a biological life-form can. However, a carefully designed computer simulation system can imitate or mimic the processes of life and consciousness to some degree (just as a mannequin imitates or mimics the human form), and such imitations will improve as computer technology becomes more powerful.

However, simulated consciousness is still simulated; a computer simulation system is not conscious the way animals and human beings are conscious; it merely imitates some of their actions to make a better human interface.

All computer systems are electronic machines built by human beings, machines that manipulate binary bits for human goals according to human logical rules specified in the computers' automatic programs. They are neither alive, nor conscious, and they certainly do not have "free will."

Strictly speaking, all computer systems do is manipulate binary bits in the form of electrical capacitances; they do not add, do not do word processing, do not process information, do not sort records in databases, and so on. Only people do all these things, and it is only in reference to a human mind, a conscious human mind that the terms such as information and knowledge can apply; it is the human mind that sets the context for these terms.₁

In addition, all computer systems must first be programmed in order to function at all; that is, every action the computer's processor is to execute must be described in a suitable programming language in exacting detail by a human programmer. This is why the programs can run by themselves in the first place, why they can be called automatic, as opposed to requiring human intervention at every step, that is, as opposed to being manually operated. Any breaks in the program definition are an action that is undefined and will cause a computer to "crash," with absolute certainty, if the undefined section of computer code (also known as a "bug") is encountered by the processor. Even the so-called "self-programming" capabilities of genetic algorithms and neural network computer systems (Holland) rely on specific programs to pre-define their basic functionality; only after these programs are running properly can they interact with their environment to automatically modify themselves from a program that is too general to perform any task except change its own parameters, to one that does some specific task that is defined by the data the program inputs.

The basic problems of all attempts to design computer systems capable of Artificial Intelligence (AI) (McDermott) or Artificial Life (AL) (Pattie Maes at MIT and Los Alamos laboratories) is that computers are not alive, not conscious, and they require a programmer to specify in advance not only what actions the computer system will perform, but also where, when, and how it will effect the actions. By contrast, the intelligent behavior of biological life-forms is largely self-defining and is so without the requirement of human intervention. The mechanistic, pre-definition of action, therefore, is at the same time the reason computer systems can run automatically, and their downfall as life and consciousness simulators; it is their downfall because it is impossible (a contradiction) to attempt to pre-define the actions of a self-defining system, yet that is exactly what some in the field of AI have attempted to do.b

Note - This statement does not mean that no part of a self-defining system can be defined by human programmers, an idea which implies the system must completely recapitulate evolution. The statement simply means that the self-defining aspects of such a system must be designed in a way that the system has the control and energy to define itself at some level, and that could mean using a set of basic actions that have been programmed by human beings.

In fact, it is precisely the attributes of consciousness and volition, or "free will," that enable human beings to perform manual tasks at all, such as to invent and build computers or to write their code in the first place. This is possible precisely because the human beings that AI and AL computer systems are supposed to emulate are not automatic: Computer programmers are not another kind of computer program; they are people with free will who are capable of performing manual tasks! The behaviors of the human programmers who write the automatic programs that computer systems run do not have all their behaviors pre-defined. The computer programs they write may be automatic, but they are written manually.

How does one break this paradox? How does one design a computer system that is not automatic?

Genetic algorithms attempt to get around this paradox by emulating or recapitulating evolution in the hope that eventually intelligence will emerge; neural networks attempt to do something similar by emulating the function of neurons in higher life-forms. It is possible that these methods working together and with help from programmers to adjust their operation from time to time could, eventually, produce a simulated life-form that could eventually simulate consciousness and volition; but if the history of life and human evolution is any indication of how long such a process takes, it will undoubtedly take a long, long time before we see any results. For example, in the Blue Genes project at IBM®, one protein folding took a year of processing time to simulate with a petaflop computer that was 1000 times better than the average super computers in the year 2000.

The invention described in detail in this chapter makes it possible to build a simulated life-form with the attributes of simulated consciousness and volition in a much shorter time, probably in two to three years of focused effort by a team of two or three programmers using off-the-shelf computer hardware and an object-oriented programming environment. Of the entire development time, approximately a third of it would be required to write the basic simulation system code; the balance of the development time would be used to train and refine the system by helping it to learn and understand the world it perceives and improving its program code and operation. The amount of data to be learned by a consciousness simulation system of this design is large, but not unmanageable; the objects and relationships found in a typical pre-school play room would be sufficient to get the system started and for it to learn basic percepts, concepts, and other aspects of reality.

One reason for the relatively short development time is the fact that this invention copies many design ideas from real life-forms instead of attempting to re-evolve them to recapitulate evolution in some manner as genetic algorithms do. In other words, just as the AL researchers did not re-evolve the gait of their insect robots, but reverse engineered their operation by copying real life-forms, so this invention seeks to reverse engineer the simulation of goal-directed behavior and the processes of consciousness rather than re-evolve them.

Another reason is that this invention solves the pre-definition of action problem by programming a simulated life-form with attributes similar to those that enable biological life-forms to define their own teleological actions. The invention breaks the paradox mentioned above by using the more complex form of causality found in biological life-forms, a form of causality that works by moving both the energy source for the life-form and its control into the simulated life-form itself. With the locus of control and its own energy source, the potential exists for action that is independent of either direct or indirect human control. This form of goal-directed action is different from current state of the art robotic or computer agent action as it operates in extant systems, and the differences will be explained shortly in detail.

Consider the following thought experiment crafted by Dr. Harry Binswanger to differentiate the actions of living from non-living objects: An ice cube and an earthworm are put on slanted boards and over each hangs a heat lamp. Both the ice cube and the worm will move away from the heat, but the cause of the ice cube's action is external and mechanical, whereas the cause of the worm's action is internal and teleological. If the experiment is repeated with the boards flat instead of slanted, the ice cube will not move and simply melt because it has neither life nor the capacity to preserve it; the worm, however, moves on its own internally controlled power because the heat is inimical to its life, and its goal is to survive.₂

I will explain how teleology or goal-directed behavior can be simulated in more detail below; for now, consider that this approach enables a new kind of behavior that is not automatic in the usual computer or mechanical sense in which this term is applied to state of the art computer systems (that is, pre-programmed, in the sense of an automaton), but automatic from the perspective of life-forms and a life simulation system, a system that behaves like an earthworm rather than an ice cube.

However, simulating teleological behavior does not preclude the predefinition of some of a simulated life-form's actions. Like state of the art genetic algorithms and neural networks, this invention is partially pre-defined and partially emergent, but the design is also different from current state of the art ideas. By pre-defining a system with an internal energy source and internal control like a life-form, I have essentially copied the logic of the causality of life that has taken billions of years to evolve in order to cause the behavior of an earthworm. I have applied a strategy similar to that which some in the field of AL have used to very good results in the development of robots that simulate the capacity of insects to negotiate rough terrain without pre-defined actions, such as some of the cock roach robots.

The difference is that I have applied that strategy to both the cellular level of life and to mental actions of complex life-forms to simulate their consciousness. How I have done this will become apparent as I describe the invention in detail.

The new behavior, along with some new data types, set the stage for the eventual emergence of simulated self-consciousness and simulated volitional behavior by that simulated consciousness. In other words, the invention solves the previously unsolved problem of needing to specify all of an AI or AL system's actions in advance by permitting all actions and then using an automatic means of eliminating the unwanted actions, where automatic here means in the biological sense, not in the sense of a computer automaton.

The essential ideas of the invention are presented, described, and explained based on the following key points:

• The problem of action pre-definition is solved by using goal-directed behavior: Goal-directed behavior is how all life-forms stay alive; it is required by the conditional nature of life processes and is an automatic means of limiting behavior without specifying what behaviors are permitted or when they are permitted; energy and action control reside inside the life-form, and actions are controlled by simulating pleasure and pain, with the digital life-form's own "life" as the standard of action. This method of control is based on a complex form of causality: In order to act, a simulated life-form must first survive; unwanted actions, therefore, do not get repeated in the long-term because the life-forms that would have repeated them no longer exist. This form of causality is complex because it involves the issue of the survival (continued existence) of the acting object, as opposed to the simpler form of mechanistic causality in which there is no issue of the acting object's survival (because non-living objects exist unconditionally).

Note - At the risk of being redundant, I want to re-emphasize two important points: First, the fact that goal-directed behavior is the means this invention uses to solve the action pre-definition problem does not mean human programmers cannot pre-define basic actions such as Look, Find, or Eat to build a starter simulation system; goal-directed action refers to actions (or sequences of basic actions) selected by a life-form for survival purposes; it does not mean the recapitulation of the evolution of actions from no actions. Second, in biological systems, automatic goal-directed behavior is still goal-directed, not automatic in the sense of computer or robot automation, which operates on mechanistic causality.

• Intelligent action presupposes consciousness: Awareness of the world outside a digital life-form is made possible by simulating consciousness. Simulated consciousness begins with the simulated perception of objects, and these perceptions provide a processing unit economy (or content-oriented form of data compression) that has a survival advantage to digital life-forms; the need to process fewer units saves processing cycles (energy and time), properties of importance to a life-form whether real or simulated. The perceptual form of simulated consciousness is automatic (in the biological sense), partially pre-defined, and is programmed in the conventional way; the processes that produce the content for simulated perceptual consciousness are not modifiable by a simulated life-form, only the sequencing of its actions in reality are. This statement goes back to the earlier explanation that this invention is partially predefined and partially self-defining: How the simulation system described herein senses the world, resolves objects in the world, and forms its percepts of them by extracting their attributes is pre-programmed; what objects the system "looks at or focuses on," the content of its perceptions and how it reacts to those percepts (which actions in its basic action set it selects and in what order), it defines for itself. This simulation cannot reprogram its perceptual processes, but it can reprogram how it makes use of those processes to aid its own survival by the action sequences it selects.

• Volition (free will) implies the ability of self-regulation: Another level of control is possible for a digital life-form by making some of the processes of simulated consciousness self-modifiable, unlike the perceptual processes just described. This enables a digital life-form to change the way in which the content of its simulated consciousness is stored in memory and to have simulated volitional control over its memories by organizing them with a new data type called a concept. Concepts (when formed by a process defined by Ayn Rand, see note below) are symbolized by natural language words, and the specific way in which concepts are calculated define their meaning and the digital life-form's simulated abstract view of reality; concepts also set the stage for natural language to emerge in a DLF in the mid to latter stages of development. The emergence of natural language will occur approximately half way through the 2-3 year development process for the simulation system, since it depends on the system having formed (calculated) several hundreds to perhaps a thousand conceptual chains; this is necessary because it is the conceptual chains that the system uses to calculate the meaning of the natural language words that symbolize its concepts. The details of this process are explained later in this chapter.

• The need to process fewer data units equals a survival advantage: An additional survival advantage for a digital life-form comes from simulated conceptual consciousness due to the time independence and even greater processing unit economy attributes of concepts, and the fact that concepts can represent in conscious awareness normally invisible phenomena such as relationships between perceived objects; this fact makes possible a stable, unitary world view for a digital life-form, a world view that is in the form of abstract symbolic information in addition to percepts of specific objects (though every abstract is connected to specific percepts via chains of "manual" calculations).

• Concepts make simulated self-consciousness possible: The consequences of the self-modifying nature of simulated volitional control over memory at the conceptual level in a simulated consciousness of a simulated life-form enables the emergence of a concept of "self" for that digital life-form. Once formed, the concept of "self" enables a simulated rational self-consciousness to emerge, as well as the use of natural language in the form of simple sentences. A digital life-form can then use natural language primarily as a means of gaining further knowledge and, secondarily, for communication with human beings and other digital life-forms; this greatly improves the system's interface. As with the description of simulated volition above, these attributes will emerge in the mid to latter parts of the 2-3 year system development process, and for the same reason: Simulated natural language and self-consciousness depend on chains of conceptual calculations to operate, so they cannot emerge until the system has formed those concepts and can follow the calculation chains of the concepts to use the knowledge of reality the concepts contain (including knowledge of "self").

Note - The term concept as used in this description means only concepts as defined by the method described in the book Introduction to Objectivist Epistemology by Ayn Rand (see references). None of the common uses of the term "concept" that are found in philosophy, science, or the field of AI, such as how the term is used by Lenat, Shank, Collins & Stevens, and so on apply to this description; a more detailed explanation of what this means follows later.

The details of the ideas introduced so far and how to reduce them to practice is what the description in the balance of this chapter contains. Once these basic ideas contained in this description of the invention are learned and integrated by an expert programmer, that programmer will be able to design and build a digital life-form which can simulate rational self-consciousness. The simulation system can be created by the following steps:

If done correctly, when these steps have been completed, the capacities of simulated rational, self-consciousness and volition will have emerged in the digital life-form in the mid to latter part of the 2-3 year development process I have been describing. Moreover, it cannot be emphasized enough that in order to be successful, the simulation system must perceive objects in their natural relationships in the world to be able to simulate human consciousness in order to form concepts of those objects and relationships that are similar to the corresponding human concepts.

While this process may be started with a simulated world such as a game world, if the digital life-form is to be conscious of and function in the human world, then it must ultimately perceive in, act in, and form concepts of the same world we humans perceive, act in, and form concepts of, and it must do so directly (for itself).

Before I can go into the specifics of how to implement the steps listed above, the ideas that were described and explained in detail in the first four chapters of this book must be reviewed to set the proper context. This invention's reduction to practice depends on those ideas, many of which are new to the state of the art and therefore must be reviewed for even the most experienced computer programmers.

5.1.1 A Few Prerequisite Ideas

As explained above, computer systems are man-made machines that are neither alive nor conscious and can never be so in the exact same sense as biological life-forms.

Moreover, there is a huge difference in the ways in which computer systems behave as they are designed to operate in the current state of the art and the way life-forms operate. In computer systems, actions are programmed, which means they are caused and regulated by humans and are therefore very predictable, whereas life-forms are independent entities and much less predictable.

Note - To whatever degree computer systems are unpredictable at all, it is only because of pre-programmed randomness (such as from the use of random numbers), or that some systems are so complex that human consciousness cannot process all the possibilities in order to predict what all of causal consequences of the programs will be as they are run. (However, unpredictability due to this latter limitation does not imply that state of the art computer systems operate by something other than simple, mechanistic causality.)

Computer systems are action platforms that are capable of an endless variety of potential actions; they are like a blank sheet of paper, which is a "platform" on which an endless variety of stories can be written. But computer systems are like "causal paper," which instead of merely describing reality like ordinary paper, can animate aspects of reality in both actual and virtual forms.

Computer programs are what limit the endless potential actions (causes) computer systems are capable of, limit their actions to some set of specific actions: The actions that constitute a specific computer program. In current state of art systems, actions are limited by the programmer by pre-defining them in exacting detail when a program is written.

The actions of life-forms, on the other hand, are self-caused and self-regulating like the earthworm in Dr. Binswanger's thought experiment. Only certain specific actions get repeated, but this does not imply that life-forms are "biological computers," as many people have assumed. While it is true that some actions are pre-defined, biological pre-definition is not the same as computer programming. The alternative is that the actions of life-forms are limited by something other than simple, mechanistic or "billiard ball" causality (though not something supernatural). In order to understand what that "something" might be, it is first necessary to look at the identity of life-forms in more detail, especially the conscious ones.

Ayn Rand identified a metaphysical fact that is implicit in every action of every object: What a thing is, its identity, determines what it can do, its action capacity in reality.₃ This fact is as true for computers as it is for rocks or conscious life-forms: In order to understand how to simulate consciousness, it is first necessary to understand what consciousness is.

Metaphysically, consciousness is a fundamental state of being; it is axiomatic, and it is an attribute of some life-forms, a relationship they have with reality. From a metaphysical point of view, conscious life-forms are in a state of awareness of the world they live in, as opposed to life-forms without consciousness that are not in that state (the latter operate my means of simple sensations, not perceptual consciousness, or awareness of objects).

Note - Axiomatic, as the term is used by Ayn Rand, means implicit in your awareness of some fact, and therefore inescapable. For example, the statements: "Existence Exists," "A is A," and "Consciousness is Conscious" are axiomatic because they are implied and affirmed by any attempt to deny them.₄

Operationally, consciousness is an active process supported by other subconscious processes in living organisms. These subconscious processes are the brain functions of conscious life-forms, and they are all causal processes, as is consciousness itself. Consciousness is the subset of subconscious processes that are being activated at a given time; being activated makes them conscious processes. Both conscious and subconscious processes are brain processes that operate because of the activity of the neurons of which the brain is made.₅

There is evidence that the encoding of new memories may be a physical process, that at least part of the cycle of consciousness is caused or at least supported by new neurons growing in a life-form's brain.₆

Consciousness is a limited process with a specific identity (consisting of several subprocesses), just like any other attribute of a life-form is something specific; for example, consciousness has properties and values as do the attributes of size, number of limbs, and the process of digestion; this is how consciousness can be a causal process: It is made of the same "stuff" as the rest of the life-form, so it can interact with the rest of the life-form. The identity of conscious processes interact with the identities of objects and other processes in a living body just like any other life processes do, so there is no "mind-body" dichotomy. However, the result of conscious processes is awareness, instead of, say, the nutrition of digested food. Unlike digestion, consciousness is an pro-active process, but more on that later.

Perceptual consciousness in biological life-forms is a process of sensing reality, integrating sensations into percepts by isolating objects and extracting their identifying attributes, evaluation, action selection, memory, and action in reality, a process which repeats in an endless cycle (as long as a conscious life-form is alive and awake).₇ In most life-forms that possess it, consciousness is largely an automatic process (in the goal-directed or teleological sense of biology).

The exception is Man, who has the capacity of free will; he is capable of manual conscious behavior.

Rational consciousness is a more complex form of consciousness that is non-automatic (in the biological sense) and is made possible by a special "data structure" called a concept. Concepts are open-ended categories based on similarities observed in reality and symbolized by words; concepts make fully developed human volition and natural language possible. Concepts are also the basis for a virtual entity called the "self," but concepts as defined and used herein are not the same as those commonly used in the field of AI.

Simulated consciousness and its use of concepts as is described for this invention is not "model building" as that term is used in state of the art AI research, which uses a rationalistic method of designing AI systems.

In this invention, percepts are neither little pictures of objects nor mathematical models in the traditional sense, but rather they are identifications, sets of attributes of objects-in-the-world resulting from the calculations to make the properties and values of their attributes explicit, properties and values that are implicit in relationships in sensor outputs. Concepts are not words linked to arbitrary definitions in a database, but rather they are open-ended data structures that embody relationships calculated by a specific process (to be described later) based on the differences and similarities of objects perceived by the system earlier; the result is that in this invention, natural language words mean the entire conceptual calculation chain, including the percepts of objects at its base, not just the concepts' definitions. This is a very different arrangement from that found in extant systems in the current state of the art.

Note - This point cannot be over emphasized: The use of arbitrary definitions for creating "concepts" and "building models" is pervasive and the accepted method in our scientific culture. People do it unthinkingly; in fact, most people do not understand what it means to define a concept objectively, unless objectivity is explained in excruciating detail. The reason for this is that most people have never been exposed to Ayn Rand's formulation of the idea of objectivity, a formulation which is new to epistemology, and has only been in existence since the 1960's.₈

Concepts formed objectively in the manner described later in this chapter make possible complex volitional interaction of the conscious with the subconscious and with reality in verbal form, interaction that involves asking questions and evaluating the answers. For example, stop reading for a moment and ask yourself the following:

In each case, your subconscious provided you an answer (or not) which you can judge as true or false, good or bad, relevant to what you are thinking about, important to your life, and so on, as Dr. Binswanger has pointed out. As an on-going process, this kind of verbal interaction between the conscious and subconscious processes in your mind is called thinking.₉ The simple natural language sentences of the questions are incomplete thoughts, but with the answers, they become complete thoughts.

The "Q and A" above and the process that produced it are neither practical magic nor the work of supernatural spirits; the phenomenon is caused entirely by goal-directed processes in your brain, each with a specific, limited identity. These processes can be simulated using a properly designed computer simulation system to enable them in virtual form.

To build a computer system that simulates life, consciousness, and thinking will require the design of a new kind of computer simulation system that does not exist today, a new system design that is not found in the current state of the art.

Note - When these new systems have been perfected, they will probably no longer be called "computers" or "machines" because they will be recognized as a new kind of man-made system, one that is capable of manual behavior, and one that is yet to be named.

If computer systems are thought of as active, symbolic representations of reality, virtual reality, or reality simulators instead of accounting systems, databases, or word processors, it should become more understandable how computer systems can be designed that simulate the attributes and functions of life-forms to some degree of causal accuracy. Exactly what that degree is can only be determined experimentally after some are built and have been improved upon as new devices normally are. (Not many people imagined in the 1950's that the sophistication found in today's virtual reality systems would be possible only some 45 years later. (Jaron Lanier, VR))

Computer systems can be designed to simulate consciousness for the same reason they can simulate a jet airplane design in flight, the vegetative processes of life-forms, genetic evolution, or an entire eco-system: A good computer-based simulation system is an accurate and dynamic, symbolic reproduction of part of reality. In other words, the virtual entities and their actions in the computer correspond to the real entities and their actions in reality. Such a simulation is an objective formulation by human consciousness of the dynamic relationships between the objects involved in a given aspect of the real world, including their causal relationships, all of which are then translated into the form in which they operate in the computer simulation system.

Computer simulation systems substitute variables, calculations, and logic for real world objects, their relationships, and causality. I call this process causality substitution.

As I pointed out in the Introduction to Chapter 1, it is well known in the state of the art that computer hardware and software can be functionally substituted for each other, that they can be equivalent processes in different forms (physical vs. logical). The idea of causality substitution is only slightly broader in scope; causality substitution includes not only computer hardware and software, but also the complex causality of teleological systems. The key is to duplicate the causal activity of goal-directed behavior using logic and virtual entities, just as state of the art software logic duplicates the mechanistic causality of computer hardware (or airplanes) to make possible the substitution of one form of a given process function for the other, of the virtual and logical for the physical and causal.

In other words, a simulation system is the objects and the relationships of some part of reality represented in a symbolic and a logical form as measurements and calculations, a form that is animated by a computer system as a mechanistic platform; this base or supporting part of the simulation system that embodies the invention described herein is governed by mechanistic causality and pre-defined by conventional object-oriented programming methods.

But the life and consciousness simulation system I am describing is also more than the hardware and software platform that animates it; it is the hardware and software plus the teleological interaction of the simulation system as a whole with reality; it is more than the sum of its parts. The simulation system substitutes for and replaces the real biological processes (the causes and effects) of a life-form with measurements and calculations (in a logical and teleological form) that produce identical (or nearly identical) effects in reality. This is the part of the system that must be teleological and self-defining if it is to successfully imitate life and consciousness.

The key here is to identify the essential elements and program substitutions that need to be made; this is the pre-defined part of the simulation system. This aspect of the design involves identifying the necessary and sufficient set of elements to develop the substitutions for, and then writing the software code for those elements, making sure they can interact with reality teleologically. The self-defining stage of the development of the simulation system is the management and tutoring of those basic elements as they simulate the active processes of life and consciousness, and as they recursively operate on the simulation system itself.

The bottom line here is that to make simulated life-forms, a properly designed computer simulation system is substituted for the physics and chemistry that animates, and is the "platform" for, real life-forms, just as in other simulations, computers are substituted for and replace the physical systems that animate real, non-living objects, such as jet airplanes in flight. But in order to simulate life-forms, the system must also take into account and simulate the teleology, the goal-directed behavior of biological life-forms. This requires more complex software logic than is used in state of the art simulators.

The important point to grasp is that the simulation of a life-form, and especially its attribute of consciousness, is not just a computer program alone. Rather, it is the correct design of a complete system of computer programs and hardware in continuous, teleological interaction with reality itself. The simulation program is only one part of a complex system of mechanistic and teleological causes and effects, just as biological life-forms are. This complex system interacts with reality to digitally reproduce as many of the relationships (both mechanistic and teleological) that exist between biological life-forms and reality as is technically possible.

Note - Obviously, the quality of life-form simulators will get better with improvements in technology according to Moore's Law and other factors, as well as from the experimentation with various simulation system design strategies.

The programs in a life-form simulator must be designed to interact with reality much like real life-forms do. This latter point is important because much of what life-forms are capable of when they are mature comes not from some intrinsic property they have from birth, but from their causal interaction with reality as they live and function in the world, and the changes that occur in them as a result of that interaction. Information processed from such interactions is added to their memories as they learn their environment; the result is changes to the life-forms' identities. And changes to what they are amount to changes in their action capacities for future interaction with their world (changes to what they can do).

The world that life-like simulation systems must interact with is reality; it is the Existence of which all objects, including life-forms, are an integral part.

Existence is the natural, the metaphysical, the primary; it just is in one sense, but in another sense when perceived in the form of objects, it provides the content, the data for consciousness to process, either automatically and infallibly as percepts or by choice and fallibly as concepts.₁₀

The content of consciousness, the content that is its data, is either perceptual or conceptual information. The conceptual information is the epistemological, a man-made, volitional form of information at the conceptual level of awareness; it is what it is partly because of its nature as part of reality (the identity of objects), but also partly because of the choices that were made by the people who formed the concepts, made up the symbols (words) that represent concepts mentally, reasoned out the logic, and invented the grammatical structures that enable the content to be represented in their minds in abstract form as sentences. This fact makes conceptual knowledge fallible: Some of the choices required to form concepts could be mistaken, could be non-objective; they could have been made in error. Conceptual knowledge must therefore be checked against reality to make sure it is correct, that its content and meaning accurately reflect reality. It is the infallible nature of perceptual knowledge that makes it possible to do so.₁₁

Note - As to the validity of sense perception, the idea, for example, that a stick which appears "bent" when half submerged in water shows that percepts are inaccurate and fallible is wrong. This idea drops the context of the identities of light, water, and consciousness, focusing only on the appearance of the stick. But the identities of the light, the water, and the stick interact in a way that causes light passing through water to refract. When observed by human consciousness in that form (including the entire context), the stick is bent because of the refractive nature of water interacting with light. Our percept of it is, therefore, an accurate awareness of that fact of reality. This is a fact which must be taken into account when spearing fish in a stream, for example, and therefore has direct survival value, as well as philosophical and scientific importance.₁₂

Objective conceptual knowledge, however, is certain and infallible and used in combination with percepts, is a powerful survival tool. Such knowledge is objective because it corresponds to reality and is therefore useful to select survival actions from alternatives.

Human beings can even reanimate information from memory to a degree by imagining their abstract ideas in action. That is how new hunting techniques, agricultural techniques, animal breeding, and new machines were probably invented by early cultures for example. (Don Norman, Things That Make Us Smart).

The process of the invention described in this book carries the chain of logic of animating information from memory in the imagination one step farther: It converts information about life and conscious processes from static, abstract description (in a natural language or a mental simulation in a human imagination) into a form that can be executed on a computer system, making the information dynamic again, but in a different form that is more independent of a human mind. In addition, the invention includes the idea of adding teleological causation to the simulation system so its simulated life-forms can act for their own goals, instead of acting only for human goals.

In the current state of the art, computer simulations of new jet airplane designs, for example, automate descriptions of the planes that have been created by the human imagination, putting the designs into virtual operation to test not only the interaction of parts such as the landing gear, but the entire airplane in simulated flight, and even whether a person's hand will fit into a small space to replace a part. This invention does the same for life-forms: It does so by simulating in specially designed simulation software some of the attributes of biological life-forms such as goal-directed behavior (teleology) and the processes of consciousness.

Experiments that simulate life-forms (known collectively as the art of Artificial Life (AL)) have shown that it is possible for computer systems to simulate various aspects of life processes, though all the work I have been able to find in the current state of the art uses non-teleological (mechanistic) causality to do so, and is therefore quite limited. And of course, the field of Artificial Intelligence (AI) has shown it is possible to automate some complex behaviors of humans and other life-forms, though also in very limited ways in the work done to date, and always to satisfy some human goal.

Note - These experiments are well publicized and common knowledge in the AL and AI communities. For example, there is the Big Blue chess program and the Blue Eyes project done by the IBM Corporation, the "animats" of Pattie Maes of MIT, and the robots of Rodney Brooks, also of MIT. There are many other examples.

The work in AI especially, is not so much the simulation of consciousness as it is the automation of certain complex human behaviors, especially the attempted automation of reasoning. Even in the AI field itself, this approach has produced mixed results that have not been very life-like, though successful in very narrow domains._b

The IBM Corporation's Big Blue chess program, for example, does not simulate a life-form, but mathematically calculates the consequences of various moves in the game of chess. In this way, it is not much different from any other large computer application program such as those that model complex financial systems or attempt to predict weather cycles. Big Blue does not simulate a thinking life-form or the process of choice performed by a human chess player, it is merely a very powerful forecasting program with the ability to automatically move chess pieces in reaction to moves by human players or other chess programs using a series of pre-defined actions activated by mechanistic causality.

Other examples of non-teleological AL and AI work in the current state of the art are the Blue Eyes project at the IBM Almaden Research Center in San Jose, CA, and a robot called "Kismet" that has been developed by Cynthia Breazeal in Rodney Brook's lab at MIT._b The Kismet robot is especially interesting because it has a "face" that resembles a human face and is capable of simulating human facial expressions such as "calm," "disgust," anger, "happiness," and so on. Kismet makes these "expressions" in response to what it "sees" or "hears" with its camera "eyes" and microphone "ears." While this is useful work in the field of robotics, and could in fact be used by the simulated consciousness of my invention to show simulated feelings in the real world, it is important to note that Kismet is neither conscious nor goal-directed. Kismet is a collection of hardware and software that automatically and mechanistically mimics certain human facial expressions in response to various scenes and sounds captured by its sensors using pre-defined actions. It does not simulate a life-form (except in appearance like an automatic puppet), it is not goal directed because it has no values of its own, nor the means to evaluate. Kismet is not conscious; it is mechanically sensing its environment and comparing those sensations to a database, much like a robotic welding machine or a waldo, but one that uses light instead of pressure for activation of its actions.

Rodney Brooks own robot, COG, is a similar example of the same mechanistic processes, and was the inspiration for Kismet._c COG is a mechanistic replica of some of the functions performed by the autonomic nervous system in a human. It is impressive, useful work, but it is not teleological. COG is performing its actions to achieve the values of Rodney Brooks, not its own.

This is not an artificial distinction: Rodney Brooks is a human being and as such must act to gain and keep values to maintain his own life, his own survival. One of his values is making COG act like a robot; this is not a "made up" value, but a real one possessed by Rodney Brooks: He gets paid for doing the work; with the money he can buy the food and other things he needs to survive; if he does not do so (or find some other means), he will soon be dead and cease to exist. COG, however, has no values because COG is a machine not a life-form, so it can take no action to achieve values it does not have. COG is an example of mechanistic, "billiard ball" causality operating in a very sophisticated machine using pre-defined actions activated strictly by mechanistic causality that has been pre-defined by a human being for human purposes.

What has not been shown in the work in AI or AL is the simulation of a complete life-form based on the theory of the teleology demonstrated by real life-forms, life-forms with consciousness as an attribute. That kind of simulation must be built around the goal-directed behavior of the simulated life-form, not that of its human programmers. This means the simulation must be internally powered and motivated by the value significance of its actions to its own survival. That type of design is not found in the current state of the art.

The description in the following sections will show someone skilled in the art of computer programming how to build a computer system that simulates the complex causal aspects of biological life-forms from the teleological point of view; it will also show how to simulate perceptual consciousness, conceptual consciousness, rational self-consciousness, volitional behavior, and natural language understanding at the level of simple sentences; these are all attributes that are possessed by some biological life-forms, not by machines.

The most important word in the previous paragraph is the word "simulate." Life processes in general and consciousness in particular are attributes of living entities, not attributes of machines. Strictly speaking, a machine such as a computer system can never be conscious in the exact same manner as biological life-forms because machines are not alive.₁₃

The description in this chapter will show that life and conscious processes operate with a different form of causality than do machines, and that all we can hope to accomplish is to simulate life and conscious processes, to imitate them using computer technology to store teleological process relationships to reality that are similar to those that life-forms possess, and to then animate those process relationships to some degree of accuracy.

The other important issue to hold firmly in mind, as was mentioned at the beginning of this introduction, is that consciousness is a limited, quantify-able, causal process with a specific identity, and that identity determines what consciousness is capable of doing in reality as an attribute of a life-form. Consciousness is part of everyday reality and neither supernatural nor a transparent, empty non-entity. As with the objects that are its data, consciousness itself is identity; it is a collection of properties and values in the form of an active, teleological process with specific content that causes specific effects in reality.

The design and functionality of the consciousness simulator described on these pages depends on this fact as its modus operandi. But before that description can begin, I must better differentiate the work described herein from the current state of the art, and it must be determined where the description of my invention should start, what processes should be described, and in what order.

5.1.2 Differences from the Current AI/AL State of the Art

It should be apparent to the reader by now that the invention described herein is designed in a way that is very different from the current state of the art AI and AL systems.

Most of the differences stem from differences in the invention's theoretical basis. In fact, the invention is so radically different theoretically that I will deal with the major differences mostly in this section, and only mention them in a few other places through the chapter where they are particularly important to the reader's understanding and to differentiate the details of the description. It should also be noted that to see the differences clearly, a significant investment must be made in reading and thinking about the passages cited in the primary references. In fact, it is strongly recommended the references be read in detail because the ideas they contain are not widely known.

State of the Art Concepts vs. Objective Concepts

I will describe the nature of concepts and how they are calculated in this invention in detail in the latter part of this chapter to make clear the specific kind of concept required for the invention to work; however, it is crucial to the understanding of even the earlier parts of the description for the reader to differentiate the term concept as I use it in the following description, and the more common usage of the term in the field of AI because the two uses are incompatible.

In the context of this invention, I mean a very specific type of concept and only that type: I mean concepts as formed by the method proposed by Ayn Rand in her book Introduction to Objectivist Epistemology. Concepts formed by the Objectivist method are the only kind of concepts that this invention will simulate because they are the only kind that will enable the invention to work. In addition, Rand's method provides for a specific and objective way of abstracting concepts into a hierarchy, and then validating that hierarchy; it is a methodology that is found in no other epistemology and is essential to the simulation of thought.

It is Rand's method that makes concept formation and validation a calculate-able process that is based on and connected to reality, as opposed to being arbitrary constructs as are found in the current state of the art.

In the current state of the art in AI and other fields there are many ideas that have been advanced as to the nature of concepts and various "cognitively plausible processes that lead to new concepts," such as induction, deduction, abduction, analytical reasoning, and so on. None of these ideas apply to this invention as they are described and in those contexts.

Whatever those processes lead to, they do not lead to concepts as they are defined by Objectivism and simulated in this invention. That is not to say that DLFs will never perform any of the processes listed in the previous paragraph, just that they will not use them as a method to form concepts.

If one studies the history of epistemology, one will find that no matter what their identity, all concepts ever formed fall into one of three categories as a result of how they are formed:

Of the three ways of forming concepts, the objective concept formation method is used by this invention because it is the only one of the three types that is a method. To simulate concept formation on a computer system, neither mysticism nor generating arbitrary concepts at random will work very well.

Only an objective method that enables the calculation of concepts based on the similarities and differences in real objects measured in simulated sense perceptions can connect a simulation system to the reality outside of it in order to simulate an awareness of that reality, do so objectively, and give meaning to the system's natural language words. That is why I chose the Ayn Rand's method to use for this invention.

However, that choice also makes the ideas described herein theoretically incompatible with most of the rest of AI and AL, due to various contradictions between my theoretical premises vs. theirs.

Theoretical Differences

In the current state of the art, AI and AL systems are designed around the rule bases of expert systems, genetic algorithms, neural networks, contextual ontology spaces, and other similar computing techniques commonly used in the fields. For example:

Patrick Henry Winston is the director of the Artificial Intelligence Laboratory at MIT. His book Artificial Intelligence (Third Edition) is described in a Web page on the Internet as follows:

"Part I is about representing knowledge and about reasoning methods that make use of knowledge. The material covered includes the semantic-net family of representations, describe and match, generate and test, means-ends analysis, problem reduction, basic search, optimal search, adversarial search, rule chaining, the rete algorithm, frame inheritance, topological sorting, constraint propagation, logic, truth maintenance, planning, and cognitive modeling."_e

All of these computing techniques (and more that are listed for the other parts of the book) were designed to solve various computer programming problems. None of them, however, has anything to do with life processes, survival values, or the alternative between life and death for biological life-forms, or objective concepts which are central to the invention I have been describing.

Doug Lenat's Cyc system is another good example of state of the art AI technology as it developed through the 1980s and early 1990s:

"... During the 1984-1989 time period, as the CycÆ common sense knowledge base [Lenat&Guha 90] grew ever larger, it became increasingly difficult to shoehorn every fact and rule into the same flat "world." Finally, in 1989, as Cyc exceeded 100,000 "rules" in size, we found it necessary to introduce an explicit context mechanism. That is, we divided the KB (Knowledge Base) up into a lattice of hundreds of contexts, placing each Cyc assertion in whichever context(s) it belonged."_c

This system, as with the previous example, is based on an enormous set of rules devised by human beings and their interpretation of how various aspects of human intelligent behavior works. However, it is not based on human consciousness as an attribute of a life-form, nor does it take into account the biological basis of consciousness, or the formation of concepts based on sense perception of reality.

The Dynamic Memory of Roger Schank

Another well known researcher in the field of AI is Dr. Roger Schank, and one of his books that is representative of his views on AI is entitled Dynamic Memory.

The views on AI that Dr. Schank expresses in his book on human memory as he thinks it relates to AI research are consistent with those of other researchers in the state of the art. For example:

"We hypothesize that the basic entity of human understanding is what we have termed the personal script. Personal scripts are our private expectations about how things proceed in our own lives on a day to day or minute to minute basis."_d

Dr. Schank explains how he arrived at his hypothesis of personal scripts and other memory structures by studying various references in human psychology and why, in his opinion, these entities are a useful model for writing AI computer programs.

He goes on to explain how scripts and various other memory structure entities he and his research team have invented, such as memory organization packets (MOPs) and thematic organization points (TOPs) can be useful in both storing and processing information at various levels of generalization in both humans and computers. He also explains examples of various program structures he and his team have created to demonstrate his hypothesis._e

As with the other AI researchers who's work I have cited, Dr. Schank is a Materialist. It is clear from his book that he believes consciousness is a form of information processing that can be replaced by a computer program. It is also clear that the design of such a program can simply be invented by programmers using their interpretation of human language, memory, and thought processes, and that this interpretation does not have to be connected to the survival of a life-form.

The "Animats" of Patti Maes

Of all the people working in the fields of AI and AL, Patti Maes is a good representative of those who have been closer to using biological processes as a basis for developing computer technology that simulates life-forms she called "animats" in some of her papers on the subject and "autonomous agents" in others. The following is quoted from one paper she wrote that surveys some good examples from the AL field:

"The relatively new field of Artificial Life attempts to study and understand biological life by synthesizing artificial life forms. To paraphrase Chris Langton, the founder of the field, the goal of Artificial Life is to `model life as it could be so as to understand life as we know it.' Artificial Life is a very broad discipline which spans such diverse topics as artificial evolution, artificial ecosystems, artificial morphogenesis, molecular evolution and many more. ... The goal of building an autonomous agent is as old as the field of Artificial Intelligence itself. The Artificial Life community has initiated a radically different approach towards this goal which focuses on fast, reactive behavior, rather than knowledge and reasoning, as well as adaptation and learning. Its approach is largely inspired by Biology, and more specifically the field of Ethology, which attempts to understand the mechanisms which animals use to demonstrate adaptive and successful behavior."_a

I have added italic emphasis to this quote in four places to indicate some key phrases:

• To focus on "fast, reactive behavior" and "mechanisms which animals use" of an "autonomous agent" imply that mechanistic causality underlies the actions of animals, not the more complex teleological form of causality of biological life-forms. The term "autonomous" as used here, means "automaton" in the mechanistic sense.

Neither in the quote nor in the balance of the paper is there any consideration of consciousness as an attribute of these artificial life-forms, no description of values possessed by them, no explanation of goal-directed behavior based on the alternative of life and death from the perspective of the life-form, no description of objective concept formation, and no explanation of consciousness as a means of identifying reality for survival. There is only description of mechanistic causality operating as "fast, reactive behavior" in the form of computer software written to satisfy human desires.

The Unintelligent Robots of Mark Tilden

Another, somewhat different approach AL has been taken by Mark Tilden of Los Alamos Laboratory in New Mexico. Mark has designed a number of robots that behave remarkably like insects and other small life-forms.

The unique thing about Mark's work, is that no computer software is involved, merely some simple, but interesting electrical circuits he designed. One in particular works as follows: "Within the core of the circuit, Tilden explains, transistors oscillate, thereby producing rhythmic pulses of electricity. A robot's different walking behaviors thereby result from different rhythms as oscillators fall into sync."_f

Tilden's robots are strictly simple electro-mechanical devices, and he explains their life-like behaviors as resulting from the mathematically chaotic behavior of their oscillating circuits. However, while these may be useful devices, that simulate how some life-forms walk, they are mechanistic, not teleological, and serve only to satisfy human goals.

Conclusions About the Current State of the Art

The approaches summarized above are representative of, and leading ideas in the fields of AI and AL in the current state of the art. Yet they are all designed around various computer system constructs or other mechanistic systems that have very little to do with the teleological processes that are the basis of biological life-forms or conceptual chains calculated from the similarities and differences of real objects based on the kind of sense perception that biological life-forms use; these are not examples of teleology and concept formation as a form of conscious identification of reality for the purpose of survival.

The reason for this is quite simply that the theories on which the current AI and AL state of the art systems are based, have very different roots in the genealogy of ideas from the ideas which form the theoretical basis of the invention described in this book. The fact is that most scientists are unfamiliar with the Objectivist ideas and their consequences in the fields of AI and AL, so they base their ideas on other theories.

The net effect, however, is that there is very little basis for comparison of the ideas in this invention and the current state of the art other than to say that they are very different and incompatible.

Design and Operational Differences in the Invention

The differences I have been describing are not only theoretical, but carry over into the actual operation of AI and AL systems.

Most state of the art AI systems have been designed by modifying techniques originally created and used for other computing purposes, such as databases, expert systems, or control systems. State of the art AL systems that have been designed to more closely imitate life-forms have, however, also avoided the simulation of teleology as it operates in biological life-forms.

For the design of the simulation system embodied in this invention, I have focused on theoretical consistency with ideas of the metaphysics, teleology, and epistemology of Objectivism, and that fact has had the following practical consequences on my simulation system design:

• The specification of teleology as the essential differentiating factor of life processes from the simpler mechanistic processes that make them possible by serving as an animation platform, that life-forms are conditional and therefore have values (the ice cube vs. the earthworm), that they therefore must survive by gaining their values in order to act in the future, and the use of Dr. Binswanger's three test criteria for goal-directed action:

• Automatic action for this invention means the simulation of goal-directed behaviors and instinctual behaviors exhibited by biological life-forms such as nest building or web weaving by spiders, or the automatic processes of sense perception in higher animals, and that this is not the same as computer automation of actions.

• Pre-definition of action for this invention means the programming of the means to effect actions such as finding and moving objects, finding and eating food, using sensors, and so on, but not the order or timing as to when or if these actions will be selected and activated by a simulated life-form; action selection and activation is teleological and based on value-significance to the life-form and its interaction with reality. This is different from state of the art AI and AL systems where every aspect of actions must be pre-defined using mechanistic causality, including their order and timing, and where the acting agents are mechanistic automatons.

• The specification of the simulation of consciousness as a reality based, teleological process that is a specific, finite, natural, identity as an attribute of a simulated life-form, an active process that uses the measurements of sensors as its only data and the identity of the sensed objects as its only form of content.

• The specification that it is the nature of concepts as produced by an objective method that uses optional, goal-directed actions that leads to the simulation of rational consciousness; and further, it is precisely because such actions are optional to the simulated life-form that leads to the emergence of the attributes of fully developed simulation of volition, self-awareness, natural language understanding and reasoning ability, and that these attributes derive from the nature of reality-based concepts, especially their hierarchy and context.

• The specification that natural language understanding in particular is the process of thinking, and as such, is primarily a survival strategy and means of more efficiently processing and storing information about reality; that natural language is only secondarily a communication system. And further, that the meaning of sentences is the connections embodied in the calculation chains of concepts formed by the Objectivist method. Moreover, the meaning of sentences is the entire chain of calculated concepts traced all the way back to percepts of the objects used to form the concepts in the first place, not just the definitions of words stored in a dictionary.

The reader will find it helpful to keep these differences clearly in mind as an aid to their understanding as they read the remainder of this chapter.

5.2 Biological vs. Digital Life-Forms

In order to make the distinction between state of the art mechanistic automatons and the simulation of the teleological attributes of life-forms clearer, I have created the following explanation.

The processes and behaviors of biological (real) life-forms and the Digital Life-Forms (DLFs) that are simulated with a computer system designed to the specifications of this invention can be divided into layers according to their function in the overall simulation system. In other words, they can be broken into layers of subsystems, so they are easier to understand, as long as the layers reflect the natural dividing points in the information (somewhat like carving a turkey at its joints rather than cutting it randomly).

The higher subsystem layers are causally dependent on the lower ones. The practical effect of this is that the upper layers cannot function without the lower layers. In the world of life, a cell cannot function without DNA, energy, and chemistry, and an animal cannot live without cells; in the world of machines an operating system cannot function without computer hardware, and an application program cannot be run directly on the hardware without the operating system layer (assuming a modern computer). In the world of teleology, life process functions such as consciousness cannot function without cellular processes and food digestion, which in turn depend on the mechanisms of physics and chemistry.

However, within subsystem layers, each layer has its own uniquely independent types of causes and effects; it is a system unto itself and can only interact with the other subsystems indirectly across some kind of interface or intermediary.

For example: In a computer system, one cannot mix application program functions (which are logical and digital) with electrical resistances and capacitances of circuit oscillations (which are physical and analog), without a processor and operating system to interface the interaction between these two different types of subsystems. Similarly, in biology, one cannot mix the functions of processes like blood circulation (which is mechanical) with cell reproduction (which is molecular and chemical).

In each of these cases, objects with certain specific identities at a certain scale interact with certain causes and effects, and those causes and effects, taken as a subsystem, collectively cause effects in other layers, also taken as a subsystem, though they cannot interact directly with each other; instead they act through one or more intermediaries or interfaces. This is true of both mechanistic systems such as computers and the entire life process. The processes of life, though based on mechanistic causality, are teleological processes that interact with the mechanistic world through specialized chemical reactions at the molecular level and sensors and muscles at the human scale.

Note - What this idea means in effect is that while some systems of different forms can interact, they may not do so directly. For example, the systems may contain dis-similar objects, the attributes of which are incommensurable, and therefore they cannot interact directly. The reason is that causation is enabled by the identities of the objects that interact, and incommensurable identities cannot interact with each other; a capacitance, for example, which is electrical and analog cannot directly call a digital program method, though that capacitance may maintain some value in memory for the digital program. A computer processor must serve as intermediary between the two incommensurable contexts of analog and digital to enable their interaction.

I introduced this idea in Chapters 3 and 4. Below, Table 5-1 reviews the layers of subsystems into which life-forms can be divided, as well as some essential functional similarities and differences between real and simulated life-forms.

The fact that various processes are on the same layer in the table does not mean they are exactly equivalent, but rather, for the purpose and context of this invention, they are functionally equivalent. They are similar enough in function so that they can be causally substituted for one another.

For example, in the lower three layers, while the computer system attributes shown are not the same as their biological counter-parts, in the context of animating a digital life-form, these subsystems perform similar enough functions, such as providing power and control, that within the context of this invention they can be considered functionally equivalent.

Likewise with the upper layers; simulated perception, concept formation, and reason are not the same as in humans, just special, calculated imitations of these functions, but imitations that are similar enough to be useful to the digital life-form to aid in its simulated survival just as the real forms of these functions are useful survival tools for human beings.


	Biological life-forms	Simulated, Digital life-forms
Layer 7	Conceptual Consciousness (Reason)	Simulated Conceptual Consciousness
Layer 6	Perceptual Consciousness	Simulated Perceptual Consciousness
Layer 5	Goal-directed Cellular Processes	Simulated Goal-directed Behavior
Layer 4	Mechanistic Cellular Processes	Digital life-form Simulation Program
Layer 3	RNA, Protein, ATP Synthesis	Object-Oriented Prog. Environment
Layer 2	DNA Processes	Computer Operating System
Layer 1	Electro-chemical, Physical Processes	Computer Hardware

To carry the analogy a step farther, just as a computer application program cannot function without the operating system and hardware in the layers below it that cause its operation, so the attribute of consciousness cannot function without the lower teleological layer 5 on which it depends causally either, or without the mechanistic layers of 1-4 that support layer 5.

Consciousness is an attribute of certain types of life-forms and is caused by their teleological processes; it is not a "stand-alone" process. Likewise, natural language and reason, which are the most complex behaviors in humans, are not stand-alone processes either, but are caused by and depend on conceptual consciousness, perceptual consciousness, and ultimately life processes operating in cells, to function.

To simulate complex human behaviors such as natural language and reason, all of the subsystems on which they depend must therefore be simulated as well.

5.2.1 Computer Systems vs. Teleological Systems

Computer systems and life-forms are two distinct categories of systems. The essential difference between them is that one is alive and the other is not; the form of causality that operates in state of the art computer systems is simpler than the form that operates in life-forms. Computer systems, as currently designed, are merely machines that automatically execute pre-defined actions worked out by human consciousness for human goals. Life-forms, on the other hand, are teleological; they have their own goals; they also initiate and sustain their own actions to maintain their very existence.

Note - The concept "causality" as used here means that the identity of an object determines its action capacity in relation to other objects, such as with the case of the difference between dropping a bowling ball and an air-filled balloon from a tall building: The two very different identities of these objects cause two very different effects. This view enables more detailed interactions to be described, as opposed to the over simplified action-reaction view of causality.₁₅

The form of causality that operates in life-forms is more complex than simple mechanistic causality because a life-form's existence at any given time is dependent on all previous instances of its existence and the success of the survival actions it took in those instances, whereas the existence of a state of the art computer system or any other non-living object is not.

In other words, the existence of a computer system is unconditional; it may stop working, but it remains part of reality and can be restarted. Whereas, the existence of a life-form is conditional; furthermore, the conditionality is part of the life-form's identity as a life-form. The condition is that the life-form must cause its own future existence (survival) precisely because it is a goal-directed or internally driven teleological entity, as opposed to a rock, which is not goal-directed; any actions of a rock are simply the result of outside forces. This is a metaphysical difference. Failure to attain its goals causes the life-form to cease to exist, a condition in which it is no longer part of reality and one that is irreversible.₁₆

For example, since state of the art computer systems use simple, mechanistic causality; if there is a causal failure in a program that makes it stop functioning, nothing happens to the program (in most cases) or the computer hardware, other than part of it stops working; it can be restarted later by human intervention. But if there is a causal failure in a life-form (sickness) and the immune system is unable to cope with it, the life-form not only stops working, it also ceases to exist. It dies; its physical form putrefies and disintegrates because life-forms are driven and supported from the inside, not the outside like all other kinds of objects. This is a profound and fundamental difference between state of the art computer systems and biological life-forms, between mechanical/logical systems and teleological systems.

Note - Computer systems are conditional in the sense that they are man-made, and man having volition could have chosen not to make them at all or to design them differently or choose not to repair them; but that is a different issue and not relevant to the point being made here. The same is true for computer code that ceases to exist if it is only stored in Random Access Memory and there is a power failure; the code is not alive and does not have the capacity to save itself (being controlled from the outside by humans), whereas life-forms do have the capacity to save themselves (within their functional and internal energy supply limits).

Any number of alleged counter examples can be created that supposedly disprove the conditional nature of life-forms is fundamental; the reason usually given is because other objects can be said to be conditional too, such as clocks with bombs as part of their structure or submarines designed to operate in acid and disintegrate if they stop maintaining their hulls, and so on. However, these mechanical examples miss the essential point about the conditionality of life-forms.

First, the conditional nature of life-forms is not a "design standard," it is a metaphysical fact that anyone can observe for themselves. The design and function of life-forms is not determined by some arbitrary, revocable human choice, but rather by the basic nature of the life itself, and the requirements of the reality in which life exists and must ultimately survive. Life is an objective fact of reality.

Take the robot submarine example: The main problem with this example is that the submarine is a man-made machine, a human-goal-directed object, not a self-goal-directed object.

The reason the submarine is not a goal-directed object, is that it operates conditionally only because human beings made it that way, and they only made it that way so it could act to further human values; those values are external and not the submarine's own values; it has no values of its own. The conditionality is contrived, not natural.

The reason life-forms are conditional is because they evolved that way, naturally. The reason life-forms are goal-directed is because they are conditional: They are naturally and intrinsically unstable and will cease to exist without continuous, internally caused action. Biological life-forms' internal energy source, means of action control, and values came into natural existence before the life-forms did as extensions of mechanistic causality; these complex causal processes are what make life-forms possible in the first place. They are not the same as conditional processes invented by human beings for the sake of argument.

The second way machine counter-examples miss the point is that the biological function of the survival vs. death alternative for life-forms is not just some non-essential attribute they happen to possess, but one that goes to the very core of their existence: Conditionality is a means of natural behavior selection that works by wiping out the life-forms with any behavior that is not survival efficient. Life actively uses conditionality to naturally alter identity (remember, identity determines action capacity); the result is only life-forms that encode efficient survival behaviors exist over the long-term. Conditionality causes evolution.

To illustrate the nature of conditionality, Dr. Binswanger quotes biologist Albert Lehninger as follows: "A living cell is inherently an unstable and improbable organization; it maintains the beautifully complex and specific orderliness of its fragile structure only by the constant use of energy. When the supply of energy is cut off, the complex structure of the cell tends to degrade to a random and disorganized state."₁₇

The nature of life-forms is that they are conditional objects, and their continued existence depends on the condition that they themselves continue to act to attain the goal of survival, to cause their own future survival, by internally powered and controlled causation. Hence Ayn Rand's definition of life as "self-initiated, self-sustaining action" as explained by Dr. Binswanger.₁₈

"Life and goal-directedness are intimately related. For consider the three requirements of goal-directedness: self-generation, value-significance, and goal-causation. Each implies the others, and all are a consequence of the essential nature of life: conditionality. Self-generation means there is an internal store of energy. This store must be replenished-hence the need to obtain energy-hence the phenomenon of value-significance. What underlies goal-causation? The fact that only valuable actions get repeated. Why do only valuable actions get repeated? Because the value here is survival value, and to repeat the action, the agent must survive."₁₉

Robotic vs. Goal-Directed Causality

The previous section notwithstanding, what if people design an artificial life-form using biological components instead of a computer simulation? Would such an entity be teleological even though it has been designed by human beings?

The answer is "yes." It is true that being designed by humans would mean that to some degree, the artificial life-form would share human values because the humans who designed it would want it to do certain things or they would not have built it in the first place. But if it is a real life-form, that is, if it is alive, it must be conditional because that is the essential attribute of all life-forms. The artificial life-form would therefore be goal-directed, which means be internally driven by its own values, energy source, internal locus of control, and the value-significance to itself of its own values. The fact that it also shared some values with the humans that built it is irrelevant:

The essential facts are that it has values, it acts to gain and keep them on its own power, and its survival depends on it doing so.

Note - Nanotechnology may someday make this possible, as pointed out in the previous chapter.

To build a simulation of mammalian life and the most complex attribute of biological life-forms, conceptual, rational self-consciousness, the causal context of the rest of the life-form cannot be dropped, especially its conditionality and goal-directedness. As an attribute of a life-form, consciousness cannot exist without the complex, teleological causes that make it possible, any more than it can exist without the mechanistic ones that underlie those; the causes on which conceptual self-consciousness depends are all the layers of causes, all of the subsystems below the top layer as shown in table 5-1.

Based on the foregoing explanation, it is now possible to see where the description of a consciousness simulator must start. It is the nature of teleological causality that determines the starting point, because this is what serves as the interface between the mechanistic causality that rules in the subsystem layers below it, and the conscious behaviors that operate in the layers above it.

Just as various mechanistic cellular processes such as RNA, protein, ATP synthesis, DNA processes, electro-chemical processes, physical molecular processes are the mechanistic basis of the goal-directedness of biological life, there must be some mechanistic processes to be the basis of the goal-directed processes of simulated life.

The point of this discussion is not to show that machines can exist conditionally (which they cannot), but to show that in order for life processes, consciousness, and volition to be simulated, the conditionality that occurs naturally and controls behavior in biological life-forms must be emulated by a simulation system if it is ever to simulate intelligence. The design strategies and logical processes that work for mechanical automatons will not work for simulating life-forms, only teleology will.

This invention is not a "conditional machine." It is a new kind of virtual entity that emulates the conditionality attribute found naturally in the identity of all life-forms, and it does so by simulating the complex teleological causality of life.

Computer vs. Teleological Action Definition

Causes are not causes without effects, and effects in the context of life are the actions of some teleological agent acting on some object(s) in reality in order to survive.

There is a huge difference in the way state of the art computer agents (automatons) and goal-directed agents as defined in this invention operate:

• Computer agents and other state of the art automatons are all externally driven. They do not have values, they do not have their own energy source or internal control, they do not have to survive by their own actions; state of the art computer agents are completely defined and controlled by the parameters of their programs to implement human values. Any situation that puts them outside their pre-defined parameters will necessarily cause them to fail to operate.

• Life-forms and simulations of them as used in this invention are internally driven. Teleological agents are driven to act by their own internally held values, with their own internal energy supply and means of controlling it. Moreover, the actions they take are primarily and necessarily on their own behalf (though they may also coincide with human values); the purpose of those actions is to maintain their own existence, to survive. Only after survival is assured can teleological agents engage in optional actions: This is because without survival, the agents do not exist, and non-existent agents cannot act. Moreover, it is why only survival actions are necessitated; survival actions are necessitated precisely because they are required for survival (and hence cause all future actions of any kind); whereas optional actions are not necessarily part of the causal chain required for survival, hence the term optional.

These primary and fundamental differences in the nature of acting agents lead to some additional differences in the way actions occur in computer agents vs. teleological agents in:

• The nature of automatic actions: Automatic action selection is by means of a goal-directed, simulated pleasure-pain system that simulates that biological subsystem which evolved to safe-guard the lives of biological life-forms by providing rapid calculations of the relative value of objects perceived by a life-form and communicating the results to the consciousness of the life-form with pleasure and pain feelings.

• The existence of "optional" actions: When survival needs are met, and only then, does the survival value of actions calculate much closer to being equal because the life-form has built up some energy reserves and there are no immediate threats to its life. Optional actions that are not part of the causal chain required for a life-form's survival, and hence it matters little, if at all, which one a life-form selects and performs, though these actions can become crucial to survival in the future of a life-form.

State of the art AI and AL computer programs are not designed in the manner just described.

5.2.2 The Starting Point for Describing the Invention

Just as it is necessary to start with the hardware, then move to the operating system, application programs, and so on to build a state of the art computer system, it is necessary to start by simulating the electro-chemical and physical process of life-forms, followed by all the layers of subsystems to build a simulation of consciousness. The layers of subsystems not only contain implementations of all the necessary and sufficient concepts that must be in the system in order for it to work, but the layers must be causally linked as well.

This can be accomplished by causally substituting and linking a computer simulation system shown in layers 1-4 of Table 5-1 to animate the objects and relationships being simulated, namely the digital life-forms in layers 5-7. The animation is of a different form than its biological counter-parts, but it is functionally similar enough to serve the purpose of this invention.

One cannot run a modern application program directly on computer hardware, leaving out the operating system layer; it is necessary to proceed through all the subsystem layers in between without leaving any out to have an unbroken causal chain that connects simulated consciousness to reality.


	Biological life-forms	Digital life-forms
Layer 7	Conceptual Consciousness (Reason)	Simulated Conceptual Consciousness: open-ended categories formed by calculating differences and similarities using perceptual measurement ranges, symbolized by natural language words, connected to percepts through calculation chains of other less abstract concepts, volitional actions, virtual self for self-consciousness
Layer 6	Perceptual Consciousness	Simulated Perceptual Consciousness: automatic feature identification, unit economy, reality interface, physical actions to cause changes in reality
Layer 5	Goal-directed Cellular Processes	Simulated goal-directed behavior: automatic self-regulation (in the biological sense), self-generated energy, values and value significance, survival, optional behavior
Layer 4	Mechanistic Cellular Processes	Digital life-form Simulation Program
Layer 3	RNA, Protein, ATP Synthesis	Object-Oriented Prog. Environment
Layer 2	DNA Processes	Computer Operating System
Layer 1	Electro-chemical, Physical Processes	Computer Hardware

Let's look at Table 5-1 again, this time with the necessary and sufficient concepts added.

In the case of digital life-forms, this means simulating the more complex form of causality on which they depend for their existence, goal-directed behavior (layer 5), followed by perceptual consciousness (layer 6) and conceptual consciousness (layer 7), to simulate rational self-consciousness, a capacity that is an attribute of human beings. It is only in this way that a consciousness simulator can be built that will come as close as possible to behaving like a biological life-form that mimics some human-like traits.

Moreover, to simulate consciousness in a manner that resembles what is observed in higher animals and humans (a goal directed process), a teleological program (layer 4) must overlay the mechanistic state of the art computer system (layers 1-3) to serve as an interface between the two forms of processing because they depend on different logic: the mechanistic, logical computer processing (layers 1-3) and the inherently conditional, teleological form of processing (layers 5-7) of life and consciousness.

Note - Layer 5 also makes optional behavior possible, and it is optional behavior that both makes concept formation possible and forms the foundation for volitional behavior, or free will, in digital life-forms. The "how" of this will be explained in detail later in this chapter.

Just as the analog world of the electronic components of a computer cannot interact directly with the digital world of its programs without the processor as the intermediary, so the mechanistic world of the digital program cannot interact directly with the teleological world of a simulated life-form without the teleological program as an intermediary or interface between mechanistic and teleological process logic.

Now that it is clear where the description of the invention must begin, I can continue and describe how to build a computer system to simulate consciousness as an attribute of a simulated life-form.

The content of the description depends on the axiom that consciousness is a metaphysical primary, that it is a relationship between a certain type of life-form and reality that cannot be broken down into simpler components in order to reduce it to mechanistic causality, and that this is true even though consciousness is ultimately based on mechanistic causality. Consciousness is a relationship that either exists or not, period.

Consciousness as a state of awareness just is, is what it is (A is A), and is conscious. It is an attribute of certain life-forms; it is the relationship of awareness of reality as a collection of objects as opposed to awareness as discrete sensations or to non-awareness (meaning "nothing" or "void," not a contrary kind of relationship). Consciousness is one of a number of relationships that the life-forms that possess it have with the world.₂₀

The content of the description is also based on the corollary axiom that consciousness is causal, that it is a process with a specific identity that interacts with the identities of other objects in specific ways; in other words, consciousness is an active, limited, teleological process like life itself.₂₁

Consciousness is caused by one set of teleological subprocesses, and life is caused by a different set, but both phenomena are caused processes. When some of the subprocesses of awareness are active, consciousness exists; when they are not, a life-form which has the attribute of consciousness is unconscious. This is analogous to the way a life-form is alive when its teleological life subprocesses are operating, and dead when they no longer function.

The description of my invention will show that to simulate automatic consciousness amounts to enacting its biological causes in the form of measurements and logic in a teleological computer simulation system that has an interface with reality that is as causally equivalent as is technically possible to that of a biological life-form. In other words, the state of simulated consciousness will be an attribute of a simulated life-form in a simulation system.

In addition, the description will show that simulated rational consciousness is a virtual subprocess of simulated consciousness that emerges from simulated automatic (perceptual) consciousness reorganizing itself with "data structures" called simulated concepts that are formed using optional mental actions. Further, the description will show how optional mental actions, the identity of concepts, and their logical relationships make simulated volition and simulated natural language possible.

Finally, the description will show how the repeated use of simple, first level simulated concepts leads to the emergence of simulated self-consciousness, how this latter leads to the simulated ability to process simple natural language sentences, which in turn enables the simulation to achieve the power of simulated volition and the capacity to initiate first causes.

In other words, the description will show how the simulation system bootstraps itself.

5.3 A System Design for Simulating Conscious Life-forms

If a life-form simulator had to be designed by simulating all the mechanistic process on which life depends, such as various chemical and molecular processes, the DNA processes, the RNA processes, protein synthesis processes, ATP cycle processes, and so on in exact detail, building such a simulator would be very, very difficult indeed. Luckily, this difficulty can be avoided by using something called causality substitution.

I explained this topic briefly in previous chapters, and will describe it in more detail in this section.

5.3.1 A Computer Network Analogy

In the field of computer networking, there is a model for network design specified by the International Standards Organization called the OSI model. The OSI model is a layered model similar to the one shown in Table 5-1 for life-forms in that the upper layers depend on the lower ones, except that the OSI model, shown in Table 5-2, describes network technology.₂₂

The OSI model is interesting in this context because it provides for something called layer substitution, a concept that is very useful for network design and which I have extended for use in the field of life-form simulation.

The bottom two layers of the OSI model are the Physical layer and the Data Link layer. These subsystems and how they are used in network design is the idea that led me to think of the idea of causality substitution. So that the reader may understand how I arrived at this idea as well, we will make a brief digression with an example of how layer substitution works in the field of network design before continuing.

The example is as follows: In networks that follow the OSI model, the Physical layer is the network cable (or other medium such as radio frequency or infra-red light), along with its interface card that physically puts the binary bits on and off the media, and the Data Link layer is the software program that sends and receives data packets from one network node to another using a protocol such as Ethernet or Token Ring or some other proprietary network protocol.

Each layer in the model causes the functions that occur in the layers above it.

Table 5-2 The OSI Reference Model
Layer	Name	Function
7	Application Layer	Program-to-program communication.
6	Presentation Layer	Performs data representation conversions.
5	Session Layer	Opens and maintains communication channels.
4	Transport Layer	Maintains end-to-end integrity of the data transmission.
3	Network Layer	Routes data from one node to another.
2	Data Link Layer	Transmits bits as packets from one network node to another.
1	Physical Layer	An interface card that puts data on & off the physical media.

There are various kinds of network cables such as twisted pair, coax, or fiber optic, and there are other media such as radio frequency or infra-red light, each requiring a different computer interface card.

Layer substitution in network technology allows for the mixing and matching of these different technologies, depending on the needs of a given network application.

For example, a company using a proprietary network protocol may have twisted pair cables; they may wish to replace these with fiber optic cables which have a much larger capacity, effectively substituting one type of Physical layer for another. To effect the change, they would replace their twisted-pair cable with fiber optic cable. They would also need to change the interface cards because fiber-optic cable uses light to carry the bits, whereas twisted pair uses electricity.

They may run their original, proprietary network protocol at this point and their network would function, but they may also want to change to the more common Ethernet protocol. In order for the network to function with this second change, the Data Link layer must be substituted in addition to the physical layer. This is necessary because different software is needed to manage sending Ethernet data packets from one network node to another vs. whatever software routines their proprietary protocol used.

Once the substitution of the Data Link layer software is made, network users would probably notice better performance as the only difference, because the upper five layers of their network as specified by the OSI model would be unchanged, and their network would continue to function as before, only faster. The substituted bottom two layers simply cause the same effects as before in the upper network layers, but in a different manner, one that uses a different kind of network packet and uses light instead of electricity to physically transfer the data bits. In other words, one set of causes is substituted for another set, which lead to the same effects in the overall system, and this is accomplished by substituting layers or subsystems.

5.3.2 Substituting Layers

In a similar manner, I have extended this idea to build and animate a life-form simulator using a computer simulation system.

A causality substitution can be made to simplify the simulation of the mechanistic processes that make biological life possible in the real world; instead of attempting to simulate chemical and other molecular processes directly, standard mechanistic computer code is substituted to run the teleological software, which in turn animates the simulated life-form.


	Biological life-forms	Digital life-forms
Layer 7	Conceptual Consciousness (Reason)	Simulated Conceptual Consciousness
Layer 6	Perceptual Consciousness	Simulated Perceptual Consciousness
Layer 5	Goal-directed Cellular Processes	Simulated Goal-directed Behavior
Layer 4	Mechanistic Cellular Processes	Digital life-form Simulation Program
Layer 3	RNA, Protein, ATP Synthesis	Object-Oriented Prog. Environment
Layer 2	DNA Processes	Computer Operating System
Layer 1	Electro-chemical, Physical Processes	Computer Hardware

As shown in Table 5-1 above, to accomplish this goal, the mechanistic computer causality of the bottom four layers of digital life-forms is substituted for the biochemical mechanistic causality of biological life-forms.

The fact that the mechanistic processes involved operate differently is irrelevant in the same way and for the same reasons as in the computer network example. It is not the differences that are important between the biological processes in say, layer 3 of ATP synthesis and an object-oriented programming environment, it is one key similarity that is important: The fact that both of these processes are mechanistic; the same is true of all the other processes in layers 1-4. The processes on both the biological and DLF sides of the table are mechanistic processes that serve to animate the life-forms. That fact makes them functionally equivalent for the purposes and in the context of this invention.

Hence, one set of causes can simply be substituted for another set, which have the similar effects at higher levels in the system, namely the effects of animating it. The substituted causes and effects function in a manner that imitates or is similar enough to the biological causes and effects in the final outcome that, for the purposes of the simulating life-forms, one can be substituted for the other. As with computer networks, it makes no difference causally whether you use fiber optic cable and Ethernet or twisted pair and Token Ring (two very different systems), the information packets still get sent transparently from computer A to computer B as far as network users are concerned.

So far as I am aware, this concept of causality substitution is a new idea to the field of AL. The point of the concept is to demonstrate a means of simulating the animating effects of a tremendously complex molecular chemistry by means of using a computer system to supply the causes.

In biological life-forms, molecular and other mechanistic cellular processes cause energy to be generated and stored. This animates the life-form on a short-term basis. In a similar manner, the electrical energy of the computer hardware, the functions of the operating system, and software application environment and its programs will animate digital life-forms on a short-term basis just as molecular processes do in biological life-forms. Causality substitution makes the animation possible using existing computer technology so the digital life-forms can be made to imitate or function as much like biological life-forms as is technically possible without using chemistry.

The more interesting processes to simulate are those in the teleological, goal-directed behavior layers that sustain energy generation and life-form animation over the long-term. These involve more complex forms of causality. They are the subject of the next section.

5.4 Setting Goals in a Computer Simulation System

Having explained the substitution of the lower layers in the model shown in Table 5-1, it is now time to describe the processes in detail that occur in the upper layers that simulate the actions of biological life-forms and that make the long-term "survival" of digital life-forms possible.

This section and the other sections that follow contain additional new ideas not in the current AI or AL state of the art, but they are ideas that a skilled programmer can implement on any reasonably powerful desktop computer, using an object-oriented programming environment.

5.4.1 What is Goal-Directed Behavior?

On the one hand, it should be noted that the goal-directed or teleological behavior of life-forms is caused behavior, not some kind of intrinsic vitality or other sort of miracle. On the other hand, goal-directed causes are not merely the simplistic, "billiard ball" kind of mechanistic causes either: The causes of goal-directed behavior are of a different form that is more complex than mechanistic causes, but still, it is important to understand that teleological causation is based on mechanistic causation and could not exist without it, just as a computer network cannot exist without its lower layers that put binary bits on and off some wire or other physical media. The essence of goal-directed behavior is that it is a form of causality that makes possible the existence and animation of conditional objects: life-forms (as opposed to the mechanistic causality by which everything else in existence operates, and which also serves as the foundation for teleological causality as shown below in Table 5-3).


	Mechanistic Causality	Teleological Causality
Objects existence type	Unconditional: objects will exist until natural erosion or other mechanical, chemical, or nuclear processes break objects down into simpler parts. They may last seconds or billions of years.	Conditional: life-forms will die and cease to function immediately if specific conditions are not maintained or at the end of a predictable life-span, then putrefy within hours of death and disintegrate in days to weeks.
Source of action	External	Internal, self-generated
Need for action	None	Must act to remain alive and in existence, self-sustaining
Source of energy	None or external (excluding nuclear reactions as in stars)	Internal, but derived from an external source by own action
Locus of control	None	Internal
Values and needs	None	Value energy and other things their conditional existence requires to sustain
Value significance of environment	None	Crucial: without gaining values such as food and avoiding disvalues such as predators, death and non-existence result very soon.
Goal	None	Survival: which is achieved by means of continuous internally generated, sustained, and controlled action to gain and keep values and avoid disvalues.

A goal-directed process is self-generated as well as self-sustaining, which means it must be internally powered and controlled. It causes its own future existence in spiral fashion: A life-form's actions in the present cause its survival in the next instant, and its survival enables it to act again in the following instant, and then the cycle continues, so each instance of action can be said to cause all future instances of action. Just one break in that chain, and the life-form will die and cease to exist.

Mechanistic causality, on the other hand, is a simpler form because the existence of the natural, non-living objects it enables to operate exist unconditionally; non-living objects just are. Causes do not have to be continually enacted to keep non-living objects in existence, to keep them part of reality. In addition, the actions of non-living objects are externally powered and controlled only by the mechanistic laws of chemistry and physics; they have no internal means to act (unless put there by humans as in the case of powered machines).

Note - Time-frames vary and are relative to the type of object as to when natural non-living objects will break down into simpler ones. Notice, however, that they have no capacity to prevent the break down, whereas life-forms do (recall the example of the ice cube and the earthworm). Some non-living objects can even "grow," such as crystals, fires, or storms, but these changes are driven from the outside, not controlled from the inside as they are with life-forms. Stars are internally powered by nuclear processes, but they are certainly not alive.

According to biologist Walter Bock, life-forms have three essential characteristics:₂₃

In other words, life-forms exist on the condition they take the actions necessary to cause their own future existence (survival), and they cause their future existence by enacting the causes listed in items 1 and 2 above. As pointed out earlier, life-forms are inherently unstable and must continuously act to maintain their existence.

Since life-forms need materials and energy to survive and only certain specific types of these items will cause survival (others cause death), some things in any environment are values life-forms must act to acquire, and other, alternative things are disvalues to be avoided. Goals, therefore, are values to be acquired (values that are not in hand); they are the standards or criteria for action selection needed for survival, to get the life causing things in an environment. These are the things that have value-significance to life-forms based on the alternative of value vs. disvalue, which translates into life vs. death.

To review the list Dr. Harry Binswanger has identified as the conditions for goal-directed action that I introduced earlier in this book:

Note - The term "agent" means teleological agent, as in life-form or some major system thereof, not mechanistic automaton as is does in the current state of the art.

It is these three conditions that distinguish the goal-directed actions of life-forms from the mechanistically caused actions of non-living objects and computer automatons. Life-forms' continued existence depends on values, the attainment and maintenance of which are caused by their own actions. The result is survival if the actions succeed, and death if they do not.₂₄

5.4.2 Interfacing Computer Systems to Value Systems

Computer systems as they are designed in the current state of the art are machines; they are not alive and do not have values. They are governed by the simpler, mechanistic form of causality.

How then, can conditional objects who's existence depends on values be simulated on a mechanistic computer system? They can be because the more complex causality of conditional objects depends ultimately on combinations of unconditional objects and mechanistic causality: non-living molecules and the laws of chemistry and physics that govern them. All teleological systems are based ultimately upon mechanistic, logical systems.

That this is so can be shown by a simple existence proof: It can be inferred from the fact that the goal-directed behavior of biological life-forms is already "implemented" on the unconditional molecules of chemistry and their mechanistic causal processes. The fact that life is based on (not reducible to) physics and chemistry, means life is based on non-conditional objects and mechanistic causes and effects at its lowest levels. Life exists, so as a corollary, an interface between goal-directed causality and mechanistic causality must also exist for biological life-forms.

Humans form concepts and use language to represent reality symbolically, then manipulate the symbols with logic to produce objective descriptions of reality in the form of language. One purpose conceptual tools are often used for is imagining or mentally simulating parts of reality. When done by humans, this is a slow, manual process consisting of many, many choices about the properties and values of the objects being simulated and their relationships. Computer simulations are the automation of that process using symbols.

Computer systems are causality simulators that can simulate any aspect of reality using man-made symbol systems that have been translated into binary arithmetic and are manipulated by electricity. Computer-based simulations can be made to operate at whatever level of detail is desired.

Moreover, there are no restrictions as to which ideas, which conceptual, symbolic representations of reality can be simulated given sufficient computer resources and time, though some can be simulated better than others. The objective description of goal-directed causality is one such man-made symbol system.

It is possible, provided teleology is used (rather than ordinary logic), to write a computer program to serve as an interface layer between the goal-directed causality of digital life-forms and the mechanistic causality of the computer systems that animate them. Such a program is the symbolic analog of the natural interface between teleological and mechanistic causality that exists in biological life-forms as part of a living cell, and it can enable a computer simulation system to imitate teleological causality.

Writing a computer program to simulate goal-directed behavior is a matter of reproducing the teleo-logic of the objective description of the complex form of causality evident in life-forms, including their values and the fact of value-significance, and building it into the computer simulation program so the system is functionally similar (or functionally equivalent, if technically possible) to the real process observed in biological life-forms. Doing so creates an interface between the two forms of causality, and when implemented on a computer simulation system, it effectively imitates life processes by substituting the mechanistic binary causes of the computer simulation program for the mechanistic molecular causes of real life-forms. (This means in effect that, in Table 5-1, layers 1-4 in the right column are substituted for layers 1-4 in the left column.)

The result is still a simulation, the manipulation of symbols in a computer system, not a life-form, but a teleological as well as a logical manipulation that imitates life processes. A digital life-form is a virtual object (an object in the form of symbols, of information and its relation to human consciousness) and not a real life-form, but its behavior can be made very similar to that of a biological life-form if the teleological causality is duplicated by the logic of a computer program, plus the digital life-form's interaction with reality. A properly designed and programmed digital life-form will satisfy Dr. Binswanger's three conditions for goal-directed behavior cited earlier, and by doing so, will be capable of causing its own future survival just like a biological life-form.

Of course, an objection could be made that this causal process is unnecessary, since the Digital Life-Form's (DLF) simulation data could be backed up, and used to simply "bring it back to life" if the DLF "died." And while it is true that this is possible, doing so would undermine the main benefit of conditionality: namely, its ability to drive independent action and to eliminate anti-life behaviors.

Simulated "death" is the primary means this invention uses to solve the problem of the apparent need to pre-define a simulated life-form's future actions. Simulated "death" solves this problem because only pro-life actions get repeated in the long-term.

5.4.3 Goal-Directed Simulation Logic: Teleologic

Conditional causes are not new to the art of computer programming. Programmers use conditional programming structures all the time. To simulate a life-form on a computer system requires only that the conditionals that are part of the computer system be used in a manner that simulates the way biological life-forms operate so that the simulated life-forms behave as close to real ones as is technically possible. Their primary purpose cannot be to achieve human goals, which is how conditional programming structures are used in all state of the art computer programs, but the goals of the DLFs themselves.

In other words, the use of conditional causes in a program to simulate a life-form must also simulate values and value-significance to digital life-forms, as well as internal energy and its controlled use for survival. It is these ideas that provide digital life-forms with the capacity to select their actions and the motivation to use that power: Upon "pain" of simulated death.

This means recreating the inherently unstable nature of life using conditional programming structures as a basis for making virtual life-forms that are themselves inherently conditional objects, and therefore able to simulate the conditional nature of real life-forms. So DLFs must be logically structured to take action to maintain their existence, and that they must be deleted if their survival actions fail.

Of course, it could be objected that unlike biological life-forms (except humans), DLFs could learn to manipulate the computer system they run on and prevent their own "death," or they could learn to "resurrect" themselves from backed up files containing their data. It is unlikely that human simulation system managers might need to take steps to prevent these kinds of actions, though it is somewhat early in the development of DLF Simulation Technology to know this for sure.

Goal-directed behavior of biological life-forms involves the repetition of certain actions such as finding and eating food (a value), avoiding predators (a disvalue), maintaining vegetative functions like digestion and blood circulation (a value), and so on. The goal is survival for the life-form (the source of value-significance), the causes are the actions taken to attain that goal.

The process and the life-form it sustains is conditional because if the actions are taken (causes are enacted), then the life-form survives to act again in the future; if such actions are not taken or they fail for some reason, then the life-form dies (the process stops breaking the causal chain) and the life-form ceases to exist in the future; it has stopped causing its own survival. The need for action is a primary attribute of living objects.

The process, described by Dr. Harry Binswanger, that needs to be programmed to simulate the goal-directed behavior of a life-form is therefore as follows:₂₅

Life-form Identity --> "action1 --> goal1 --> survival1 --> action2 --> goal2 --> survival2" --> actionn --> goaln --> survivaln and so on, until the end of its life-span.

Each instance of survival occurs at a later time and is caused by the action with the same number subscript. The sequence can be thought of as spiraling into the future, causing the life-form's own future survival. If this sequence is interrupted in a biological life-form, it dies and ceases to exist. A simulated life-form must likewise be erased from the computer program (or otherwise made inactive) to simulate its death and status as a conditional object.

• The complex, teleological causal process of living objects involves not only the first step (1) the Life-form Identity -->> Action sequence, but also (2) a goal, and (3) maintaining the state of survival in every instance (meaning the goal is successfully achieved and the causal chain is unbroken). In addition, teleological causality is a cyclic process, with each new cycle depending on all the previous ones to create the spiral effect as time passes.

Teleological causality is more complex not because of some intrinsic "vitality" or supernatural power, but because it involves additional simple steps in a sequence that must be continuously maintained, as opposed to the simpler "Identity -->> Action" sequence of mechanistic causality. The additional steps explain the behavior observed in biological life-forms, and these steps must be duplicated in a digital life-form simulation program for it to work successfully, for it to simulate life.

The actions involved in a life-form simulation program are any action that a biological life-form would normally take to causally maintain its existence or reproduce itself, to select, attain, and maintain its values. As with mechanistic causality, the causal or action capacity of a digital life-form stems from its identity. The sequence is the same for all actions and is shown in a traditional computer programmer's flow chart in Figure 5-1 below.

Failure must always lead to death (erasure or deactivation) because life depends on values (disvalues always cause death) and success always leads to survival because values are also causes, and the same causes always have the same effects.

This process is the first layer of the complex causality of life-forms and the interface between mechanistic and goal-directed causality as a form of action. The process is the same for simple biological life-forms such as single-celled organisms or complex, multi-cellular life-forms such as plants, animals, or humans.

The process is complex causality because it must run continuously to maintain a DLF's existence, and it is internal to the DLF. The teleologic of this system is different from a simpler, state of the art, mechanistic automaton or agent because their programs do not need to be acting continuously to maintain their own existence.

Notice that in goal-directed behavior, actions may or may not be pre-defined, but are not pre-selected; rather any action or action sequence is permitted so long as it aids in the goal of survival, and the action selection process occurs in each cycle of goal-directed action; actions are only limited by death. It is the need to avoid death above all, that serves as the criteria for action selection. Goal-directed behavior is primarily active; it is forward looking and solves the problem of action pre-definition by setting life as the standard of action selection. The arrangement is built into the pleasure-pain systems of biological life-forms that allow them to function automatically (in the biological sense); it enables them to proactively do whatever is causally possible to survive.

This fact and mode of operation is what makes automatic teleological action different from automatic mechanistic action; the identity of life processes differentiates a living cell from a mechanistic automaton.

And there is another important difference in teleologic: Only the actions required for survival are necessitated, meaning necessitated by survival, by the need to act to stay in existence against the alternative of death and non-existence. This is the case because of all possible actions open to a life-form, some actions cause life, some cause death, some cause neither life nor death in certain contexts, but can be neutral, and neutral actions are optional. They have that status because survival is assured for some period of time (by necessitated actions that were taken in the past), and therefore no necessitated actions are required in the present.

Note - Actually, value-significance is a highly contextual idea. In principle, everything has either positive or negative value-significance to a life-form because everything has the potential to influence its life for better or worse. However, such evaluations can be calculated in many ways depending on immediate contextual circumstances, allowing for neutral evaluations and optional behavior. For example, putrid water or urine is of negative value-significance to a healthy life-form with a good water supply, but of positive value-significance to one that is dying of thirst. Some contexts, therefore, especially time related ones, determine which actions are neutral to survival and hence, optional.

For example, an animal near starvation in the wilderness is necessitated to live hand-to-mouth in order to survive. However, an animal living in a rich environment with no predators could afford to do nearly nothing for the its entire life-span and still lead a comfortable life (as some people's pets do). In the former case, nearly every action is necessitated by the scarcity of values; in the latter case, whether the animal slept or played or ran around a tree until it was exhausted is optional because of the special circumstances of its context: The animal is in a safe environment of abundance in which little action is required to get the values needed to maintain its life, so it has many options.

Since actions that cause death are eliminated from the DNA of life-forms over time (the dead life-forms are not around to repeat the actions), ultimately, only two types of action remain available to any given life-form: survival actions (which are necessitated), and optional actions (which are not).

This fact is the source of optional behaviors in life-forms, and ultimately, volition in human beings.

5.4.4 How to Write Your Own Goal-Directed Program

A skilled programmer needs only a reasonably powerful desktop computer with at least 64 MB of memory, a 1 GB hard disk drive, and an object-oriented programming environment to write a goal-directed program that simulates a simple digital life-form.

Writing a program to simulate goal-directed behavior on a computer system amounts to creating a Digital Life Form (DLF) and a simulated environment in which the DLF will live. Simple simulations involving a few thousands of simulated percepts and 100 or so simulated conceptual chains would require less computer resources and could be done on a high-end PC, but complex simulations of higher life-forms that involve millions of percepts and 20,000 plus conceptual calculation chains for simulated natural language understanding could require a more powerful computer system such as are used for large Internet servers.

Note - A simulated or virtual environment can be made very sophisticated and is easier and less expensive than using a real one because it can exist entirely in a computer's memory, so no external sensors or actuators are needed. To simulate high order functions such as rational consciousness accurately, a DLF would eventually have to interact with the same world human beings do, including interaction with people, but simulations of simpler DLFs do not require real world contact. However, both simple and complex simulations that use external robot technologies are possible with today's technology, and will become even more realistic in the technical improvements that will come in the near future.

To create a DLF using an object-oriented computer programming environment requires a digital life-form program object be defined with suitable attributes such as a name, an age, an initial supply of Energy Packets (EPs) to simulate ATP (the "fuel" of biological life-forms), and so on, to enable it to act by its own internally controlled power source so it can sustain its own future actions. Many other attributes or properties can be added to make more complex DLFs, but the ones just listed are sufficient for a simple life-form simulation.

The novel idea here is to replicate the essence of the physical design of biological life-forms, namely the facts of their conditionality, having an internal fuel source, and having internal control of action, in virtual form as part of a teleological system; the teleological system can then be animated by the mechanistic causality of a computer in a manner similar to the way in which biological life-forms are animated by the mechanistic causality of physics and chemistry.

Given this design, as with their biological counter-parts, maintaining an adequate energy supply becomes the basis for all other actions a DLF may be capable of performing; therefore once the DLF programming object has been created and defined, processes called methods (object oriented-computer programming code) must be defined to enable the DLF to take action and an action selection method to enable internal control of its actions to find simulated food in its simulated environment to generate more energy packets. This must be a continuous process to enable the DLF to survive, just like a biological life-form.

Some action methods such as Find Food, Eat, and so on are pre-defined and designed to be just like the ability of animals to move their sensors and limbs (so the simulation system does not have to recapitulate evolution), but action selection by the DLF is not pre-determined to any specific action. The goal implicit in the program is for the DLF to take the action necessary for it to survive, to select an action from several alternatives at each iteration of the goal-action cycle, but which specific action it activates is determined indirectly by the DLF's action selection methods based on its life status, its values, and other strategies explained in later sections. These methods are automatic (in the teleological sense), but provide a range of alternative options, and the control is internal to the DLF as is the energy to take the actions.

As with the DLF program object itself, the program objects in the DLF's simulated environment must be created and defined (to save resources and make the system simpler during initial development), but since these objects are non-conditional (non-living), most need few action methods for simple reality simulations. More complex and sophisticated simulated environments in which non-living objects are animated (or contain other DLFs), would however, require coding extensive action methods for those objects.

In the example shown here in Figure 5-2, food is the shaded objects or word objects, the other objects are not food and will not generate EPs if "consumed" by the DLF. Or, alternatively, all objects could be food, but be assigned different values so some objects are more "nutritious" to the DLF than others. How the environment is defined depends on the programmer's purpose for creating the simulation (provided that definition is reasonably consistent with reality). So keep in mind that the environment objects shown are just suggested examples, not required specifications.

To survive, the DLF needs only to act to find food and eat it, thus generating fuel and sustaining itself for future action; these actions are necessitated by its conditional nature. The computer program code for finding food is to "look" and perform a simple object search in the simulated environment to identify food objects.

Note - The ability of a DLF to "Look" for food assumes a working sensory/perceptual system (automatic consciousness), which is explained in a later section of this chapter.

The program code for the Eat method can automatically include digestion, generating EPs, and the simulated feeling of being "full." The code for the Stop method is a simple loop that continuously tests for feeling of fullness, and stops the Eat method when that condition is met. The code for the Death method erases the current DLF from the computer's memory and calls the Birth method which increments the DLF name attribute by one and resets the other attributes to initial conditions. An example flowchart is shown in the example in Figure 5-3.

One objection to a teleological design that could be made is, why go to all this complexity, when a mechanistic agent could do some task more simply and directly? After all, an automaton that is specifically designed to do something, just does it, without need for food, digestion, a pleasure-pain system, simulated death, and so on.

The answer is that this is true, so long as the mechanistic agent only faces tasks that its programmer has foreseen and put into such an agent's design.

The advantage of the complexity of teleological agents is that by interacting with reality they can find ways to do tasks for which they were not programmed, find ways of ordering their basic, pre-defined actions in new sequences for which they have no pre-defined actions in their design. What is their motivation? The DLF's need to survive.

Teleological agents can do so, while mechanistic autonomous agents cannot, because to act is their primary imperative and their simulated lives depend on it, literally. In this regard, teleological agents are like the earthworm in Dr. Binswanger's example described in the introduction to this chapter, and mechanistic autonomous agents are like the ice cube.

In other words, the complexity is necessary in teleological agents to build in the imperative to act for survival, without specifying more than a few alternative actions in advance. Whereas, mechanistic agents may be simpler, but they can only perform actions that are specified in advance by their human designers.

Modern jet fighter designs provide an analogous, though mechanistic, example: One might ask: Why design an inherently unstable plane that requires a fly by wire flight control system which uses millions of dollars worth of computers to make the plane possible to fly, when a more standard design can be flown with a simple hydraulic flight control system? The answer, of course, is that the fly by wire system is the only type that can be used with the inherently unstable designs of modern fighter jets, designs that are of value because of the maneuverability they offer pilots in dog fights.

Though teleological, simulated life-forms are inherently unstable too, and likewise require more complex system designs in order to survive and do things that mechanistic automatons cannot.

So unlike state of the art automatons, survival for a DLF means action by goal-directed behavior only, which is achieved by the flow chart shown in Figure 5-3; there are no other alternatives. The goal is to generate EPs to build up and maintain an internal energy supply, and the actions it must select and cause are to find food and eat until full. Just as with a biological life-form, securing the energy to maintain the ability to self-generate future actions must be its most fundamental priority; it is a necessary action (as opposed to other actions that are not necessitated). These and only these actions lead to survival (and indirectly make other actions possible). All other actions lead to death, and hence never get repeated.

Note - This does not mean that behavior which is not goal-directed can never occur, it simply means that optional behavior can only occur on a short-term basis, and after a reserve of energy has been accumulated. Survival actions must occur often enough to maintain survival.

The computer program methods described in this section may seem simple, even artificial as mentioned above, because simulating biological life is so complex as compared to using a mechanistic automaton, but their teleological design provides a DLF the necessary limits by which to select actions and with its own, internal motivation or imperative to take those actions in the first place. By replicating the essential processes of biological life, the design makes the DLF self-generating, self-selecting, and self-motivating; these are attributes state of the art computer simulations systems lack.

Another objection could be raised that a programmer gives the goal of taking survival actions to the DLFs and therefore they are no different from other state of the art AL programs. This objection assumes that no human intervention is what defines teleology, that humans cannot provide goals to DLFs, but that is not the case. As with state of the art AI and AL mechanistic agents, DLFs are being created for human purposes and some of the actions of DLFs will therefore be for human values. However, this is not what makes DLFs teleological; what makes them teleological is that:

These are the three essential causal aspects of teleological or goal directed behavior, and the DLF program meets all of them. The fact that the author shares the value of wanting DLF's to live and prosper or that a programmer programmed the values is irrelevant. What is relevant is that it is necessitated that the DLFs gain values for themselves, or die, and they control the selection of and provide the energy for the actions to do so.

DLFs are driven by their own internal needs that result from the interaction of the alternatives in their program design with reality over a period of time, as opposed to mechanistic agents which are driven by the actions specified in their program design alone.

The methods described for DLFs effectively add another layer of causal complexity, teleology, to standard, mechanistic object-oriented computer software: In this simulation system, the existence of DLFs is conditional and only DLFs that conform to the teleological criteria are able to live and act. Therefore all other possibilities are eliminated, and all other types of behavior are eliminated over the long-term; thus there is no need for DLFs' behavior to be pre-defined, other than to provide them with a few basic actions and the imperative to act.

In biology, this is the function of death, to eliminate non-survival or anti-life behavior, and it works precisely because life is conditional. This simulation system simply mimics the natural process in digital form.

Note - The programmer will also need to determine how to pass on pro-life behaviors learned by DLFs and maintain them between generations. This is an issue I have not addressed in detail. It could be done several ways and may require some experimentation. For example, it could be done by not erasing pro-life behaviors from memory at simulated death, simulating genetic evolution to carry the behaviors forward to the next generation of DLFs by simulating a "reproduction" system, or by some other means. The main thing is that the anti-life behaviors must be erased when a DLF "dies" or the process will be self defeating.

5.4.5 How a DLF Differs from the Current State of the Art

Now that I have given a specific description of a teleological simulation system, the basic problem it solves in the fields of AL and AI can be more clearly stated and contrasted with current state of the art.

The difference between the DLFs in this program design and current state of the art AL and AI programs, is that the DLFs are teleological by design, whereas state of the art mechanistic automatons that may appear similar in some ways, but the similarities are only superficial because they are not the result of intentional teleological design, but from copying an isolated aspect of a life-form, and life-forms are inherently teleological.

The problem of simulating a life-form is not primarily one of designing a system that behaves like an animal eating by plugging into the wall to recharge its battery, behaves like a human being playing chess, that senses and exhibits human facial expressions, that acts like a group of animals or people solving a problem as a team, modeling genetic evolution in a simulated eco-system, or in the words of Patti Maes, "fast, reactive behavior" that "models life as it could be." These are interesting computer programs that model some isolated actions of life-forms, but they are not essential to the problem that needs to be solved to animate an intelligent life-form using a computer simulation system.

The problem is not primarily one of specifying and automating specific, isolated behaviors.

The problem is to limit behaviors, to create a certain kind of relationship (conditionality) between the simulated life-form and the world it lives in, between the life-form and reality that is similar to that of biological life-forms, and then programming the DLF with the means to maintain the conditions its life requires. The problem is one of limiting the endless number of behaviors a simulation system is capable of to only goal-directed behaviors without having the impossible task of having to specify them all, and all the conditions in which such behaviors will or will not be activated.

This is what goal-directed behavior does for biological life-forms: It automatically limits their behaviors to only those that cause survival. And it does so by the drastic means of simply eliminating any life-forms that perform any other kinds of behaviors.

Life-forms have the imperative to act to satisfy their own greatest needs, whereas automatons are the simple cascade of a set of program instructions being executed.

The kind of relationship that this invention provides between DLFs and their world is the same kind that biological life-forms have with reality: a teleological relationship. To quote Dr. Binswanger: "What underlies goal-causation? The fact that only valuable actions get repeated. Why do only valuable actions get repeated? Because the value here is survival value, and to repeat the action, the agent must survive."₂₆

It is the conditional, teleological relationship between life-forms and their environment that automatically limits their behavior to specific action capacities and makes them what they are. The only actions that get repeated long-term are the valuable actions; life-forms that repeat any other kind of actions simply get wiped out and no longer exist.

That is what the teleological software of this invention brings to the current state of the art: A means of automatically limiting the endless number of actions a computer simulation program is capable of, without directly specifying what actions will be taken, where they will be taken, and when they will be taken. As I pointed out earlier, while the suggested simple action methods are pre-defined in the flow charts shown above, the specific action a DLF will select is not predefined, to say nothing of complex actions that could be created by DLFs by selecting various different actions for each action event, thus stringing together a complex action such as Find Food, Eat, Stop, Eat, Stop, and so on to build up a reserve of EPs without triggering the "pain" of being too "full." An action such as that is not pre-defined in the DLF system. Only the imperative to act to satisfy needs is specified, to select some action from a basic set of alternatives, but which specific action or string of them is not specified. That is determined by the internal needs of the DLF.

The ultimate locus of action control for a DLF is therefore internal; they can be guided from the outside, but not controlled like a robot; rather, DLFs must be trained like animals. As with biological life-forms, DLFs that perform any other kind of behavior besides goal-directed behavior are simply wiped out of memory, and therefore never get to repeat their actions in the future.

The AL and robotics researchers attempting to emulate biology cannot help but to include some teleological aspects into their software because biological life-forms are inherently teleological. Their problem is that they are focused on specifying and controlling behaviors using mechanistic programming techniques rather than limiting behaviors by replicating the essential relationship between biological life-forms and reality, by using conditional, self-generating, self-sustaining program objects as per the three test criteria listed above.

The AI researchers ignore the theory of teleology and the nature of consciousness, attempting to simply explain them away as a form of billiard ball causality that can be recreated using an ordinary computer program, and therefore attempt to specify and automate intelligent behaviors of life-forms with ordinary computer software. For controlling behaviors they use either standard programming techniques or cybernetics (negative feedback control systems), neither of which are consistent with life values and goal-directedness, or in the words of Patti Maes:

"Complex behavior is the result of interaction dynamics (feedback loops) at three different levels: interactions between the agent and the environment, between the different modules inside the agent, and between multiple agents.

For example, a simple Braitenberg creature which "loves" the user can be built by making it move in the direction of the user with a speed proportional to the distance from the user. As an example of complex multi-agent interaction, Reynold's creatures demonstrate flocking behavior through the use of simple local rules followed by each of the creatures in the flock."a

These state of the art design and programming techniques are just mechanistic causality; the cybernetics (the negative feedback loop) is a form of stasis maintenance that reduces an error to zero and seeks to keep it there (like a thermostat).

Life, however, is neither mechanistic nor static; life is active. The imperative for cybernetics is error reduction and the maintenance of a static condition, whereas the imperative of life and teleology is action with life as the standard.

What is needed to emulate intelligent life on a computer simulation system or as part of a robot system is causality substitution, to substitute the mechanistic causality of a computer system or robot for the mechanistic causality of molecular chemistry and physics. In addition, and running in a layer over a conventional computer program, teleological software that simulates the causality of the conditional relationship that biological life-forms have with reality, software which insures that only goal-directed behavior with life as the standard of value is possible to a simulated life-form.

The DLF Simulation Technology design represents an advance in the state of the art for AL and AI because it does precisely that: The DLF design emulates the conditional nature of biological life in virtual form.

5.4.6 Creating More Complex DLFs

Simulating simple life-forms is interesting, but simulating more complex, higher life-forms that possess consciousness is much more interesting. In order to do that a mind must be simulated in addition to a body.

In biological life-forms, the mind is an attribute of the body, and as such, mind cannot exist independently of the body. The two are integrated causally. There is no mind-body dichotomy.₂₇

Likewise in digital life-forms, simulated consciousness (the virtual mind) must be part of a DLF's teleological operation, as a state of being aware of reality, a relationship to reality that results from the DLF's own causal sequencing. Simulated consciousness in a DLF depends on a layer of automatic subconscious processes and the goal-directed subsystem layer below it in order to function. This situation is analogous to the way human consciousness depends on the functioning of the physical brain. In fact, the very reason simulated consciousness exists at all in the DLF design is to support the goal-directed process of simulated life by making it easier and more efficient for the DLF to find simulated food and attain other values, just as real consciousness aids biological life-forms in these same pursuits.

Writing the program methods to simulate consciousness is as straight forward as it is for simulating the conditional nature of life, as the next section will show.

5.5 Adding Perceptual Consciousness to a DLF

Perception is the ability of some life-forms to see reality (touch, smell, hear, and so on) as a collection of objects with various properties, and then to use that information to guide their actions for survival, as opposed to using unprocessed sensor output (data bits) as state of the art AI and AL systems do.

Unlike sensations of individual energies or forces such as a plant sensing light, moisture, or gravity, percepts integrate many sensor outputs over time and space into a foreground of discrete objects, each with its own set of attributes (properties and values), which are the objects' identities; these identities are the focus of consciousness and are seen against a background of other objects.

The identities of perceived objects are not names or symbolic representations. Names and symbolic representations are conceptual; they are part of concepts (as defined by Ayn Rand) which are another type of "data structure" that is more abstract; one that is formed by comparing percepts. (How concepts can be simulated will be explained later.)

Perceptual identities are preconceptual; they are the content, the data from which concepts can be formed with further processing. Nor are perceptual identities images. They are the objects of perception (what is sensed) that have been processed into a different form from which the objects exist outside the perceiving life-form, and while that form is a kind of information, it is not a reproduction.

No one yet knows the exact form perceptual identities have in the brains of biological life-forms; sensations and their integration into percepts are neuro-physiological processes that function automatically and subconsciously in the brains of certain life-forms. In a computer simulation of a human perceptual system, however, the most likely form of perceptual identities will be as lists of attributes (characteristics) consisting of properties and measurement values. How to transform sensor output to perceptual identities will be described in detail shortly. Suffice it to say at this point, that percepts are much more compact than raw sensations, which are very long strings of data bits that are usually processed as X,Y coordinates.

Perception is the first level of consciousness, an automatic form of it that allows higher animals to have awareness of the world around them in order to find food, avoid danger, and fulfill other survival needs. Perceptual consciousness is an attribute of the biological life-forms that possess it; it is part of their goal-directed behavior repertoire and is inherently teleological.₂₈ Therefore, perceptual consciousness must be programmed as another form of goal-directed action in any computer simulation of it.

Perception exists in biological life-forms because it offers a distinct survival advantage: Perception reduces the amount of data a life-form must deal with and that in turn saves valuable time. Instead of millions of sensor outputs, a life-form possessing perceptual consciousness sees, hears, smells, or feels food or danger directly and quickly. Having to visually process a piece of fruit or an insect that is automatically made distinct from its background by its perceptual identity is much easier and faster than attempting to identify it as sensations. To be able to recognize and avoid predators almost instantly is an even greater advantage. Imagine yourself in the jungle: Would you like to try to identify a poisonous snake or a tiger using only sound bits or pixels, or would you rather be able to hear the hiss of the snake or see the stripes and fangs of the tiger before taking "evasive action?"

Perceptual consciousness makes survival easier by reducing the number of units of data a life-form must deal with by integrating sensations into percepts. In addition, once the most commonly perceived objects are stored in a life-form's memory, they serve as a baseline for future perceptual comparisons. Since consciousness is largely a difference detector, changes can be easily identified in the relatively small number of objects in any perceptual scene. These same advantages will accrue for a DLF that simulates perceptual consciousness for the goal of its own survival.

5.5.1 Sensing and Acting in a World

The program code for sensing the environment will differ greatly depending on whether the environment for a DLF is simulated or real. The two types of environment are essentially equivalent, except that real sensors sensing reality provide much more accurate and detailed real-time data of the world, whereas simulated worlds are limited to human imagination and computing resources. Simulated environments are primarily useful for developing, testing, and proving program methods while conserving resources. Sophisticated DLF simulations intended for practical uses will need to interact with the real world you and I inhabit to be effective.

Note - Since the use of real sensors and the computer code to operate them is well established in both AI and AL, this description will focus on a simulated environment. Suffice it to say that real sensors and their software could easily be substituted for the simulated environment to be described herein, by a skilled computer programmer with knowledge of robotics.

In a simulated environment created with an object-oriented programming language, sensing amounts to reading the attributes of the environment's objects. This information is then transferred by the DLFs sensing method from the simulated environment to the DLF's perceptual processing methods. The transfer itself simulates sensors in a simulated environment.

Simulating Perception and the Identification of Objects

The identities of objects consist of some number of attributes (These are also called characteristics or features in the fields of AI and AL.), and each attribute consists of a property and its associated measurement value. Attributes must be calculated from the sensor data and integrated into an identity, which is a list of properties and values, to be a percept. Foreground objects must also be distinguished or differentiated from background objects to be identified. Consciousness operates by detecting differences in sensory data and then focusing on the identities of foreground objects.

Much work has been done in the AI and AL fields to devise programming methods to accomplish these tasks, generically known as feature, attribute, property, or characteristic extraction and recognition. In the current state of the art, however, feature extraction and recognition is performed only to satisfy the human goals of targeting weapons or building robots. These processes are not designed to attain the sensing system's own goals (the systems have none), but the human goals of hitting targets or exploring a landscape on Mars, for example. By contrast, while DLFs may be designed to share similar human values and used for the similar purposes, human values will be secondary; the primary values of a DLF must always be to gain and keep what is necessary to maintain its simulated life, such as the identification of objects in the world that can be used for survival. Remember, the locus of control for DLFs is internal, not external like it is for machines.

Note - A lack of values is true even in the case of robots that are designed to plug themselves into electrical outlets to "eat" or those designed to "work together" in packs for "common goals." In all such cases I have seen to date, to whatever degree the design of these systems is teleological, it is that way by accident because they are emulating one or two isolated behaviors of real life-forms, but the human values of acting like a life-form or accomplishing some common goal are still the primary values of such robots, instead of survival with their own simulated life as the standard of value, as it is in biological life-forms; hence, what extant systems exhibit is not goal-directed behavior. As a consequence such robots are mechanistic devices created to achieve the human goal of building a robot. The designs do not satisfy Dr. Binswanger's three criteria for a being a life-form. There are no "values," "value significance." or "actions caused by value-significance to the agent" in the extant system designs. Their "survival" is caused externally by their human builders, not internally initiated, controlled and sustained. Whereas with the design of DLFs, their own survival is their primary goal, with the energy and control originating from inside the DLF; any human goals DLFs may share are secondary to them.

For the purposes of simulating consciousness, the point is that the overall reason for differentiating objects from a background and identifying them in terms of their attributes is that perception provides a data unit economy to the perceiver; it reduces the number of data units the system must process to aid it in its identification of the reality, making it easier for the system to maintain its own survival.

In general, therefore, the process of simulating perception is to use various means to identify attributes in a scene, identify the objects to which the attributes belong as distinguished from a background of other objects, and to produce a list of these attributes for each object in a given perceptual scene (or other perceptual grouping if non-visual). The result is a simulated perceptual "image" or "snapshot" of reality consisting of lists of other lists; that is, a list of the objects that exist and were perceived, each of which is itself a list of attributes or properties and the particular measurements for each attribute.

Percepts generated in this way are not reproductions of the perceived objects, but rather they are transformations of the objects from real objects to informational objects. This transformation is possible so long as the objects' identity is conserved as their new informational (virtual) forms are calculated from their physical forms.

In a simulated environment or world using object-oriented programming (such as I am limiting this description to), a computer interface window can be used as the background for a visual scene, and various objects drawn in the window can simulate real world objects in the foreground. Obviously, more complex arrangements can also be devised where certain objects themselves form the background and others the foreground that a DLF has in perceptual focus, such as non-food objects vs. food objects, as well as dynamic scenes in which objects move in various ways. But the simplest case is all that is needed to explain how perception can be simulated. The essential issue here is to show how the identity of the objects is acquired from reality and transformed into information, into virtual form.

In the typical computer environment, objects drawn in windows are stored as X,Y coordinates, and these coordinates are analogous to sensations in biological life-forms. They are the processing units of simulated sensations.

Note - Extant systems in the state of the art for sensing the real world ultimately transduce the sensed data into some form of X,Y coordinates or other numbering systems. In fact, all types of computer "sensations" are transduced into some form of measurement system, so that every sensation produces one or more numbers or measurement values in one or more dimensions.

Perception in a simulated environment amounts to reading the X,Y coordinates (sensing) and identifying by calculation the attributes or properties the coordinates contain according to well known, established mathematical techniques.

To make a program method to perceive a simulated world, a computer programmer would create a loop that simply repeats the steps of sensing and extracting the attributes of the objects sensed, each of which provides the objects with a unique identity. The result is a cycle of repeated perceptual events, or simulated perceptual consciousness of the simulated world; that is, the awareness by the system of the objects as objects, not as X,Y coordinates.

For example, if a window simulating the real world contains drawings of several different sized circles, squares, rectangles, and triangles, these will be stored in the content attribute of the window in an object-oriented programming environment. It is a simple matter for a programmer to write a method to read the content attribute of a window to get the lists of X,Y coordinates for each object to simulate "sensing this world."

Note - This fact is also why it makes little difference for simple simulations of consciousness if the real world or a world simulated in a computer system is used as a data set. Any real world objects sensed by a consciousness simulator, would end up as ((X, Y,)... N) coordinates anyway. The only essential difference is the amount and complexity of the data, with the real world producing copious amounts of complex data, as opposed to a simulated one.

Once the X,Y coordinates have been "sensed," the next step is to calculate the attributes of the objects. This process involves straight-forward methods of determining if the objects are simple (such as a circle) or composite (such as a triangle consisting of 3 lines) and identifying other attributes (such as if a line is straight or curved, its slope if it is straight, its length, its direction, and so on).

Each of these attributes is a category of measurement and each has a specific value because every object that exists is unique in its properties, its identity. So every object processed by these perceptual methods will produce a unique informational identity consisting of a list of attributes or properties, each of which itself has a unique measurement value associated with it because every property of an object can be measured by some standard. In fact, for a given type of object, it turns out that there is a range of measurements that is typical for each property, but more on that later.

The example I have used here is of a simple visual scene, but it makes no difference for complex scenes or other sensory modalities. The process is the same for all scenes or other sensory modalities such as sound, touch, smell and so on. In every case, reality is sensed, some unique percept is the result that identifies some particular aspect of reality. It is true that not every sensory modality produces percepts of objects (such as sounds or smells), but these can be integrated across sensory modalities as additional attributes of objects perceived in the visual modality. These facts are common knowledge in the field of animal and human psychology.

Vision is the central sensory modality of the highest life-forms. Objects are parts of reality, but reality itself is one integrated whole; it is a plenum (Aristotle). That whole is only broken into objects because consciousness itself is an identity; it is a limited, specific process and cannot take in the world as a single piece. The form information takes as it is processed is the result of the identity of conscious processes; the form of reality is transformed from its natural state to informational or virtual form, but the content itself, the identity of reality, is conserved in this process.

For any aspect of reality that can be sensed with sensors, objects (by means of their attributes) can be differentiated, and a unique identity can be calculated as a means of simulating perceptual consciousness. The simulated environment, the DLF, and the simulated perceptual process can be shown graphically in a conventional diagram and flowchart as in the following figures: Figure 5-4 and Figure 5-5.

The program code for the Look method for a simulated world is simply to select one or more objects in the "reality window" and to "focus" the simulated consciousness on its object(s). This process is analogous to what a human computer user does when selecting an object on a computer screen with a mouse before issuing a command from a menu. The human cannot think of every object on the screen or every command at once and must focus on a specific ones to do something. Likewise the computer cannot process every object at once using every command at once, so the user must select the object(s) and the command the computer is to use for processing them.

The simulated consciousness of a DLF cannot process all of reality at once either, so it must also focus (select or "Look") at only what objects are to be perceived (processed); it must delimit its field of perception to less than or equal to its processing capacity and/or what is relevant to its purpose. This, by the way, is an example of how it can be the case that while a basic set of actions may be pre-defined for DLFs by human programmers so that evolution does not have to be recapitulated, both the action and the content the action processes are not pre-defined, but are determined by the DLF based on its own simulated life needs at the time the action is selected for execution.

Note - In human beings, the choice to focus or not is the essence of volition or free will.₂₉ How focus and volition can be simulated will be described in detail below; the short description is that simulated focus and volition are forms of optional mental actions for a DLF.

Once the objects are selected, sensing is simulated by a Get method that retrieves the X,Y coordinates that are the objects and passes them to the method which in turn calculates their properties and values as lists, and then a Store method that stores these in lists in memory. The resulting lists are simulated percepts: Simulated percepts are a list of one or more object instances, each of which itself contains a list of properties and values; the lists are the perceived objects' identities. These simulated percept lists are the processing units of simulated perceptual consciousness in a DLF; they are its content.

The simulation of perceptual consciousness is all of these methods operating together as a continuous process, constantly repeated by a loop, processing X,Y coordinate lists into the identities of objects, into lists of attributes of objects (properties and measurement values).

Note - The term "object" can be confusing because it is used in two different senses here. An object-oriented programmer would create classes of program objects of which the objects in the DLF's simulated world and its simulated consciousness would be instances. It is important to keep programming vs. DLF simulation contexts of this term clear and the different meanings of the term "object" separated in your mind.

For example, the following explanation shows how objects 1 and 2 in the simulated world shown in Figure 5-4, could be converted to simulated percepts (lists of lists) for storage in a DLF's memory as (P1, P2,... Pn). The length, slope and curvature attributes of these data for objects 1 and 2 in this example were generated using the DLF Program described in Chapter 3.₃₀ The shapes were drawn in the DLF's simulated reality window by the author using a mouse, and then the attributes were calculated based on the X,Y coordinates. Other attributes of line thickness, intersection angles, and object fill could be calculated to have additional or alternative attributes.

Figure 5-6 below, shows the actual X,Y coordinates for a circle drawn in the DLF Program by the author to look like the circle (object 2) shown in Figure 5-4 (without the fill pattern). The circle is small, only about a quarter inch or so in diameter on the computer screen. Obviously a large circle and one with a complicated fill pattern would generate many, many more coordinate pairs. In addition, different strategies would be needed to calculate attributes from filled shapes as opposed to line shapes.

The point of this example, however, is to show one way in which attributes can be calculated from the actual X,Y coordinate pairs of real objects that the author drew in a DLF Program window. In the form of simulated sensations, which are X,Y coordinate lists, objects such as the ones used in this example would be processed by computer sensing software as lists of one or two hundred coordinate pairs, or fewer (as in Figure 5-6). In the case of the small objects in this simple example, the unit economy gain of storing the objects' identities as attribute lists instead of X,Y pairs would also be small. However, for more complex objects or those sensed in the real world using digital cameras and microphones, the coordinates thus acquired could have three or more dimensions and could number in the millions to billions, so the unit economy of converting the objects to simulated percepts (lists of lists of attributes (properties and values)) before storing them and doing additional processing of them would be significant.

In a properly designed simulator, the coordinates could be recovered as needed later by reproducing them from the attribute lists, or by perceiving the original object again.

Figure 5-7 shows the attributes for the example triangle and circle (objects 1 and 2 from Figure 5-4) calculated from their respective X,Y coordinate sets. The closure attribute is FALSE for the three lines of the triangle (a,b,c) because the "composite shape" method was not yet working in the DLF Program at the time the author used it to produce these data. There was also no method for calculating the attributes of angles; however, the program was working well enough at the time to demonstrate the idea of calculating attributes from the X,Y coordinates used to simulate sensations.

The unit economy gained from this method is content-oriented data compression because it results from the way the content is processed and from its final form, rather than the analysis of the data's bit patterns.

Note - The very large number for the slope of line "a" simulates the "infinite" slope of a vertical line.

The process of simulated perceptual consciousness described herein, is not the same as how the consciousness of biological life-forms processes the data they sense in the real world. However, within the context defined for simulating perceptual consciousness, the process described here is causally equivalent insofar as the objects' identities are transformed from their form in reality to the form of information inside the simulation system, as they are converted from real objects to various kinds of processing units (informational objects). Given objects to sense, a DLF will always produce simulated percepts of those objects as lists of properties and measurement values. The motivation for the DLF to perform the simulated perceptual process is teleological, but the means is mathematical and always certain to produce a result.

Note - The form of the percept of an object is different from the form of the object itself, though its identity is conserved. (See the references₃₁ on the form/object distinction.)

It is a well known fact of mathematics that given X,Y coordinates of objects, various properties can always be calculated. Since computer sensors with proper software will always produce coordinates if given objects to sense and process, an identity in the form of a simulated percept will always be calculated for any object "sensed" in this manner, so long as the program methods are properly designed.

I pointed out earlier that Ayn Rand observed that "Existence is Identity." This statement means in effect that to be an object is to be some properties and values of some kind. In other words, every object is valid content for consciousness because every object is some set of properties and values and will therefore always produce a processing unit that falls somewhere in the appropriate measurement range for a given type of object. Not to do so is not to exist.

Ayn Rand also observed that "Consciousness is Identification."₃₂ Since perceived objects are the content (data) of consciousness, this statement means that consciousness transforms the identity of the objects it perceives into a form of information, into a mental, epistemological form in the human mind, as opposed to a metaphysical form. The objects exist in reality; the information exists in consciousness as content; it exists in the form of information and is true about the world so long as each perceived object's identity is conserved by conscious processing. In other words, the objects are metaphysical (part of reality), the percepts are epistemological (part of the content of consciousness), with latter being the information that makes up a conscious entity's knowledge.

Information IS identity in conscious form, as the content of the consciousness attribute of some life-forms. Moreover, it is specifically in reference to human consciousness that the term information gets its meaning.

When you or I look at objects 1 and 2 in Figure 5-4, the objects are converted into percepts that are stored in our memories by our perceptual consciousness; the identities of objects 1 and 2 are the content of that particular conscious event for us. The properties and values we see as the identity of the objects are stored in the form of perceptual information in our memories; they are the processing units at this level of consciousness. This process is what a DLF's simulated perception process imitates using various calculation and storage methods.

The method of the simulation of consciousness I have just described in this section similarly enables a DLF that is animated by a computer system to convert the identity of objects that exist in reality (simulated or real) into identities in the form of simulated (calculated) percepts, which like their counterparts in the consciousness of you and I, are also a form of information. In the case of the DLF, the processing is performed by a computer simulation system, and the result is stored in the computer's memory instead of the mind of a biological life-form, but the end result is causally equivalent as long as identity is conserved as the following example shows. Though strictly speaking, it is only the context of human consciousness that makes this content "information," in a simulated sense, it functions in a manner similar to how it would in a person's mind: It enables the DLF to be "aware" of what is in its world by conserving the identities of objects and then storing them in memory.

Let's assume for a moment that a DLF is using a digital camera to take pictures of the real world human beings see instead of sensing a simulated world, and that objects 1 and 2 in Figure 5-4 are in the view field of the camera: The identities of objects in the field of the camera are converted from real attributes into measurements; that is, attributes in the form of real objects (the objects' physical, visual identity) are transferred to the camera by light waves and are converted into pixels that consist of (X,Y) coordinates plus color by the digital camera's software (one form of information); in effect, the information about the objects identities carried by the light to the camera is transduced and stored as pixels in the camera's memory. The pixels are then output to the DLF and are further processed by the DLF's simulated perception software into attribute lists consisting of properties and measurement values (another form of information), and this is the form in which the DLF is aware of the objects in the field of the camera; all of the information about the objects carried by the pixels to the DLF's simulation of perceptual consciousness is now in the form of attribute lists, simulated percepts to the DLF, but the identity of the objects in the camera's field is all still there, just in a different form. Though the specific means of processing and holding visual information may differ in biological and digital life-forms, it is still information about objects 1 and 2 as shown in Figure 5-4, information in which the identity of objects 1 and 2 has been processed, conserved, and stored in memory.

Evaluating Objects

Once various objects have been perceived by a DLF, they must be evaluated with the DLF's life as the standard of value. To a biological life-form, since its continued existence is conditional, every percept is either a value or a disvalue relative to its life; that is, every percept has value-significance to the life-form as being information about its world that is either for or against its life. In order for a DLF to be an accurate simulation of a life-form therefore, a DLF must also be able to determine the value-significance of its perceptions.

Note - As noted earlier, everything has value-significance to life-forms, but how it is determined is highly contextual.

The pleasure/pain systems of biological life-forms are automatic, built-in value systems. In general, things that are good for a life-form cause it to feel pleasure, and things that are bad for it, cause it pain (either physical, emotional, or both). These are well known facts of biology and psychology.

In order to create a digital simulation of a life-form, a similar automatic, built-in evaluation system is required, and like a DLF's pre-defined actions, may be copied from biological life-forms and pre-defined so evolution does not have to be recapitulated. Since computer systems are not biological, but digital, simulated pleasure and pain must be calculated based on simulated values which serve as standards with the life of a DLF being the ultimate standard; this is the essence of life and goal-directedness. The idea is to make simulated evaluations as causally and functionally equivalent to the biological ones as is technically possible.

Note - Obviously, a computer simulation can never actually "feel" pleasure or pain as it is not conscious, only an imitation of consciousness; a simulator can only calculate simulated pleasure or pain measurements. And while this fact will not prevent the practical application of DLFs to many useful functions, it means DLFs can never have quite the same perspective as biological life-forms. Only experimentation can tell us how this fact will affect a DLF's behavior.

For example, in the flowchart in Figure 5-3 above there are decision boxes that test if a DLF "feels full" or not. What this means in computational terms is that a method must be written that compares the number of Energy Packets (EPs) that a DLF has with the range that its simulated life requires. Having EPs is a value to a DLF's life; without them the DLF will "die" just as a biological life-form will die without food.

Note - EPs can be thought of as the digital equivalent of Adenosine Triphosphate (ATP), which is the "fuel" of biological life-forms. Reminder: The term "value" as used in this context (as in value/disvalue pair) means value to life, and does not mean "value" as in number.

When a DLF is "born," it starts out with some number of EPs, say 100. From then on every action it takes uses EPs, and every time it "eats," EPs are replenished. If the DLF adds the same as or more EPs than it consumes it survives and "lives" on; if not, the DLF "dies." What the evaluation system does in the form of simulated feelings of hunger and fullness is tell the simulated consciousness of the DLF in a form of information it can process and use, its life status at a given time, like a fuel gauge.

Biological life-forms (including humans) are not aware of their life status directly, but through feelings such as hunger vs. fullness, and other value/disvalue pairs. Hunger is experienced as pain which eating makes go away. Hunger is replaced with fullness, which is pleasurable, unless the life-form eats too much, in which case fullness can become pain as well. If a DLF is given a range of EP values from 100 to 1000 say, with 400-600 EPs being defined to generate a neutral feeling, having less than 400 EPs will produce stronger and stronger feelings of simulated hunger, and having over 600 EPs will produce stronger and stronger feelings of simulated fullness.

A simulated feeling can be calculated for any number of EPs a DLF has at any specific time by comparing the number it actually has to this range. The simulated feelings could range from say -9 for starving to +9 for too full, and 7 or 8 for comfortable fullness. According to the definitions just described, if a DLF had 250 EPs a simulated feeling of hunger of -5 would be generated as shown in Figure 5-8, if it had 700 EPs, the DLF would feel full and feel at perhaps a +8, and if it had 950 EPs it would feel the "pain" of greater than +9 fullness.

Similarly, simulated feelings can be calculated for any number of other value-disvalue pairs, such as interest vs. boredom, company vs. loneliness, clarity vs. confusion, activity vs. laziness, confidence vs. fear, and so on. Any simulated feeling calculation would compare some numeric standard of value (usually a range of numbers) against whatever number a DLF has at any given time relative to the state of that value in its simulated life, and the result will be the simulated feeling. Pain can be simulated by using negative numbers (or positive numbers beyond a certain range), neutral feelings by zero, and pleasure by positive numbers in a range from -9 to +9 (or any other range) as suggested above. In addition, the overall "happiness" of a DLF can be calculated as the average of all of its other simulated feelings.

Note - Reminder: The term "value" as used in this context (as in property and value list) is a computer programming term, and does mean "value" as in number, a measurement value.

A programmer skilled in object-oriented programming can make simulated feelings attributes of a class of DLF program objects. For any instance of a DLF, the property and value list might look like, though would not be limited to, the following:

The simulated feelings give the DLF an instantaneous indication of its life status, (and if put into a window on the computer screen as part of a DLF program interface, a human observer can see the same status). By being conscious of its own life status, a DLF can take actions to cause its future survival, since it would have the information that is a prerequisite to such actions. Simulated feelings are the simplest form of simulated self-awareness or self-consciousness, though at this level a DLF is not aware that it is "aware" of itself.

Actions and Objects

Life and consciousness are processes; active processes that consist of a series of actions. Biological life-forms continuously act upon objects to maintain their survival by finding and eating food, finding shelter from weather objects (such as raindrops and lightning bolts), running away from predators, and so on.

Similarly, DLFs need to act out comparable processes to maintain their simulated lives in order to maintain causal equivalence with biological life-forms. In fact, as pointed out earlier, at the level of perceptual consciousness, speedy action selection is the main survival advantage consciousness has to offer biological life-forms.

Actions are not pre-selected, but selected by the simulated perceptual consciousness process, and as with its biological counterpart, this process is an automatic one (in the teleological sense): There is no other basis for making selections because options are limited at the perceptual level. However, action selection is teleological because its goal is a DLF's survival, the DLF's simulated life is the standard, and it, therefore, cannot be explained as simple, mechanistic causality.

DLFs must have a list of actions they can invoke in their simulated world to effect their survival, depending on what they perceive at a given time, such as look, eat, move an object, compare objects, draw an object, type a word, and so on, just as a biological life-form does, such as nest building or swimming. These basic actions and more complex composite actions DLFs may construct by stringing basic actions together are primarily used for survival and secondarily for optional behaviors or to accomplish secondary, human values that DLFs may share with people.

Note - Remember, what I am talking about here are the processes internal to the DLF that simulate life, processes that must be maintained by the DLF to survive (Dr. Binswanger's 3 criteria), and that this fact is what makes teleological causality a more complex form than mechanistic causality.

Actions are selected based on how a given percept and the life status of actions capable of changing that percept are evaluated. The efficiency with which a DLF can process percepts has a direct effect on a DLF's life: The more efficient its processing of simulated percepts, the more likely it is the DLF will survive. Early in a DLF's life, when there are few examples of percepts and how the DLFs previous actions changed them, most of the DLF's actions will be selected by trial and error. However, after perhaps six months of life and many thousands of perception-action events, the action selection methods will have much more data to use and will therefore be able to select actions with the greatest survival value more efficiently.

The simulated feelings a DLF calculates for itself, as described in the previous section, are the primary data for a DLF's action selection process, along with its simulated perceptions of its world. Actions that are followed by increases in positive simulated feelings can be rated with a positive index or associated with the positive simulated feelings to make them more likely to be selected in the future in similar conscious events; the opposite is true for actions that result in negative simulated feelings. As with real feelings in biological life-forms, the simulated feelings of pleasure and pain calculated by DLFs provide an instantaneous indication to the DLF of its life status.

There are a number of strategies and measurements for implementing such strategies that a skilled programmer can use to simulate automatic action selection in a DLF, and more than one will be needed for a DLF to survive even in a simulated world, because life thrives on alternatives. Some examples of action selection strategies that have been observed in biological life-forms (including the author) and copied so DLFs do no have to recapitulate evolution are as follows:

Note - Recall the screen shots in Chapter 3 from the DLF Program of one example of how to implement these strategies.

Other selection strategies could be invented and used, but these are sufficient for a skilled programmer to write an action selection method that will enable a DLF to always be able to select some action for any given percept and therefore simulate the biological imperative to act. The latter is important because life (simulated or real) is a process that cannot stop: Life-forms that have no active process are dead.

While it is up to an expert programmer to determine the specifics, the most probable design for an action selection method is a series of program conditionals. An example of how action selection can be simulated is shown in Figure 5-9 below. Remember however, that actions must be selected based on the needs of the DLF's life, not pre-programmed specifications, which are effectively arbitrary from the perspective of the DLF's simulated life.

The DLF perceives its simulated reality or world and calculates its simulated feelings as described in previous sections above to determine its life needs at that moment. Then a conditional compares its simulated feelings to find if any are in the "near death" or "very unhappy" part of the range, and hence require "emergency" action. If so, emergency actions are taken to insure the DLF's survival, such as eating if the DLF is almost out of EPs.

If not, the DLF's happiness value is compared to its typical range of values, and if it is near the highest value, no action may be necessary (for survival) so optional actions are possible, or a random action may be indicated to provide the DLF with some new perceptions. A calculation is done, and either the No_Act method (a method which does nothing) is selected, or the Random_Act method is selected (which selects an action from the DLF's action list using a random number generated by the computer). The calculation can be based on the length of time since previous new activity has occurred or some other standard; the specifics are up to the programmer and the design requirements of the simulation system that is being created.

If the DLF's happiness value is in its mid-range, then a method is called which calculates an action based on the other strategies listed in the bullet items explained above with various similar conditionals to select between them, such as continuing its current action, calling an action that is associated with an increasing positive simulated feelings in the past, a complex, pre-programmed action, and so on.

• The action selection method is teleological in that its goal is causing the survival of the DLF with its simulated life as the standard of value, and it does so by increasing the DLF's simulated happiness. It provides the DLF with a means of self-regulation using its own energy to cause its own future goals to be achieved. In other words, only necessitated survival actions and pro-life optional actions get repeated in the long-term; all others are effectively eliminated from its repertoire by the "death" method as explained earlier.

• The No_Act and Random_Act methods allow a DLF to maintain its simulated happiness for a time, provide for trial and error actions, and allow for the "unexpected" or the novel to be simulated in a DLF's life, as well as optional actions which are not necessitated by survival, but are caused by the DLF simulation system.

• At the perceptual level of simulated consciousness, a DLF is capable of difficult to predict behavior due to its complexity and that optional actions possible, but it is not capable of unpredictable, volitional behavior. Its actions, though purposeful, are largely predictable, with the exception of the No-Act and Random-Act methods. And even these are predictable, but for a narrow range.

• These strategies taken together provide a basic group of behaviors, the specifics of which are calculated at the time of their execution, and that enable a DLF to respond to conditions in its simulated world. They are analogous to the automatic, genetically determined behaviors found in many biological life-forms that have evolved over many thousands or millions of years, as opposed to the ontogenetic, learned behaviors such life-forms may acquire during their lifetimes. Since we know these behaviors work in real life-forms, it is safe to copy them and pre-define them in DFLs, so long as care is taken to maintain their teleology.

Note - The issue concerning what happens to behavior and other memories when a DLF dies will be discussed in the section on memory.

• A key idea in this invention is that of leveraging the results of millions of years of biological evolution; this is precisely the point of directly programming the simulation of the perception/action and pleasure/pain systems of biological life-forms in DLFs. These systems of automatic consciousness and action selection are the causal layer that support and cause the rational intelligence of the conceptual layer above. The hard work has already been done by biology; we need only reverse engineer it.

Once an action is selected, its method is called, any necessary data is retrieved or calculated, and the action is executed by the DLF to cause changes in its simulated reality (or the real world, depending on the simulator design), thus closing the system.

Memories

In order for a DLF to be a realistic simulation of a biological life-form, it needs to have memories of its past perceptions, feelings, and actions, of its simulated life.

For example, using the DLF described in the section on simulated feelings, we have the information shown in that section, plus the action selected (which would be the Eat method in this case because Hunger is its most negative feeling). The memory for the DLF for that instant in its simulated consciousness is therefore:

One example of a memory record for this DLF at this instant in its life is as follows:

(006023, 84,100,350, P1, P2,... Pn, Look, Find Food, Eat, Stop, Eat, -2, +3, +2, +5, -1, +2, 1.5).

Note - To make such memories possible, a programmer could use any of a large number of standard database record formats and processing methods to store the DLF's percepts, simulated feelings, the actions it selected, and so on; the specifics of these choices are up to the programmer. (The Eat method is listed twice because it is one of the available actions and the action selected to be implemented in this instance of consciousness.)

With the older technology of the recent past, the amount of data a DLF generates as memories over its simulated life would have been a storage problem, but that is no longer the case with the huge amount of storage space available now, even with today's personal computers.

Still, a skilled programmer must be smart about how the DLF's memory is designed to make the most efficient use of its memory. If a DLF perceives its world X many times every minute, it will generate X+ many percepts, most of which will be the same because the world will not have changed since the previous percept. Methods for changing the simulated world and deciding how many duplicate percepts should be kept in memory and for how long will need to be determined by the programmer, and will depend largely on the purpose of a given simulation and the memory resources available. This is not an issue that is relevant to the simulation of consciousness as such.

Note - Lack of change will be less of a problem for simulations using real world data because the real world is constantly changing; however, the amount of data generated by sensing the real world will be significantly greater than that of a simulated world.

For example, the memory may need to be broken into short and long-term storage, with much of the duplicate short term content deleted or over-written after some period of time and only the non-duplicate content save to long-term memory. Other strategies can also be employed, such as indexing schemes that are commonly used in databases to make any memory easier to find, and so on.

It is also important for the programmer to make sure the memory is stored in such a way as it will persist as long as a given DLF is alive, since in the object-oriented programming environment, not all data persists when an object instance is not active. Memories can be stored as class attributes of DLFs or in some other form of persistent data structure.

Another issue is what happens to a DLF's memories when it "dies." With biological life-forms all memories are lost, which is a terrible waste of information. On the other hand, the very purpose of death in evolution is to eliminate behaviors from memory which are not productive of survival or which are anti-life, so that they are never repeated. I have not included "reproduction" in this description because how pro-life information and actions are maintained in the simulation for future DLFs is a practical, programming matter, and it is not crucial to the description. The exception to this is that however the "death" simulation method is designed, it must be designed such that it eliminates anti-life behaviors from the action repertoire of DLFs and meets the criteria described earlier of being teleological.

Since this invention is a form of virtual reality, it will be up to the programmer who writes the methods for the DLF system to decide what strategy to use to simulate death. In biological life-forms, long term behavior memory is controlled by evolution through the gene pool. Ontogenetic behaviors, those learned during a lifetime, are controlled and modified by the pleasure/pain system. To deal with this issue, either an approximation of natural selection could be simulated, or some other method could be devised to cull unproductive behaviors from a DLF's memory when it dies, while leaving other, valuable information in place to be available for future DLFs so each new DLF does not have to re-learn everything its "ancestor" DLFs had already "discovered."

Finally, there is the issue of processing unit economy. Even though simulated percepts of objects offer much greater unit economy than X,Y coordinates or pixels, there will still be a huge and always growing number of them, even if memory conservation strategies are employed. (Another means of processing unit economy will be explained in a later section that will make DLFs more efficient at processing content by dramatically reducing the number of units they must process and will expand the capacity of their simulated consciousness.)

Action in a DLF's World

Once the results of the processing of the Perception, Evaluation, and Action Selection methods have been stored in memory by the Memory method, the Action method executes the action selected during that simulated conscious event and causes changes in the DLF's world.

Executing an action involves calling the action method selected earlier in the conscious event and effecting whatever change that method is designed to make in the DLF's world. For example, the Look method would perform a search for an object, the Move method would move an object, the Eat method would extract "nutrition" from an object, the Draw method would draw an object, the Type method would type a character or a word into the DLF's world, and so on.

Note - Some of these methods, such as Draw or Type, could be easily created by simply calling methods that are already part of an object-oriented programming environment and providing them with the appropriate data. Others, such as the Eat method, would have to be written completely by a programmer. And remember, it is why an action method is selected and its content is specified by a DLF relative to its own survival needs that determines its goal-directedness, not if the action is reverse engineered and pre-defined by a programmer.

The data for what object to draw or what characters to type would come from memories of previous perceptions of these objects in the DLF's world and the DLF's immediate life needs. Just as a human child first draws or writes what it sees around it, so would a DLF at an early stage of its development; in later stages, the data to be drawn or typed could be combinations it makes from its memories of percepts, combinations that it may have never actually observed; such "made up" combinations would be a form of simulated imagination on the part of a DLF.

When an action such as Draw or Type is selected, part of the action selection process must be to find the appropriate data for that method in the DLF's memory, data acquired by previous perceptions. If no data is available, the action selection process fails and a different action for which data is available must be selected. This ensures that every action selected is a valid action and can be executed.

A method such as Draw, is simply the reversal of the process of simulated consciousness: The Draw method takes the identity of an object, which is stored in the form of attribute information in the DLF's memory, and transforms that content back into an identity in the form of a real object or a change to a real object in the DLF's world. As with simulated perception, except for changes to objects, identity is conserved in a DLF's actions just as it is in its perceptions.

A flowchart of the essential functions of the Action method is shown below in Figure 5-10.

Note - If the DLF's world is the real world, the Draw action could activate a printer or plotter to physically draw an object, or in addition, the DLF might have an action called Make, which could employ machine tools to transform the DLF's perception of an object into an actual, physical object or change an existing one.

The final function of the Action Method is to call the DLF's life processes in the subsystem layer below it (layer 5 in the model shown in Table 5-1). The life processes then calculate the EPs used by the action just executed, the number of EPs remaining, and if the DLF has enough to survive, as well as calculate other life needs. If there are not enough EPs, the DLF suffers a simulated death (which is not shown in Figure 5-10, but was described earlier in this chapter). If the DLF survives, control is passed back up to layer 6, and the Perceive method is called to begin the next perception of the DLF's world. The percepts that result will close the system by showing the DLF the effects (changes) to its world that the action it just executed have caused.

5.5.2 The Conscious Event Cycle

From the description in the previous section, it should now be clear that simulated consciousness is a series of discrete, causal steps performed by program methods that repeat or cycle, operations a programmer turns into a process by putting them into a loop internal to the DLF to simulate its life and consciousness; the program continuously cycles through these several program methods, thus effecting the simulation. The process steps to simulate consciousness run in a subsystem layer above those of the DLF's simulated life processes (layer 6, see Table 5-1) and the program methods that implement them are:

To simulate consciousness, the methods in the list are continually repeated so long as a DLF survives, as shown in the flowchart in Figure 5-11 below, in something I call a Conscious Event Cycle (also called the C.Event cycle), and each process run through the loop is a simulated conscious event, or "C.Event."

From the perspective of a DLF, the seeming continuous nature of consciousness comes from the rapid repetition of C.Events in this process, like the seeming continuous nature of a movie or video comes from the rapid changing of frames of still pictures, from the perspective of human consciousness. This process can continue indefinitely as long as the DLF is alive.

From the perspective of human beings, the process just described is the simulation of biological life and its attribute of consciousness, a process animated by a computer simulation system that is causally equivalent to certain processes in biological life-forms (or as nearly equivalent as it is technically possible to make it).

Biological life-forms continuously face survival issues in every situation of their lives, issues that imply certain questions these life-forms must figuratively answer with quite literal actions: What to do next? How to choose specific actions from the seemingly unlimited number of potential actions open to them? A DLF simulates these issues and actions because the simulation program system interacts with reality based on the same teleologic as the biological life-forms do.

Perception automatically answers questions implicit in any situation a life-form may face, such as:

Evaluation automatically answers questions implicit in the identity information provided by perception with simulated feelings of pleasure/pain such as:

Action Selection automatically answers the question implicit in the simulated feelings of pleasure/pain provided by evaluation, which is: What should I do next? The answer is based on the DLFs values, its built-in survival strategies that have been reverse engineered from biological life-forms, and its current life status: Of all the potential actions available to a DLF, the actions that it can perform in answer to the implicit question above are limited to necessitated survival actions and optional actions. Why, because simulated death prevents any other kind of actions from getting repeated in the long-term by DLFs.

The Action method closes the loop with Reality by effecting the causes of the actions selected in the C.Event, including the simulated life processes of the DLF for which the C.Event is an instance of simulated consciousness.

When functioning, a teleological system that simulates life and consciousness involves some number of C.Events per unit time (say one per second, more or less), and these transform the identity of objects in a real or simulated world into information in a DLF's simulated consciousness. The result is a growing number of memories for a DLF, memories that constitute its perceptual information about whatever kind of world it senses.

Note - The number of C.Events per unit time could vary substantially, depending on how much processing occurs in a given event. A long memory search, for example, to recognize an object could cause a single C.Event to last a few minutes or more in a DLF with a large memory store of objects.

Note - Research at Princeton University (1999) on macaque monkeys has shown that thousands of new neurons are formed in the monkeys' brains each day, neurons that travel to the cerebral cortex where higher intellectual functions and personality are stored in humans. These neurons are then specialized in various ways.₃₃ This process is continuous and could explain some aspects of consciousness as a cycle and the physical basis of learning new subconscious processes in monkeys and potentially in humans. It could also explain how conscious events are stored in memory.

Since some of the information in a DLF's memory (perceptual information) corresponds to the identity of the objects in the DLF's world (simulated or real), the DLF can use the information to act in its world to attain its goals of survival and simulated happiness.

Simulated consciousness and the information it provides to the Evaluation and Action Selection methods is causal, and therefore it has survival value to the DLF just as real consciousness has survival value to real life-forms. There is an unbroken chain of causality from the world through perception, through the DLF and its actions, back to the world, an unbroken chain of the identity information that is conserved as the content of simulated consciousness.

Note - Consciousness of relationships and other complex phenomenon will be described later.

In biological life-forms, consciousness is an active, teleological, life process that transforms the identities of objects in reality into the form of perceptual information, a form of identity suitable for storage and processing inside the life-form.₃₄

Actions performed by life-forms in reality transform the identities of objects in the form of information in a life-form's memory back into the form of objects in reality, or at least changes to such objects. Consciousness is a causal process; it has causal efficacy; this follows directly from the fact that consciousness is a limited process with a specific identity and hence a specific action capacity.₃₅ As simulated in this invention, consciousness has an analogous function because it provides the DLFs in the computer simulation system with a form of causal efficacy that is different from state of the art computer programs.

Note - The author's view contrasts sharply with the view that consciousness is either mystical or a totally transparent, empty process that lacks any identity. In the science of the current state of the art, only the latter view is taken seriously (since computers are not supernatural). That view translates into the false idea consciousness equates with brain function, and that therefore (as Herb Simon and others have suggested), that computer hardware is the "brain" with software being the "mind." This false idea is then further extended to mean that an ordinary computer program can somehow become conscious without considering the complex causality and teleology of life processes or considering the identities of the objects which are the content of consciousness.

The attribute of simulated consciousness provides a DLF with a relationship to reality, a level of awareness, that is analogous to that of conscious biological life-forms, as opposed to entities which do not have that level awareness, such as viruses or state of the art computer systems.

Described in this way, it can be seen that simulating consciousness is a straight forward process that a skilled programmer can reproduce with the appropriate computer hardware and program code to interact with reality, provided the rules of the complex causality of goal-directed behavior are followed as part of the system's design.

The key to a successful design is causality substitution, to separate the various types of causal processes into subsystem layers as shown in Table 5-1 (reproduced below).

Then it is to substitute the appropriate computer hardware and software for the mechanistic causality that underlies real life-forms, in order to animate the entire system.


	Biological life-forms	Digital life-forms
Layer 7	Conceptual Consciousness (Reason)	Simulated Conceptual Consciousness
Layer 6	Perceptual Consciousness	Simulated Perceptual Consciousness
Layer 5	Goal-directed Cellular Processes	Simulated Goal-directed Behavior
Layer 4	Mechanistic Cellular Processes	Digital life-form Simulation Program
Layer 3	RNA, Protein, ATP Synthesis	Object-Oriented Prog. Environment
Layer 2	DNA Processes	Computer Operating System
Layer 1	Electro-chemical, Physical Processes	Computer Hardware

With this type of design, the software and interactions with the world that simulates the complex causal functions of teleology and consciousness in layers 5-7 in Table 5-1 are supported by mechanistic causality of non-biological kinds in layers 1-4 shown on the right side of the table, but supported in a manner analogous (meaning causally similar or equivalent, depending on the process and level of technology) to their biological counter-parts (taken collectively) shown in layers 1-4 on the left side of the table.

The net effect of the design is that in both columns, the processes of simulated life and consciousness are causally connected to the world in all respects by a causal chain. There is simply no magic involved, only levels of causal complexity supported by physical reality in each case:

• With biological life, the behaviors of conscious animals and other higher forms of life are caused by the teleological behaviors of cells, the teleological processes of cells are internally driven by Dr. Binswanger's three criteria of the cells' actions being self-generated, the cells' actions have value-significance to the cells, and the cells' actions are caused and internally controlled by there value-significance to the cells. All of these teleological causal forms are themselves caused by the simpler mechanistic causal forms of the electro-chemical and mechanical processes by which the cells physically operate, such as those listed in the left column in layers 1-4.

• Likewise, with DLFs, their simulated consciousness and other behaviors are caused by the teleological simulation methods running in the simulation system, and these are driven by Dr. Binswanger's three criteria of the DLFs' actions being self-generated, the DLFs' actions have value-significance to the DLFs, and the DLFs' actions are caused and internally controlled by their value-significance to the DLFs. All of these teleological causal forms are themselves caused by the simpler mechanistic causal forms of the logical, electronic, and mechanical processes by which the computer simulation system physically operates such as those listed in the right column in layers 1-4.

The bottom line is that the causality in both columns of Table 5-1 parallel each other.

5.5.3 Automatic Survival is at the Foundation of Life

In the simulation of consciousness, the automatic (in the teleological sense) and efficient survival value provided by perceptual consciousness is the foundation for the development of the simulation of more complex conscious functions such as forming abstract concepts, generalized ideas, more complex information, and the use of natural language as the tool to accomplish actions.

As a basis for the description of the simulation of these processes, there are some key points that need to be emphasized.

Interacting with Memory

Much of the processing for the simulation of more complex conscious functions involves much more interaction with memory than occurs at the perceptual level of consciousness, and subsequent processing of those memory contents.

For example, if suitable recognition and memory association methods are written for a DLF's action repertoire (to enable it to compare its memories of different C.Events and thereby recognize objects and feelings it has had in the past, and to associate the past successful actions with specific objects and feelings), then it will be possible for a DLF to remember a past C.Event such when it felt "fullness" from eating. This fact will lead to purposeful behavior in a developing DLF because such associations will result in simulated feelings of desire to repeat successful past actions, actions that brought the DLF simulated pleasure in the past.

Recognition and Purposeful Action

As memories accumulate, recognition of objects, scenes, relationships, past actions, and situations become more complex and important to action selection. The purposes of actions also become more complex.

For example, if in a given C.Event a DLF feels hunger, it may automatically recall associated events or actions in the past and then the action strategy method processing of that memory can calculate a simulated desire for feeling "fullness." One of the action strategies built into the system and explained earlier is: Select the action that resulted in pleasure in the past when a given object was perceived.

To be conscious of the "desire to eat" as a human is, a "desire" simulation method would need to be written that calculates a generalized simulated feeling of "desire." The new feeling is a general causal factor for any kind of purposeful action, usually to bring other simulated feelings back into an acceptable range. In the case of simulated hunger, for example, the desire will cause the action selection method to select actions such as "find food" and "eat." The result is the simulation of purposeful behavior: The DLF will try to find food in order to eat, and eat in order to feel full, thus satisfying the simulated desire.

To accomplish this result, the desire simulation method must be designed to give priority to or cause the selection of any action that had previously achieved a given goal, and resulted in, say, "fullness" in a case of hunger in the past. These past instances are found by ordinary memory searches.

Let's assume the "select an action that resulted in pleasure in the past" action selection strategy has been called in a given C.Event. A search of memory would be initiated looking for actions that lead to "fullness" in the past. When one is found, a simulated feeling of "desire" for that action can be calculated to increase the likelihood the action would be selected over other actions that may be in the queue for selection. The Figure 5-12 below shows an example, starting with the calculation of a feeling of hunger.

In the flowchart, the specific actions are not shown, but they would be the actions of a DLF such as Look, Find Food, and so on. This simulated desire feeling must be made an attribute of the action it is associated with. That way it is always part of the action, and available for the action selection method to compare to the desire attributes of other actions when processing its queue.

The same teleologic that works for simulating purposeful behavior for a DLF to eat, can be applied to enable a DLF to simulate any other kind of purposeful behavior, provided it has the necessary objects in its environment and a programmer has written the necessary methods for the DLF to interact with the world and its memories, so simulated desire attributes can be calculated.

For example, many higher animals have instincts, which are automatic forms of complex purposeful behavior such as nest building or making certain call sounds. Once the causal sequences involved are diagramed with flow charts or similar analysis tools, the process steps included in these behaviors are certainly no more complex than the behaviors of some non-living objects that have been simulated on computer systems, such as building and flight testing jet airplanes; the latter are simply of a different causal and logical form.

By studying the various instinctual behaviors of animals, programmers can write methods to enable DLFs to simulate similar behaviors in their environment and therefore save the time and resources that would be required to re-evolve them, as some are attempting to do with state of the art systems such as genetic algorithms. The reverse engineering approach will work as long as the programmers take care to simulate the more complex teleological causality that biological life-forms exhibit by keeping in mind the action limiting effects of conditionality and death, and then making sure all program methods are consistent with such effects.

Note - For the sake of clarity, I must point out once again that human programmers writing action methods for DLFs is simply a means to provide DLFs with a basic action repertoire. DLFs must then select these simple actions in various sequences in the context of goal-directed survival behaviors to be purposeful. The simulation of goal-directedness does not require the recapitulation of evolution. One of the key advantages of the DLF simulation system is the fact that one can start in the "middle of the evolutionary scale" so to speak.

When the simulation of purposeful behavior is combined with simulated perception as described previously, the result is useful simulations of some animal behaviors.

Automatic and Infallible

At the perceptual level, simulated consciousness operates automatically like its biological counter-part. Its control is exclusively by means of automatic, goal-directed action selection strategies. It is entirely predictable and infallible within the range of its action potential as an automatic survival system and will always select actions to promote life. Remember, survival actions are all necessitated; neutral actions may be optional, but by definition, are not against survival. Actions that are anti-life will be stopped by simulated pain, and if that fails, by "death."

Assuming normal computer operation, the fact that reality and simulated consciousness both have specific identities that interact only in a specific, causal relationship, then the arbitrary is precluded from occurring. Only survival behaviors and neutral behaviors can be caused by a system with this design in the long-term. The reason: DLFs that cause any other kind of behaviors are wiped out.

The two forms of teleological action, goal-directed behavior and purposeful behavior, enable a DLF to simulate predictable, automatic behaviors similar to those observed in biological life-forms.

Note - Even the random actions provided by a Random_Act method as described in the section on action selection above, are only unpredictable in a narrow range that depends on the design of a DLF program and a given computer random number generator. At a deeper level, these actions are not random at all; all the actions are caused, though not necessarily predictable down to the last detail, due to the limited nature of human consciousness and the complexity of modern computer systems.

In biological life-forms, automatic self-regulation of behavior is caused naturally by genetic evolution and refined by ontogenetic learning. Many different action strategies are tried by many different life-forms; the ones that work to help the life-forms survive persist, and the ones that do not are wiped out of existence along with the life-forms they killed; these behaviors, therefore, do not get repeated over the long-term, not in the sense of genetic algorithms as in the current state of the art, but in the sense of Dr. Binswanger's three criteria that were explained earlier.

With teleological causality as a model, programmers can apply analysis tools to specific biological behaviors and reverse engineer them for the action selection methods of DLFs, thereby avoiding the need to create them by recapitulating genetic evolution as genetic algorithms attempt to do. A programmer can write program methods directly that simulate the goal-directed, automatic survival strategies of biological life-forms based on simulated consciousness at the perceptual level. The result will still be different from that in extant systems because the DLF simulation system is teleological in its basic design.

In order for DLFs to have manual (non-automatic) behavior and the large degree of control over their own self-regulation the higher level of consciousness observed in human beings provides, additional processing must be done on the content provided by perceptual consciousness. Furthermore, the additional processing can be neither pre-defined, nor automatic. It must be volitional; that is, the highest level of consciousness of a DLF must be self-regulated, actions must be defined and selected by choice.

The volitional mode of simulated consciousness can only occur in subsystem layer 7 of the model shown in Table 5-1, at the level of simulated conceptual consciousness, and will be described in a later section. The conceptual level enables DLFs to modify their own behavior in a manner similar to the way human beings do.

5.6 The Emergence of Volition in a DLF

The main purpose of this invention is to simulate the most advanced form of conscious behavior known, volitional or rational self-consciousness as it is observed in human beings, and to animate it with a computer-based simulation system; all the description to this point is designed to support the simulation of consciousness at this high level. The value in doing so is that this form of consciousness is not only self-regulating, but also self-defining, and therefore a simulation system with this capability could be used to replace human beings in some situations where non-automatic, self-regulated decision making is required, such as in space probes and other robotic systems. Volition is a valuable capability that is not possible with extant robotic systems that operate by simple billiard ball causality because they have no means to initiate action outside the scope of the automation pre-defined by their programs.

As subprocesses, the simulation of goal-directed behavior and automatic perceptual consciousness that have been described so far in this chapter are useful inventions in themselves. They provide the content for the next level of this system, the simulation of volitional consciousness, but they are not nearly as powerful as the volitional form of consciousness. Being automatic (in the teleological sense), their functionality is largely defined in advance as built-in functionality, not self-programming functionality, and it is therefore more limited.

Note - What I mean here is that the behavior of the simulation system is automatic in the sense that the operation of the perceptual system in animals or instinctual animal behaviors such as nest building is "built-in" and functions automatically in the biological sense. I do not mean automatic in the sense that computer programs or state of the art agents, automations, or robots are automatic.

It may be difficult at first glance to see how any form of consciousness could be volitionally self-defining, let alone a simulation of it. Yet both the automatic and volitional forms depend on the designs that have evolved naturally to help biological life survive, designs that do in fact already exist. You and I are living examples.

Remember, all extant forms of consciousness are causal attributes of various life-forms and only exist because of the survival advantages they provide. If you look at the scale of complexity of biological life-forms, it is obvious that a key survival advantage that consciousness offers them is not only immediate awareness of reality, but also the speed with which the behavior of the life-form can change, and hence the speed with which a life-form is able to deal with environmental changes.₃₆

Life is action and depends on having both the right behaviors to survive in the present, and on being able to change behavior as the environment changes in the future, so as not to break its own causal chain.

Genetics and goal-directed behavior automatically limit and modify the behavior of life-forms by means of evolution and death; evolution provides the mechanism for changing behavior and death insures behaviors that are counter productive to survival do not get repeated over the long-term.

Perceptual consciousness provides a means of automatically converting the identity of objects and some simple relationships from physical form in reality into the form of information in the memory of certain life-forms. This conversion offers the additional survival advantage of enabling ontogenetic changes in the behavior of these life-forms through learning, as opposed to the slowness of genetic changes to behaviors; this advantage is analogous to the ease and speed of modifying software vs. modifying hardware in computer systems.₃₇

Both of these capabilities enable life-forms to adapt to their environments, the latter offering faster adaptation than the former with the additional survival advantage that speed of adaptation provides. Both of these capabilities are also automatic, non-introspectable, non-evaluative, and non-modifiable by the life-forms that possess them.₃₈

Note - As explained earlier, the automatic behaviors of biological life-forms and those simulated by the DLFs in this invention should not be confused with cybernetics and control theory. The maintenance of life is a positive, value-seeking process, not the negative-feedback, stasis-seeking process of control theory.₃₉

Over many conscious events, the capacity of perception, evaluation by the pleasure/pain system, and taking action, conscious events which are repeated over and over, enable biological life-forms to make limited changes in their day-to-day life process, mainly by a learning process that maximizes pleasure and minimizes pain. Similarly, a DLF can do the same with its simulated life processes over many C.Events following the processes described earlier; by making trial and error changes to which action methods it selects or the numerical settings in them, a DLF can make small modifications to the way it looks for food, the objects it draws, the way it interacts with a human teacher, and so on.

Perception and memories of past C.Events enable a DLF to "see what it is doing." What a DLF with only perception cannot do is introspect, evaluate, and modify the individual processes within a C.Event, to rewrite its Look or Draw action methods for example, just as an animal cannot modify its eyesight or hearing ability.

Recall the important distinction that must be made between mechanistic and teleological necessity: While all teleological actions are caused, the only actions that are necessitated teleologically, are those that cause survival, such as eating, sleeping, or finding shelter.

All other actions, though also caused, are teleologically optional actions, such as a chimp playing with sticks or smelling flowers. A life-form can perform optional actions or avoid the effort depending on the state of its life and its pleasure/pain system. Once a life-form gets past living "hand to mouth" and builds up an energy reserve, if it is healthy and its other survival needs are met, it can spend some of its energy on, optional, non-survival activities. both physical and mental. Its motivation to do so is simply the pleasure the activities generate for it. Optional actions are ends in themselves, or at least the pleasure of doing them is. Optional actions can be pro-survival or neutral (non-essential for survival), but they cannot be anti-survival (at least not for long!) because pain or death will soon result.

Optional actions are causally necessitated in the sense that a given cause always leads to its effect, but not teleologically necessitated: Optional actions are optional precisely because they are not essential to the specific causal chain that a life-form's survival depends on; they can be enacted or not, depending on pleasure of the life-form, depending on its internal state, not some external factor; optional actions are not required for a life-form's survival.

For DLFs specifically, the only action necessitated by teleology is the "Eat" method; if a DLF does not eat, like a biological life-form, it dies when it runs out of EPs. However, if a DLF has sufficient EPs and it performs a physical action such as drawing a shape or mental action such as comparing several objects it has perceived, these are optional actions; they have no effect on, are neutral to, the DLF's survival (except that they consume EPs).

A DLF automatically selects optional actions based on the state of its simulated pleasure/pain system. As long as it has enough EPs, a DLF can select optional actions that are available and that its memory shows will increase its pleasure, it can select them by trial and error, or not select them. It can select either optional physical actions or optional mental actions.

It has been common knowledge in biology and anthropology for many years that there is only a small difference between the higher apes and human beings. Some apes share many behaviors with people and their brain weights differ by only a small amount. Yet human beings have a much faster means of modifying their behavior than apes or any other kind of conscious life-form: The behavior of apes has not changed much for millions of years; in the past few thousands of years human beings, who were once living not much different from apes, have built an industrial civilization and explored part of the solar system. What is the difference in the human form of consciousness that has made such a fast, self-modification of behavior possible?

What is different is that the highest level of consciousness in human beings is not automatic and therefore not pre-defined; human beings and only human beings can introspect, evaluate, and modify their mental behavior; humans can modify the behavior of their consciousness, and do so quickly, whereas the apes cannot.₄₀

Volition is an action capacity of human beings that makes such fast behavior modification possible; humans have a powerful ability to initiate optional mental actions. Volition consists of the ability to focus the mind, to self-regulate some of the human conscious mental processes, and most importantly, the ability to form concepts.

In human beings, to form concepts or not, is optional mental action based on the choice to focus one's mind.₄₁

Identity determines action capacity. The ability of human beings to form concepts is the ability to change their own identity, to redefine themselves and therefore cause changes to their own action capacity.

As I have already explained, percepts are formed automatically by neuro-physiological processes in biological life-forms, including human beings, and simulated percepts are calculated in DLFs to mimic that biological process.

In human beings, the only biological life-form with the capacity to form them, concepts are formed volitionally by an act of free will; that is, they are formed by an optional mental action to focus the mind, an action of consciousness that results in the modification of memory content in the mind of a human being, content that has been automatically put into memory by the perceptual system during earlier conscious events. The result of the modifications human beings make to their memory, by choice, is a new data type: the concept (as defined by Ayn Rand).

Note - The free will vs. determinism argument is beyond the scope of this patent description; suffice it to say that determinism is self-refuting. (And nothing can be indeterminate because all actions by all objects are caused.) In the context of the description of this invention, simulated "free will" or "volition" is a causal process and a means to modify a process of simulated consciousness. See the following reference for more detailed information on this topic.₄₂

In the context of simulating consciousness, the capacity to modify a conscious process is similar to the capacity to modify a physical process. In a DLF, the latter is done automatically (in the teleological sense) at the perceptual level of simulated consciousness by selecting and optional physical action method. This means that as long as a DLF has a moderate amount of EPs, it can select any of its actions that cause changes in its environment. For example, a DLF could select the "Say" method to type words instead of the "Draw" method to draw shapes in order to achieve some non-survival goal; alternatively, a DLF could set the numerical values in the "Draw" method for drawing a circle instead of a triangle.

These changes are caused by automatic, teleological action selection strategies in the DLF's simulated pleasure/pain system as already described, but they are optional behavior because they are not necessitated by the DLF's survival. They simulate a very simple, limited form of "volition" or "free will" in a DLF, but their effect does not go beyond the specific action taken, so their consequences are almost trivial.

The capacity for a DLF to modify a process of consciousness or symbols in its memory is similar in that it too is an optional action, except the change that is effected occurs in the organization of the DLF's memory instead of external reality, so it is an optional mental action instead of a physical one. But optional mental actions have more far reaching effects because they change the DLF's identity, and hence its action capacity, and that fact makes their consequences are far from trivial.

In other words, the effect of an optional mental action will be to change the DLF's action capacity for future C.Events. By doing so, the DLF has just redefined itself by exercising that optional mental behavior. It has simulated a subconscious choice.

Large changes in a DLF's action capacity do not happen all at once, but add up over a period of time. The ability of a DLF to redefine itself consciously bootstraps itself by a process I will describe later.

The point to grasp here is that this is the entrance to the conceptual level of simulated consciousness (subsystem layer seven in Table 5-1) and that it is caused by the action of consciousness on itself; it is a recursive process of selecting optional mental actions that allow a DLF to subconsciously redefine itself (for the simulated pleasure of doing so) and thereby simulate a limited range of volitional consciousness (though it cannot yet simulate volitional self-consciousness). In order to see how and why this process works, lets look more closely at concepts because they play an important role.

For a human being, and according to the Objectivist theory of concepts, a concept is: "A mental integration of two or more units possessing the same distinguishing characteristic(s), with their particular measurements omitted."₄₃

The details of how the concept formation process can be simulated by a DLF will be described in the next section. The point to grasp here is that the concept formation methods in a DLF are action methods that operate internally on a DLF's simulated consciousness instead of externally on the world it perceives; they further process the measurements that have been calculated for simulated percepts and stored in a DLF's memory; they change the way information is stored and accessed in the DLF's memory, instead of changing external reality as most of its other action methods do.

The methods that simulate concept formation are optional mental actions for DLFs.

The changes they make modify the DLFs identity, its action capacity, and hence the DLF is capable of new behavior in future C.Events. This is how a DLF can be self-defining like a human being is.

Concept formation occurs over several C.Events, and unlike the perceptual process, the process of concept formation is introspectable, "looked at" by the DLF as part of its simulated conscious processes, can be evaluated, can be modified, and the whole process can be remembered. This means concept formation is a purposeful mental action that is optionally in the direct control of the agent performing it. A DLF can introspect, evaluate, and change its concept forming actions and their results, it cannot do the same for its automatic behaviors such as perception; it can only look at different objects.

The specifics of how changes to simulated conceptual processes occur will be described in the next section, for now I want to point out some important consequences for this optional mental action capacity of DLFs:

In the next section, I will describe the specifics of how simulated concepts are calculated in a DLF simulation system.

5.7 A Simulation System Design to Calculate Concepts

As with perceptual consciousness, conceptual consciousness offers humans a survival advantage. Not only do concepts reduce the number of units that must be processed, they make it possible to be conscious of aspects of reality that are inherently invisible, such as relationships and mental processes, and thereby provide the ability for faster changes in behavior than genetic evolution or the ontogenetic changes that perceptual consciousness makes possible.

As explained at the beginning of the chapter, by concept formation and concepts I mean only Ayn Rand's theory of concept formation and the type of concepts that method produces.

Given the Objectivist theory as a basis, the process of forming concepts can be simulated by enabling the content of the simulated perceptual consciousness of a DLF be introspected, evaluated, and modified by the DLF initiating its own optional mental actions.

Concept formation is not necessitated behavior in human beings and it will not be in DLFs; it is an optional form of simulated mental behavior that recursively changes the identity of the agent performing it, and its power to do so increases as the number of concepts formed increases.

This latter fact is what produces the dramatic difference in observed and potential behavior between other primates and human beings, even though their brains are nearly the same size.

5.7.1 The Nature of Concepts as a Data Type

Concepts in the Objectivist theory of concept formation are analogous to file folders in offices or the electronic database files of modern computer systems; they are a means of storing information in a different, abstract form by taking advantage of certain measurable relationships in the data to separate some data records and to store others together.

Filing systems offer the advantage of compressing large amounts of information into the small space of a filing cabinet or computer storage media in an organized way so it can be easily retrieved when needed.₄₄

Filing systems, whether manual or electronic, work by differentiating some objects such as papers or data records, and grouping or integrating others according to attributes of their content they share in common. For example, sales invoices go in one folder or computer database file, vendor invoices into another, marketing information into still another, and so on. The choices of what attributes to use to set up a filing system are based partly on the usefulness of the attributes of the objects being filed and partly on the human needs of what the filing system is to be used for, but they are all measurable attributes of the objects being filed. In other words, some of the criteria come from the identity of the objects being filed and some come from human needs, purpose, and values.

Not only are there an endless number of ways objects can be filed, once a filing system is in place, there are certain bonuses that accrue due to the identity of filing systems in general. For example, files about invoices of various types can be themselves grouped together as financial information, as opposed to other kinds of files such as those containing marketing information.

Information about business paperwork and process can be distinguished and separated from business objects, such as the physical plant that produces the product or the office furniture, all defined in the form of a genus which indicates the next more abstract file and differentia which uniquely identifies a given file. (The system works like a taxonomy in biology.) If necessary, all files can be further abstracted into more and more general files, until a single file is reached that summarizes the business in abstract form: the file containing the balance sheets for the business, which is the ultimate genera summarizing the contents of the entire filing system.

In addition, within any filing system there are an even larger number of cross-classifications that can be made, such as between invoices generated by a particular marketing program or for the purchase of computer equipment as opposed to those for consulting services.

These attributes of filing systems enable a manager to see information about a business that ranges from the "big picture" to the most specific detail, such as how much a pencil costs.

It is because of all the potential choices in designing a filing system (and the workflow issues that go with it) that office automation systems must be worked out manually by business people and systems analysts first, analysts who detail all the choices implicit in business operations, before they can be programmed to run on computer systems.

Filing systems were originally devised to make information manageable for people. They have survival value, especially for businesses, because business managers are better able to make the decisions necessary to compete efficiently and maintain profitability.

Life-forms face an analogous challenge; they are inundated with information and the faster they can process it or retrieve it when it is needed, the better chance they have of surviving. Moreover, environments are not static; if changes occur that threaten a life-form's survival, it must change its behavior to adapt to the new environment, change the environment, or die.

I have already discussed the survival advantage of being perceptually conscious of the world as a collection of objects, of using percepts as processing units as opposed to using sensations. Concepts offer another even larger advantage to one type of conscious life-forms: human beings.

The primary survival advantages of concepts are: Processing unit economy and rapid self-programming.

The processing units of concepts as symbolized by natural language words are small by comparison to the percepts of most other objects, like the labels on ordinary file folders are small by comparison to their contents. Concepts offer a means of filing and organizing percepts that is similar to what an office filing system does for business papers. In using this same analogy to a filing system to explain Ayn Rand's theory of concepts, Dr. Binswanger has called the concepts the file folders of the mind and the folder labels the words of natural language.₄₅

In addition, concepts provide a means of making the invisible visible: Relationships are not objects, but shared attributes or other kinds of identity links between objects; they are the informational analogue to the physical connections and interactions of objects in the real world. But the vast number of relationships that are not physical are inherently invisible to perceptual consciousness. Conceptual consciousness, on the other hand, can represent any kind of information, including these "invisible" abstract relationships.

These facts plus the ability to recursively modify conscious behavior make a whole new level of action possible to a simulated life-form. Concepts enable a DLF to redefine itself and its action capacity.

Concepts, as they are defined and used in this description, are a new kind of data structure to the fields of AL and AI. In the next several sections, I will describe how concepts can be formed in a computer simulation system and the new capabilities they will provide.

5.7.2 Concept Formation as a Calculation Process

Concept formation is neither arbitrary and subjective nor intuited from some "intrinsic feature" of objects. In human beings, it is a process based on the perceptual comparison, and then differentiation and integration of percepts based on the attributes of objects perceived and the needs of human cognition; it is the fact that some attributes are similar, that is, they differ only in quantity or measurement that makes the process possible. In addition, concept formation is based on the fact that some attributes are distinguishing or unique to various types of objects, a fact that enables the objects possessing the unique attributes to be grouped together.₄₆

The best way to understand how to form a concept is by example. Earlier, I showed the simulated percepts of a triangle and a circle. To continue with that example, I will go through the process of forming or calculating a simulated concept of a triangle in this section using the data created in the DLF program.₄₇

In figure 5-13 below, there are three triangles, a circle, and a square. One of the triangles and the circle are like the ones shown in the example percept used in an earlier section. I have eliminated the fill on the circle for the sake of simplicity. The other two triangles are new and of a different shape than the first one. The X,Y coordinate pairs are not shown in this example, but they are the processing units for the objects at this part of the process, the data units that will be transformed into the attribute lists that simulate percepts as described previously.

Next, in figure 5-14, the windows that show the attributes that were calculated for objects 1 and 2 by the percept simulation methods in the DLF Program I have been writing to prove out these ideas. These attributes, that is, lists of perceived properties and measurement values are the data that will be used to form or calculate the simulated concept "triangle." The simulated percepts are the processing units for the process that calculates the measurement range definitions of simulated concepts.

The resulting simulated concept will be calculated from the content of actual simulated percepts, not intuition or from some arbitrary construct.

The process of simulating the formation of a concept involves comparing the attributes of the objects, that is, comparing their property and value pairs. Since these are all physical objects of the same basic type (simple line shapes), they all have the same (trianglelist) attribute lists.

Note - Other objects could be used for comparison as long as they share at least some attributes. If not, they would be incommensurable and could not be compared.

Also not included here are the attribute lists for the lines, end point connections, and angles in triangles 4 and 5 that shown in Figure 5-13, again due to the fact that the composite object method in the DLF program was not complete at the time this book was written. However, it is obvious that the calculations of the attribute lists for triangles 3 and 4 will produce similar lists for each of the three lines in the triangle; the lines would also have the attribute "connected at end points" (because each line has some end points in common with the others), and the attribute "angles" (because the end point connections of the lines would necessarily be at some angle to each other).

Even without this information in the figure, however, it is obvious that in the results of a comparison, each triangle would calculate as a shape composed of three lines, and their composite closure attributes would calculate as TRUE, due to the end point connections. The square would calculate as four connected lines with 90 degree angles, and the circle as one line with no angles.

The differences in the lists are the values of the attributes, primarily in the line numbers, positions, lengths, slopes, curvature, end point connections, angles, and so on.

Looking at the attribute lists and comparing them, it can be seen that the difference between these objects is one of measurement: The values of the commensurable attributes are all different. On the other hand, there are more similarities in the measurements of the triangles, as opposed to the circle and the square, the attributes of which are much more different.

The triangles all consist of three lines with end point connections and three angles less than 90 degrees, whereas the square has four lines with four 90 degree angles, and the circle has only one line with no angles; the triangles and square are straight lines, the circle is a curved line. The triangles all share a range of measurements that is different form the circle and the square, and would be different from a trapezoid or octagon or other shapes if they were drawn and processed into simulated percepts like the ones shown. These differences are used to differentiate the triangles from the other shapes and integrate them as members of a group of similar members, namely those shapes that fall into the measurement range described.

When the comparison is complete, the attributes unique to triangles as opposed to the other objects in the scene are that the triangles are all three straight lines connected at their end points. This means that the specific values of the attributes do not have to be specified because all triangles will always calculate an identity (property and value list) that includes them in the range of values unique to that group. All the other attributes and their values exist as part of the identity of each triangle, but are not relevant to defining the concept because they are not unique to it, though all the other attributes continue to exist and are included in the identity information the simulated concept stores.

Since any triangle will always calculate attributes within the "triangle" measurement range, they will be always distinguished from all other objects (non-triangles) in the scene by their unique measurement range (a subset of their property/value list).

Note - As used in this description, the term "attribute" means a type of measurement, a property/value pair (such as EPs=25 or Length=30) and is synonymous with the term "characteristic."

By using the word "triangle" (provided by a human teacher) to symbolize the measurement range of these shapes, a DLF performing this process has formed a simple first (perceptual) level simulated concept; the unique measurement range defining a word integrates all the triangles like an ordinary file folder integrates a pile of invoices. This is the process of simulating concept formation, of calculating a concept from a group of simulated percepts of objects that a DLF has previously perceived and stored in its memory.

The word "triangle" stands for, symbolizes, or designates the concept; the word and its concept mean any object (in this case triangle) that falls into the specified measurement range is part of the concept (whether the DLF has actually perceived a particular triangle or not) and will be recognized as such in future C.Events.

In her theory of concept formation, Ayn Rand calls the observed similarity between objects in a group of two or more similar members a Conceptual Common Denominator (CCD). The objects' shared attributes serve to both differentiate them from other objects and integrate them as the units of a new group.

The subset of CCD attribute(s) and/or measurements that distinguish the objects as units of a new concept are their Distinguishing Characteristic(s) (DC). That they must have some measurements is based on her "some but any" principle: Because to exist is to be something, the relevant measurements "must exist in some quantity, but may exist in any quantity."₄₈

As explained earlier, the method Ayn Rand identified in the early 1960's for forming concepts is new, unique, and objective, as opposed to the intrinsic or subjective approaches to forming concepts found in most extant systems of thought. In addition, because her method uses perceptual measurement as its basis, the method lends itself to calculation and use in a computer-based simulation program.

Calculating the concept "triangle" is a simple example, however, it is representative of forming a concept of objects. Furthermore, the same process can be extended to and repeated over and over for any commensurable objects, such as circles, squares, octagons, tables, chairs, the letters of the alphabet, and so on, anything a DLF can "perceive." The identities of the objects "sensed" by a DLF will always produce simulated percepts which are the processing units that can then be processed into simulated concepts as long as they have commensurable attributes and there is some survival advantage or simulated pleasure for the DLF in doing it. Other kinds of objects would require different attribute lists, but the process is always the same to calculate their properties and values from their X,Y coordinate lists (or other simulated sensory measurements), and then to form simulated concepts of them.

All the other objects not included in a given concept are the context for the concept's uniqueness, and a concept must be re-formed and updated if new objects come into the context which would produce a different comparison result. Simulated concepts are therefore contextual, and their calculated definitions may need to be changed to account for new information, but they are absolute within the particular context they are calculated.

Assuming a DLF lives in a reasonably rich world of objects (simulated or real), that it processes all the simulated percepts in its world, and asks a human teacher for the names of the objects it perceives, using the processes described herein, such a DLF would have the capacity to recognize these objects by name (using a natural language word) as well as by example. The word that is the simulated concept name serves as a symbol that stands for an unlimited number of objects of a certain type, that is, that fit the measurement range, the DC, for the objects subsumed or integrated by a simulated concept. The DC calculated for the concept is the definition that indexes the word to all the simulated percepts of objects it subsumes.

For example, in Figure 5-15, the world the DLF perceives contains a triangle and the word "oval." Once perceived (and assuming the DLF has sufficient EPs to engage in optional actions), the recognition methods compare the new percepts to those stored in memory from previous C.Events.

The triangle percept (trianglelist) is compared to the measurement range definition or Distinguishing Characteristic (DC) for the concept "triangle." Since the perceived shape is, in fact, a triangle, it will match the measurement range of the DC for the concept "triangle" and be recognized. The word "triangle," (which is indexed to the concept and thereby all perceptual instances of triangles) is accessed by association if an object fits the DC. Other program methods later in the C.Event enable the DLF to select the action method "Say" to output the word "triangle" to a human teacher to show it has correctly recognized the shape.

The other object in the example, the word "oval," is perceived and recognized in the DLF's list of concept names, and is indexed to its respective DC, the measurement range for ovals; for this example we will say the DC for ovals is the attribute "Curvature=TRUE." Other methods later in the C.Event enable the DLF to select an example oval from perceptual memory and the action method "Draw" to output the example oval to a human teacher to show it has correctly recognized the concept from perceiving the word "oval."

The specific size of the oval the DLF draws could be any size within the measurement range for ovals and the capabilities of is output device. The specific size would be defined in an optional mental action by the DLF and actually drawing the example oval would be executed as an optional physical action.

Both of these actions are examples of simulated volition, and the latter is an example of a simulated first cause.

The conceptual method of storing information is efficient. In the case of the triangle example, the word "triangle" stands for and is a single processing unit that integrates every triangle the DLF that formed the simulated concept has perceived in the past, may be perceiving in the present, or will ever perceive in the future. All of the DLF's knowledge about triangles is indexed by that single symbol: the word "triangle." That is what storage efficiency or processing unit economy means: Potentially billions of bits of information are reduced to about 64 bits for the word, whatever is needed to store the measurement range definition for the concept (less than 500 bits for many objects), and some example triangles. Thereafter, in future C.Events, the DLF can specify all the information about triangles in its memory with a single processing unit, the word "triangle."

Computer systems and robots in the current state of the art store the data of objects they sense as pixels of X,Y coordinates and color information. Using simulated percepts and concepts will offer an efficiency of many orders of magnitude over the capabilities extant systems by means of the processing unit economy just described.

As you may have noticed, much of the processing involved in forming simulated concepts is internal, in memory, with the simulated percepts of objects serving as the data. And additional processing can result in additional useful concepts, once the concepts of objects are in place.

Just as with an ordinary filing system, files can be grouped in more general categories such as, say, financial records and marketing records; in a similar manner, more abstract and general concepts can be formed using first level concepts of objects as data. For example, by comparing concepts of triangles, circles, squares, octagons, and so on, the attributes that distinguished these objects from others to form first level concepts (the DC) become the data that is compared to form the more abstract concept (the CCD). Within this data, a new DC is identified to form the more abstract simulated concept, and a new word is acquired from a human to symbolize it.

Consider the DCs for triangle (three straight lines connected at their end points), for squares (four straight lines connected at their end points at right angles), and for circles (one curve line with its end points connected); these become the CCD for the new concept "closed shapes," as opposed to "open shapes." When these measurement ranges are compared, a new DC can be calculated, lines connected at their end points, that defines the concept "closed shapes," which is a second level and more general simulated concept, an abstraction from an abstraction.

After simulated concepts of many types of objects have been formed in a similar fashion and concepts of perhaps two or three more levels of abstraction formed, the simulated concept object itself can be formed, which is an "ultimate genera" or the top of the hierarchy for a large group of concepts.₄₉

Note - Large conceptual hierarchies are too complex to draw. The example shown is intended only to give the general idea of how a few levels concepts are related.

The point to grasp here is that all of the simulated concepts shown are calculated by the same process as described in the triangle example of comparing the attributes of commensurable objects against those of all other known objects; they are all calculated the same way by widening the measurement range for the attributes in the objects' identities. The measurement range is based on data sensed by the DLF in every case; none of it is arbitrary programming constructs as are simulated concepts are in state of the art systems that may use them.

The operating principle is that the DLF's memory is organized and indexed according to certain relationships that are carefully calculated to be consistent with the actual relationships of the objects sensed in reality. In this way, the DLF builds up an organized, objective, simulated conceptual knowledge of the world it exists in, but as with simulated percepts, identity information is carefully conserved.

Note - In an object-oriented programming environment, new object instances inherit their attributes from a pre-defined hierarchy of object classes. Simulated concept formation is the opposite of inheritance: The properties of objects (instances) that exist in reality are perceived, the percepts compared, CCDs identified, and DCs selected in order to form the classes of the hierarchy.

Organized, simulated conceptual knowledge facilitates object recognition in future C.Events for DLFs. Prior to having simulated concepts, a DLF could only recognize specific objects by matching its current simulated percept with one in its memory; finding a match would mean that a particular object had been perceived before, but it would have to be an exact match. Simulated concepts enable a DLF to recognize objects by types, such as triangles, circles, squares, and so on.

More often than not, improved recognition ability will result in greater simulated pleasure for a DLF and therefore encourage more simulated concept formation as an optional mental action.

Another interesting comparison to ordinary filing systems is that they also have the feature of providing sub-classifications and cross-classifications, such as all invoices of customer Smith, or all items purchased in the past month by customers Smith and Jones. Simulated concepts have this same feature, so more abstract and specific concepts can be formed, as opposed to more abstract and general concepts as described above.

For example, if in the simulated concept triangle the attribute identifying angles were as part of the attribute list, then the DC for the concept triangle, three straight lines connected at their end points, can be narrowed by adding an additional "angles" attribute to the DC. This would enable the concepts of equilateral and scalene triangles to be formed by effectively narrowing the measurement range of the original simulated concept.

A similar process could be used for cross-classifications; for example, if the value of the number of lines calculated for an object's identity was omitted and the curvature attribute which is always FALSE for certain objects was included, then the triangles and squares could be combined and integrated in a cross-classification simulated concept called "straight shapes" as oppose to curved shapes (for which the curvature attribute is always TRUE).

Any sub-classification and cross-classification simulated concepts are all calculated the same way by narrowing the measurement range for the attributes in the objects' identities, as opposed to calculating more general concepts in which the measurement range is widened.

All of these second level and higher simulated concepts, whether more general or specific, are more abstract concepts as opposed to first level simulated concepts of perceived objects.

Ayn Rand's method of concept formation is the one this simulation system implements, and it is the only conceptual system to provide both a widening and narrowing feature as part of its concept forming method. As Rand points out: "Starting from the base of conceptual development--from the concepts that identify perceptual concretes--the process of cognition moves in two interacting directions: toward more extensive and more intensive knowledge, toward wider integrations and more precise integrations." (Italics mine).₅₀

Simulated concepts of relationships can be formed by DLFs by comparing object attributes. For example, concepts of spacial relationships can be formed by comparing the position attributes of objects relative to each other. Of the objects shown in Figure 5-13 and the attribute measurements of some of them in Figure 5-14, it can be seen that the circle (object 2) has an X position coordinate of 53, while the lines in the triangle 1 have X position coordinates of 16, 15, and 30. This means, of course, that the circle is to the right of the triangle in the simulated world of the DLF. If the data were available, the other objects would show X position coordinate values greater than that of the circle.

By including the specific values of the position attributes into ranges of values relative to each other, the simulated concepts of right, left, next to, over, under, and so on can be formed; the concept "next to" might be defined as "the unoccupied range of X+50," where "unoccupied" means no other object is in the positions of X to 50, and the concept is literally defined by a search of the image to check for that condition. Since position values are always calculated as part of every object's identity, they will always fall into predictable ranges, and it is easy to calculate which spacial relationship concept subsumes the various objects in any scene.

The simulated concept "in," for example, can be formed by a DLF by perceiving instances of one object inside another, such as an oval in a rectangle, a circle in a square, and so on, as opposed to empty shapes or those in other spacial relationships such as "next to," "over," "over-lapping," and so on. The CCD for the concept "in" is the range of positional measurements shared by any group of shapes in close proximity to each other (including their relative sizes), and the DC is that all the position measurements for a smaller shape are encompassed by or are contained by the position measurements of a larger shape. This calculated relationship, which can be applied to any objects that fit the DC, is the meaning of the concept that is symbolized by the natural language word "in" for the DLF that formed the simulated concept. That DLF now has an objective definition for the meaning of "in."

As should be obvious at this point, simulated concepts of relationships are calculated as ranges of measurements in the attributes of one object, relative to those of another. A similar approach can even be used with a DLF's value relationships so it can form concepts with objective, calculated definitions of its simulated feelings: As explained earlier, a DLF's simulated feelings of "pleasure" and "pain" are ranges of measurements that are calculated relative to the DLF's values; these ranges are the CCD for simulated concepts of simulated "feelings" for a DLF. The DC for the concept "pleasure," for example, is any simulated feeling in the range between zero and 8 (but not as high as 9), as described previously for simulating "fullness." Thus, using simulated concepts, a DLF can learn to name its own simulated feelings.

With the help of a human teacher to provide the words and demonstrate the context, two other important simulated concepts can be calculated: A DLF can form the concepts of "place" and "world."

The definition of the simulated concept "place" is calculated by focusing on the fact that objects have a position attribute relative to other objects, a location, and that this is true of all objects in a DLF's memory; in other words, the CCD for the concept is having a position attribute, and the DC is that position measurements are unique for every object relative to others. Based on this, objects are integrated by the fact that they all have some specific place, but that place could be any position coordinates in the entire range of available positions in the DLF's universe. The concept "place" is therefore a relational concept like "in," except that place is a relationship that applies to virtually all objects. To paraphrase Henny Youngman: "Everything has got to be someplace."

The simulated concept "world" is another relational concept that is sort of the "flip side" of the concept "place:" It's measurement range is all the places of all objects that exist, the "place" measurement range in totality; a "world" is neither an object nor a place, but a relationship, the group or collection of all places within a certain boundary. This fact has an interesting consequence: The collection of places a DLF is able to perceive has a finite size because all DLFs have a limited, finite action capacity, and that limit forms the boundary of its world. All objects are related because they all have a place, and the concept "world" not only integrates them all by defining the collection of all places, but it provides the DLF with a boundary to its "universe," which is marked by the places of the two most distant objects it "knows" of. (The boundary of the DLF's "universe" is therefore a relational, not a physical boundary.)

To humans, the concepts of "place" and "world" are quite abstract. For a DLF they are as well, but like all the other simulated concept examples I have been describing, these two are also calculated based on data in the DLF's memory and that it is perceiving in its world. They are not arbitrary constructs, but are connected to the DLF's world by an unbroken, causal chain of calculations.

At this point, it should be clear to an experienced programmer that while a consciousness simulation system is complex, it is a system that is in no way magical or arbitrary; it is a system can be created in a straight forward manner with the appropriate programming methods. The key is that the program methods must be able to transform, by calculation, the identity of a world of objects into an equivalent world in the form of perceptual and conceptual calculations, calculations that are defined by and connected to natural language words. As with simulated percepts, identity must be conserved. This principle is absolute, and must apply to every part of the DLF simulation system.

To do so, the system must calculate lists of X,Y coordinate pairs for each object and then calculate a unique identity in the form of attribute lists for each object which is stored in its memory; these attribute lists are the objects in the form of information. The lists in memory are further processed by simulated concept formation methods which calculate concepts as just described, and after interaction with a human teacher, the result is a hierarchy of concepts symbolized by natural language words. (The required choices for this simulated volitional process consisting of optional mental actions will be explained in a later section.)

The simulated concepts (words and measurement range definitions) are information that represents and integrates the objects and some of their simpler relationships; they are the world of objects represented in symbolic form (with identity conserved), as opposed to perceptual form or physical form.

It is important to differentiate simulated concepts as defined in this system, as opposed to state of the art databases, which are modeled after the idea that human concepts mean only their definitions. In state of the art computer databases that simulate natural language, the words mean only whatever database entry has been indexed to them as defined by some programmer or based on real-time input on how people may be using them for some particular purpose. They are not calculated from objects perceived in reality (simulated or real) by a DLF for its own purposes as they are in this simulation system.

Simulated concepts formed by DLFs in the system I am describing mean all the entities (objects, relationships, and so on) that they subsume, including all the attributes of these entities and their specific measurement values. There is a mathematical connection that can be traced, reproduced, and adjusted for context changes between the objects in a DLF's world and the natural language words that symbolize its concepts of that world.

• Simulated percepts of some objects (or attribute subsets) when compared calculate as similar, as opposed to other objects (or attribute subsets), and can be indexed by a natural language word as members of a group of two or more similar members (including all their attributes); that the group so defined is in fact an open-ended category containing all instances of a given type, past, present, or future: A simulated concept.

• Simulated concepts themselves can be compared and conceptualized in two directions: More abstract and general concepts can be calculated until ultimate genera are reached (such as the concept "object" or "action"); more abstract and specific concepts can be calculated as sub-classifications and cross-classifications of earlier formed concepts contained in a DLF's memory (such as "octagon" or "jogging").

Figure 5-17 shows a block diagram of one design for a conceptual consciousness simulator. The arrows show the data flow. The object comparison and other methods for simulated concept formation are part of the Action Methods process box and not shown separately. Concepts are formed over several C.Events (a complete cycle through all the processes and reality), and they result in a causal chain of relationships that serves to index the simulated percepts in a DLF's memory.

There is no breaks in this calculation chain, up or down this simulated conceptual hierarchy: There are no arbitrary constructs which are unconnected to a DLF's world; this is a closed system. An object sensed always calculates X,Y coordinates, which always calculate an identity of some kind, which (when compared to other identities of objects already in memory) always calculate a concept which is symbolized by a natural language word that is provided by a human. A word always symbolizes a measurement range definition, which always calculates a complete instance for that type of less abstract concept or object, which always calculates instances of specific percepts of specific objects, the percepts always calculate a set of X, Y coordinates, and a draw method always calculates and then reproduces some specific object that was originally perceived (or a variation within the appropriate measurement range for that object). Arbitrary connections do not have survival value and are not repeated because DLFs that do repeat them "die."

In every case, the calculation chains linking words to reality are unbroken, reproducible, and objective. These calculation chains are the meaning of the natural language words as used by a DLF simulated life-form.

Note - Variations can be calculated based on the identities of perceived objects and relationships by recalculating their attributes to simulate "imagination" and "creativity" in a DLF, calculations which can be partially random, though not arbitrary. Actions of this type are limited by their survival value to the DLF. Errors are also possible, and will be dealt with in the section on simulated volition.

In practical terms, words that are connected to a DLF's world in the manner described offer a DLF a huge unit economy (and survival advantage) to future calculations about the conceptualized objects. Consider that after a simulation system has operated for some practical purposes for a long time, there would be thousands or millions of instances of objects encountered by a DLF, each of which would need to be recognized, evaluated, and perhaps acted upon. With simulated concepts connecting all these instances into a hierarchy, the DLF can use words in a manner similar to the way variables are used in mathematical equations. It can do so to calculate and simulate its survival strategies before acting on them.

The conceptual simulation system just described is easily adjustable as reality changes or new percepts are discovered by a DLF. For example, if a new shape is perceived such as an "oval", the concept "circle" as formed in the description above would no longer be objective because the new percept changes the contents of the memory used for the comparison to form the concept; its DC of "curvature = TRUE" would no longer distinguish the circle from all other objects because an oval would also have that attribute. The comparison conceptual calculation must therefore be re-done to re-form the simulated concept and update its definition with a new DC to account for the results of the new comparison in an expanded context. The new DC would be a narrowing of the concept "circle" to objects with "curvature = TRUE" and a constant radius. By observing additional examples of ovals and ellipses, concepts of these new objects could be formed as well.

Though this is an extremely simple example, it applies to virtually all simulated concepts because they are all calculated the same way, and it demonstrates how by using optional mental actions to form concepts, a DLF can better identify its world as that world changes, and thereby aid its effort to survive. Forming concepts for a DLF is therefore advantageous, though not necessitated behavior.

The simulated concept formation process described herein is also scalable. There is both an enormous number of concepts possible, and an even larger number of sub-classifications and cross-classifications possible. The only limitations are the memory size of a DLF and the survival value of the concepts produced by the calculations to form them; in other words, the value of the concepts to help a DLF maintain its own life and therefore have a chance to form more concepts in the future.

Note - It is important to point out that once a conceptual memory has been organized and validated as objective in one DLF, unlike with human beings, it can be simply copied to other DLFs, or communicated over the Internet. Assuming many DLFs operating worldwide, every time a DLF formed a new concept or updated an existing one, the information could be instantly copied to and shared with the other DLFs over the Internet, and they would then have the new knowledge as well.

The key thing that a programmer must remember is that this system is not just a computer program. The process of simulated concept formation is not automatic and pre-defined, only its subprocesses are. Conceptual processes must be designed such that they can be internally controlled and caused by a DLF as optional mental actions, not automatic, necessitated behavior.

Finally, the simulated concept formation process described in this section, as with the simulation of goal-directed behavior and perceptual processes described earlier, is straight forward to program. Any expert object-oriented programmer could build the simulation system using the description and explanation provided in this book. As with any new system, there would be some developmental problems, experimentation needed, and bugs in strategy and implementation, but these could be easily worked through and fixed by a competent programmer following the architecture I have provided.

5.7.3 Concepts, Memory, and Action Capacity

Now that I have described how to simulate the concept formation process, some of the basic attributes of concepts, and the survival advantages they provide DLFs, it is necessary to explain some special consequences of a DLF organizing its memory conceptually.

As with ordinary computer systems, the potential of a given design is not always readily apparent from its basic description. Few people, for example, were able to foresee the advent of the graphical user interface from the programming environments and screen technologies that made it possible, or the Internet as it has evolved from simpler network systems. A similar situation exists here: Simulated concepts are a new and unique kind of data structure in the field of computer science; they have some additional features and uses beyond those already described, features that lead to some interesting capabilities for DLFs.

Simulated concepts are timeless, as opposed to percepts, which are time dependent.₅₁ The latter is true because reality is constantly changing; a simulated percept of an object at time T1, would not calculate the same percept as on at time T2 in the real world.

Concepts on the other hand, change only when the new information of an expanded or narrowed context requires that they be updated, such as the update I just described of the simulated concept "circle" to account for ovals and ellipses. This means that the concept "triangle" that people use is essentially the same one as Pythagorus used thousands of years ago. The changes to update a concept are calculated in a specific manner, so that concepts change only in non-arbitrary ways: The calculations to change them are specific and the old concept is still in a DLF's memory for comparison and can even be reused if the updated concept is found later to be in error.

The timeless nature of concepts also enables a DLF to identify the temporal nature of individual objects. Being aware of a timeless category makes it possible to notice that individual objects come and go. That is, sometimes objects can be perceived in reality, and sometimes they cannot be perceived, but only exist as percepts in the DLF's memory. This fact enables a DLF to make explicit the concept of "existence." An object "exists" if at time T1 it is in reality and can be perceived by a DLF; the same object does not exist at time T2 if the DLF is looking at its world, but the object is gone, is only in the DLF's memory, and cannot be perceived in the DLF's world. Figure 5-18 below illustrates this point.

At time T1 the rectangle object exists; at time T2 it does not. Many observations of this phenomenon and the nature of simulated concepts enables a DLF to calculate two new concepts: "existence" and "negation" (meaning non-existence or "nothingness"). The CCD for the concept "existence" is any attribute, and the DC is to have some attributes. The concept of "negation" is a relational concept like zero, it is a placeholder for the lack of attributes, for the lack of an object in a relationship with other still existing objects.

The concept "nothing" or "non-existence" (literally "no thing") is the flip side of the concept "existence," except it means only a void, not a thing: To exist, an object must be part of the DLF's world, have a place, a relationship to other foreground and/or background objects. The negation of existence in the perceptual context is "nothingness," the lack of something or the absence of an object in the perceptual field as indicated by only background objects. As with the concept "existence," the CCD of the concept "nothing" is any attribute; the DC is that there are no attributes, no attributes in relation to other objects which do have attributes. This is an example of how a concept can make the "invisible," "visible" to a consciousness. The word "nothingness" is a percept that stands for a relationship which is itself inherently invisible. (The word is also connected to the DLF's world by calculations in the same way as all its other words are.)

Earlier, I discussed how a DLF simulating conceptual consciousness could form the concept of "world". Having explicit concepts of existence and its negation symbolized by natural language words enables a DLF to recognize when objects are subsumed by those concepts, just as it can recognize conceptually that an object is a rectangle or a circle using the measurement range definition and words for those concepts, or that the position of a rectangle is to the left of a circle by using the measurement range and words for spacial relationships. The process is the same for the relational concepts of "existence" and "nothingness," but the level of importance is not: Being able to recognize whether an object exists or not is much more fundamental, and it is necessary to simulate higher order conscious functions. The point to grasp here though, is that the nature of concepts is what makes this capability possible at all.

A DLF capable of simulating conceptual consciousness has, therefore, a new capability which a DLF capable only of simulating percepts does not have: The conceptual DLF is conscious of its world as a timeless entity as opposed to a succession of specific, time-dependent percepts or objects the world contains, as shown in Figure 5-19. Using its concept "world," a conceptual DLF can be aware of every object, past, present, and future, (whether the objects still exist or not) using that one, single word.

Forming simulated concepts adds organization to a DLF's memory by means of the chains of calculated, measurement based relationships between concepts, percepts, and objects in reality described in the previous section. Memory is a part of the identity of a DLF, and identity determines an object's action capacity.

Note - Recall the example from the beginning of this chapter of how the identities of a balloon and a bowling ball would affect the sidewalk if dropped from a tall building, the action capacities of each identity causing very different results.

The new properties that concepts add to the identity of a DLF's memory must also be reflected in its action capacity. However, before the specifics of that fact can be described, a bit more explanation of how conceptual consciousness emerges is required.

5.7.4 How Simulated Conceptual Consciousness Emerges

To build a simulation of conceptual consciousness, it is easier to start with a simulated world of simple objects such as the one I described using in the example of how to form the concept triangle. The point of doing so is to work out the methods in more detail as to how to form large numbers of concepts efficiently for various objects and their relationships, as well as abstract concepts of other concepts, including the concepts of "place," "world," "existence," "nothingness," concepts of various spacial and other relationships, and so on as described in the previous sections.

This being done, a more sophisticated simulated reality can be developed, or the appropriate sensors and processing methods put in place to process the part of the real world that you and I perceive. The DLF simulation system can then be "turned loose" to form concepts on its own.

At first, like a human child, a DLF forms simulated concepts more or less randomly, that is, as it encounters various objects in its world. It will do so because calculating concepts is a series of optional mental actions; it is a behavior that can only be performed when a DLF has sufficient EPs, and one that is not necessitated by a DLF's survival needs.

If the DLF's pleasure/pain methods are properly designed, however, the simulated "pleasure" they calculate will "encourage" a DLF to form more and more concepts as it can spare the EPs. This will eventually result in a large number of them being formed because concept formation will become a habitual behavior, one that results in simulated pleasure for the DLF more often than not. It will result in simulated pleasure and that will be remembered; automatic action selection in DLFs, as in biological life-forms, is biased to select pleasure producing actions. So even though concept formation is only one form of optional mental action, there is a high likelihood it will be selected frequently.

Though optional, forming concepts is not an arbitrary behavior either. Concepts make it easier for DLF's to recognize objects, identify the existence of the world and themselves with a single symbol, make implicit information explicit, offer processing unit economy, and so on. Concepts therefore offer a DLF real survival value.

The additional information about the world a DLF inhabits that concepts provide, will lead to more and more concepts being formed because the information will make survival easier and more pleasurable for the DLF. Finally, at some point it will become much more efficient for the DLF to be guided by a human teacher, so it does not, in effect, have to rediscover all of human knowledge in order to learn advanced concepts.

While the process of learning about its world conceptually will take some time, it is obvious that by repeatedly performing the concept formation and updating processes I have described, and doing so over many, many C.Events, a DLF can eventually be able to identify every object in its world and many of their relationships with natural language words, words which mean the calculations that connect them via the indexed organization of the DLF's memory to the real objects in the world.

Note - Forming all these concepts is a big project for both a DLF and a human teacher; there is no doubt of that, but it is not an impossible project; there are a finite number of concepts that must be formed. The number is probably over 20,000 concepts, those that human beings use routinely, and could take several years to complete. Of course a protoype DLF could be developed using only a subset of this number in 1-3 years. Some concepts, such as "freedom," may require a robot "body" for DLFs to experience first hand the perceptual differences between being restricted and not restricted physically in order to form them.

Having reached the conceptual stage of redefining itself, a rudimentary form of simulated conceptual consciousness has emerged in that DLF.

This condition leads to a slightly more advanced stage of conceptual development for a DLF. For example, up to this point, I have been describing the use of only one concept per C.Event. However, one of the things a human teacher can show a DLF is that it can use more than one word (and hence the concept the word symbolizes) at one time in a C.Event, as this is a natural progression. Given that fact, and assuming the concepts of some objects such as triangles, squares, circles, ovals, spacial relationships, and so on have been formed along with the concepts of "existence" and "non-existence" by the process I have described, a human teacher can then do the following:

Draw or point to a triangle and type the words: "triangle exists," then erase the triangle and type the word "nothing", and repeat this process for other objects. As the DLF processes these C.Events, its recognition methods trace the calculation connections between the objects perceived and the conceptual chains in its memory that connect the objects in its world to the words.

The result is that the DLF confirms by calculation that the percepts of the objects are subsumed by the concepts for the object shown, "triangle" and that of "existence," but not for the relational concept "nothing." In other words, the DLF recognizes that the "triangle exists."

At that point a slightly more advanced form of simulated conceptual consciousness has emerged in that DLF.

5.8 The Emergence of Simulated Self-Consciousness

The simulation of self-consciousness requires both goal-directed behavior as well as the simulation of perceptual and conceptual consciousness before it can emerge as a form of behavior by a DLF interacting with its world; the simulation of self-consciousness also needs a human teacher to be part of that world to facilitate its emergence, to guide the DLF to perform the optional mental actions required to form the simulated concept of "self."

As with a DLF's simulated concept of the "world," it is the timeless nature of the concept of "self" that makes self-awareness possible, by integrating all the time-dependent C.Events of itself that are in a DLF's own memory into a single processing unit: The word "me."

The simulated concept "self" is formed by a DLF perceiving itself along with other objects in its world, including other life-forms. This can be accomplished by giving the DLF a simulated body or a robot body to perceive with and the ability to introspect, that is, to monitor its own internal processes using optional mental actions, so it can monitor itself as it perceives the world and build up a large number of instances of its own C.Events. The simulated concept "self" integrates these instances, just as the concept "world" integrates all the instances of the DLF perceiving objects in reality into a single processing unit (the word "world").

The simulated concept formation process is the same as it is for "triangle" or any other concept. The CCD for the concept "self" is all the attributes of the DLF that are similar to those of other objects, and must be because the DLF is an object in the world too. The DC for the concept is the fact that instances of "self" have as attributes part or all of the DLF's own identity, as opposed to percepts of other objects which are not part of the DLF, and these attributes therefore uniquely distinguish the DLF from all other objects with the same CCD.

The concept of "self" timelessly differentiates the DLF from the rest of its world and integrates all the instances of its identity, whereas all the particular C.Events the DLF has of itself are time-dependent.The concept "self" groups and integrates all the C.Events of a DLF into a single category represented by the single symbol, the word "me." The word can henceforth be used as a single processing unit when the DLF wants to process information about itself, including awareness of itself. The DLF can henceforth symbolize its entire stream of simulated consciousness with that one single word.

Only simulated concepts formed using the Objectivist method and symbolized by human supplied words can provide such a perspective; they can do so because of their open-ended, timeless nature: The concept "me" contains all instances of a self: past, present, and future, thereby making a virtual, temporal object visible to itself; the simulated concept provides the static continuity that makes the "self" continuous, real and distinguishes it as an object, from the rest of the DLF's world.

The simulated concept "self" is what causes self-awareness for a DLF by using the power of a concept to transform all the individual instances of "self" into a single processing unit, to integrate them into a single, virtual object that is symbolized by a single simulated percept.

Having formed the simulated concept "me," the DLF can view itself as an object instance in its concept "world" and locate its place in its world and its position in its conceptual hierarchy as shown in Figure 5-20. Remember, unlike with object-oriented programming inheritance, a hierarchy of concepts is calculated from reality upward, from the concrete to the more abstract.

With the simulated concepts of "self," "existence," "world," and "in" having been calculated and indexing content in its memory, as well as the ability to recognize the units of multiple concepts in a single C.Event, a DLF has reached the point where it can identify a complex fact conceptually, the fact that: "me exist in world." This natural language phrase is neither arbitrary nor trivial.

The DLF can make this identification because each of those words connect via their chains of calculations to the respective objects and relationships that they mean in reality.

The result is calculated with data from reality. It is not a "canned," arbitrary computer programming construct.

This "thought" by a DLF is only a phrase, not quite natural language, but more like a proto-language; it is a "conceptual identification" that has been calculated based on the way the contents of its memory are indexed, a calculation that conserves the identity of the part of reality the "thought" means.

The fact the words identify is not the result of stringing words together, but rather the recognition that the specific content of the DLF's simulated consciousness for that C.Event is subsumed by these four concepts, taken together. In the current C.Event, the DLF's conceptual recognition methods calculate a result, and that result is that the instance of the DLF is simultaneously subsumed by the concepts "me," "existence," "in," and "world."

The words are strung together quite simply as a by-product of the fact that they cannot be spoken or printed all at once, but only in a series.

Note - The word order that is unique to a given natural language must be learned by a DLF as the result of interaction with a human teacher. Whether a DLF can learn the subject-verb-object order of English, for example, by mimicry, whether it will have to be conceptualized, or specified by some other means will probably have to be determined by experiment.

Conceptual recognition methods identify the DLF as a unit of the concept "me" (self), a unit of the concept "world," a unit of the relational concept "existence," and a unit of the space relationship concept "in." In other words, this multiple conceptual recognition is possible because the DLF is separately subsumed by each of the concepts listed (just as a triangle is subsumed as a unit of the concept "triangle"); the DLF calculates as one of the units of each of these concepts. So self recognition is just another form of conceptual recognition.

While the ability to do multiple concept recognitions in a single C.Event and output a string of words that symbolize the recognized concepts is not simulated natural language understanding, a simulation system with this capacity is one in which has the prerequisites for "learning" to simulate natural language in its future C.Events.

5.8.1 The "What if" Capacity of Conceptual Information

Before I can describe how simulated natural language understanding emerges from the DLF simulation system, there are a few more ideas that need to be explained. These ideas derive from the nature of conceptual information and its relationship to reality.

First, there is an important distinction between reality (object) manipulation vs. symbol manipulation.

Note - This statement does not imply that symbols are outside of reality (which is meaningless), but that symbols are a special case because they represent aspects of reality to a consciousness, aspects of its identity in the form of symbolic information, as opposed to percepts.

Reality manipulation by a life-form or a machine is both physical and metaphysical; that is, reality manipulation is limited by the laws of the physical sciences and, in a more abstract sense, the basic nature of what things are, the identities of the objects involved.

Symbol manipulation, on the other hand, is limited primarily by metaphysical laws, mainly by the identity of consciousness, not by the practical laws of the sciences. Symbol manipulation is primarily optional mental behavior performed by human beings that can be completely arbitrary (as in fantasizing), or it can be limited by the laws of logic (as in simulating), depending on the choices of people doing it.

Except for fantasies and games, arbitrary symbol manipulation is a useless exercise. However, the ability to arbitrarily manipulate symbols makes possible the creation and use of logical symbol systems for concept formation and the mental simulation of reality within a conscious mind. This latter fact is the basis for current state of the art simulation systems, systems which are originated in human consciousness. Then logic and experimentation are used to keep simulations of airplanes or weather systems consistent with reality, while the arbitrary nature of symbols make possible limited deviations from reality to test new ideas, to provide the "what if" capability that makes simulations so useful.

State of the art simulation programs use X,Y coordinate systems, mathematics, logic, and so on to simulate many aspects of reality. However, extant simulators are bound to concretes, to the specific instances of the specific objects they simulate.

The use of conceptual symbols (words defined by measurement ranges) in a simulation system adds some new dimensions to symbol manipulation in a simulator: The ability of concepts to integrate unlimited numbers of objects over wide ranges of attribute measurements, and their ability to be timeless offers a whole new approach to simulation systems that is analogous to the way human beings simulate events in their consciousness.

Words and their definitions are the essential parts of identity of concepts; the rest of the identity of concepts is all the other information about the units the concepts subsume, including all their non-defining attributes; this other information is used to create the conceptual calculation chain, the meaning that ties the words to the objects in the world of a DLF.

Note - There is a confusing, false idea in epistemology originated by Immanuel Kant called the Analytic-Synthetic Dichotomy. This idea, which has oozed into our culture, claims concepts are merely words plus their definitions, nothing more. It does so by arbitrarily disconnecting the defining attributes of a concept from all the non-defining attributes, specifically ignoring the fact that the items dropped are required to form concepts in the first place.₅₂ The DLF simulation system works precisely because the there is no such arbitrary disconnect.

Recall that in the explanation of the concept "triangle," I explained how the connections of an unbroken chain of calculations that are calculated from the measurement ranges of actual objects in a DLF's world make possible the simulated conceptual recognition of similar objects by a DLF using natural language words: Once the concept is formed and the DLF "sees" any triangular object, it recognizes the object with the word "triangle."

Furthermore, I explained the reverse calculation of an example conceptual chain that starts with a word and produces a conceptual unit (a specific object): Once the concept is formed, if given the word "triangle," a DLF can trace the appropriate conceptual chain and draw an example triangle because the chain enables the DLF to "know" what a triangle is, to know what the word means.

Finally, I demonstrated how the meaning of the word "triangle" is the chain of calculations that connect the word to actual objects in reality (its units), and how it therefore means all the attributes of all triangles for all time within some specific context: The simulated concept as a data type is the calculation chain, the non-defining attributes of the objects it subsumes, and the word that symbolizes it.

This design has powerful consequences for the DLF simulation system, because as I have also described, conceptual identifications by a DLF such as "me exist in world" are made possible by the same type of unbroken chains of calculations, calculations that enable a DLF to "know" the meaning of these concepts in exactly the same way it "knows" the meaning of the concept triangle.

Remember once more the idea that the identity of an object sets its action capacity (the balloon/bowling ball off a tall building example). Now consider this idea in the context of the DLF simulation system that uses simulated concepts, while remembering the all inclusive, timeless nature of concepts.

Simulated concepts provide a DLF with the capacity to run symbolic simulations of its world over a number of C.Events, to simulate an "imagination." Given a DLF that has formed a large number of simulated concepts as I have been describing, and given the way calculations connect those concepts to the DLF's world, that DLF can use words in C.Events to be "aware" of its world; it can recognize objects in existing relationships in its world (because it has concepts of them), as well as physically draw objects in relationships for its teacher as examples of conceptual information in its memory (examples like the one I showed with the concept "triangle").

This is possible because the word connects to the measurement range definition which subsumes the particular percepts of the particular objects (the specific instances) used to calculate the concept in the first place, and the percepts contain the attribute values needed to set the values in the draw method to draw the X,Y coordinates of the object in the DLF's world.

As I have explained, there are no breaks in the calculation chain, up or down. And while it would take many pages to describe the calculation chains necessary to produce the DLF's conceptual identification "me exist in world," I could produce that description because the principle is exactly the same as for the conceptual identification of the concept "triangle;" it is just a more complex example of the exact same process.

The capacity to trace the calculation chains of simulated concepts includes the ability to modify the measurements of the objects and relationships of the units (instances) the concepts subsume. Remember, the way concepts are calculated is by identifying that a particular object such as a triangle is a specific set of measurements, measurements that fall into a range for a given object. This means that a DLF can not only recognize objects it has never perceived before (objects that are subsumed by a concept because they fall into its measurement range), but it can also create new objects of that type, such as triangles that it has never actually perceived. As a result, DLFs can simulate "imagination," a capacity made possible by the fact that objects can have any measurements within their specified range in a given simulated concept.

Note - The case of a DLF using a simulated concept to create or "imagine" a new unit or instance of an object it has never perceived is analogous to inheritance in an object-oriented programming environment. The new object inherits all the attributes of the units subsumed by the concept. The concept specifies a range of measurements, but not the specific attribute values (except for the particular units have been perceived before). The DLF can use a random number generation method to get and set specific values for the its draw method to actually draw an object that it has never perceived. The attribute values need only be within the range specified by the concept to be valid measurement numbers.

Simulated concepts thus enable a DLF to use optional mental actions to represent aspects of reality in the form of words, and then it can use those words both for conceptual identifications and to cause changes in reality, all by following the conceptual calculation chains in its memory.

This capability applies to relationships as well as individual objects. Given the ability to process more than one concept per C.Event and having learned word order from interaction with a human teacher, a DLF can string words together in short phrases such as "circle in triangle." The phrase means the measurement ranges calculated for the three simulated concepts as explained earlier in this chapter in the section about forming concepts of relationships.

The DLF can draw an example of this particular relationship even if entirely different objects were used to form the concept "in" because any object will work so long as it does in fact fit the concept's measurement range: In this case, the new objects being drawn inherit their attributes from the concepts of "circle" and "triangle," and the circle inherits the range of its position measurement (of being contained by the triangle) from the concept "in," though its specific value may be calculated at random.

By initiating an optional mental action, the DLF has thereby simulated "creating a visual scene by imagination" and caused something new to come into existence in its world. This is an example of a simulated first cause.

What was learned about two specific groups of objects, it turns out, applies across the board to all objects of a given type because of the nature of simulated concepts, and a DLF can verify this fact by a little "experimenting" with various objects in that measurement range and tracing the calculation chains thus produced.

The point here is that by simply forming a large number of simulated concepts of objects and relationships, a DLF gains a huge amount of information about its world, and this change in the DLF's identity results in a huge increase in the DLF's action capacity.

A DLF capable of only simulated perception has the action capacity to recognize specific objects and take various simple actions in its world, such as to find food or engage in the optional physical action of drawing specific objects.

A DLF capable of using simulated concepts has an immediate and immense expansion of its action capacity due to the tremendous leverage provided by the way concepts timelessly integrate the DLF's memory contents. Only a few percepts are needed to calculate each concept, but once formed, concepts provide DLFs access to every possible instance in the measurement range they cover, and that is one major source of their power (another being unit processing economy).

Depending on the motivation calculated by a DLFs goal-directed behavior methods for optional actions, I have described how a DLF at this stage of development would be capable of identifying, recognizing, and interacting with objects and relationships using its capacities of simulated perception, conceptual recognition, word phrases, and the ability to imagine new objects. All of this taken together means the DLF can cause significant changes in its world. These changes then become part of its knowledge in subsequent C.Events. The result is an ever increasing spiral of "knowledge" on the part of the DLF.

The bottom line is that simulated concepts formed by the Objectivist method enable DLFs to alter their world, including their own action capacities, and thus define their own future identities. This fact make DLFs self-programming.

5.9 The Emergence of Simulated Natural Language in a DLF

The manipulation of human language by computer systems in the form of text or speech recognition is not new; in fact it is common. There are many examples in the state of the art of people defining thousands of words for computers to interpret. But the words have no meaning to the computers; in extant systems words are only pointers to human entered definitions in a database that in turn link the words to computer commands, data records, or pre-programmed responses. To human beings, however, the meaning of words is in their connection to reality through a chain of ideas in the form of concepts, and that is a link state of the art computers do not have because they are merely unconscious machines.

Concepts are the "data structure" that link words to reality in human consciousness. Human concepts cannot function as part of an ordinary computer program because they are not formed automatically, but simulated concepts can be used by a system that simulates the volitional consciousness of a human being.

If DLFs are ever to simulate using natural language as humans do, they cannot function like computer databases designed and programmed by people, but must calculate their own simulated conceptual chains so they can trace the connection of the words to reality and "know" the meaning of their own words.

5.9.1 The Role of Concepts in Simulating Language

The concepts and words that form the basis of human languages and the ability of people to use them seems to be a natural capacity of the human mind, but the specific grammatical systems that make up a major part of the languages are a human invention.₅₃

DLFs either have to simulate a human language or invent their own, and the former option is obviously the more attainable of the two.

As I have described in the previous section, simulated concepts are the data type that make simulating language possible for DLFs. Chains of calculations give each concept meaning in the memory of a DLF, and multiple conceptual identifications lead naturally to word strings as a by-product, with word order learned from a human teacher.

Given this level of simulated conceptual awareness, having a teacher teach a DLF natural language grammar is the logical next step in developing a DLF's simulated consciousness.

Given a DLF that already has the ability to string words together as described previously, teaching natural language becomes showing the DLF the perceptual identity of a new type of object: the sentence. A human teacher can type sentences into the DLF Program interface as examples of language, along with the perceptual scenes they correspond to for the DLF to "observe" with its simulated perceptual consciousness.

Using this perceptual subsystem that I described in detail earlier, the DLF can perceive these new sentence objects like any other objects it perceives, and it can calculate attribute lists for them consisting of the type of and order of the words, the placement of capital letters and punctuation marks, and so on. These attribute lists constitute the identity of a sentence in the form of a simulated percept in the a manner similar to that for triangles or other objects.

Then, the DLF can "conceptualize" the sentence objects by comparing them to other sentences and character strings in its memory. The Conceptual Common Denominator (CCD) for a sentence, as opposed to other character strings, is that sentences all have words separated by spaces, the words have definitions and are all concept symbols or proper names, and sentences all start with a capital letter and end with a period, colon, semi-colon, exclamation mark, or question mark. The Distinguishing Characteristic (DC) for sentences, is that at least two or three of the words they contain must be valid concepts, the function of which is that they specify a subject, an action, and optionally, an object of the action; furthermore, the order of those words must be in the form of subject-verb-object, if the language is English.

As with the concept "triangle," once a DLF has formed the concept of "sentence," it can recognize any sentence it perceives as a type of object in its world. In addition, however, since each word in the sentence stands for a concept, the DLF can "parse" the sentence conceptually; it can follow its own chains of conceptual calculations to the measurement ranges that define the concepts and "know" their meaning. The meaning of the sentence is the meaning of all of its concepts, taken together as a whole, including all the relationships between the words such as modifiers and word order.

The other grammatical relationships of sentences can be calculated into simulated concepts in the same manner as relationships such as spacial relationships are, because all simulated concepts are formed the same way. For example, the concept "the" is formed by the DLF comparing instances of the use of that word, as opposed to similar words such as "a." The CCD is the position of the word as coming before a word that names something (noun or proper name); the DC is that it specifies a specific object or thing, as opposed to any object or thing within the measurement range of a concept.

This is very different approach from state of the art natural language systems in which grammar is interpreted by a dictionary and a database of rules that have been entered into a computer by a human programmer. Since the DFL has formed its own simulated concepts, it "knows" their meaning in relation to its world, not just by a database of canned responses provided by humans.

5.9.2 Decoding Simple Sentences

For a DLF, simulated grammar is conceptual, and "learning" grammar is like learning anything else about its world; it involves forming simulated concepts. The DLF's concept of grammar is a set of abstract concepts about sentence objects that a DLF perceives, just as its concept "shape" is an abstract concept about other types of objects it perceives in its world.

For example, given a perceptual scene and the sentence: "The circle is in the triangle." that I used as an example earlier, a DLF would decode the sentence as follows:

• Then the DLF would do the same for the concepts of grammar to decode the sentence itself: It would check the word order, make sure the word "the" calculates that the circle and the triangle are specific objects (the ones in the scene), not just any circle or triangle, that the concept "is" calculates that the objects exist in its world, and that the concept "in" calculates the actual position relationship between the two objects.

To the DLF, the sentence is the symbolic equivalent of the perceptual scene because it is connected by the calculation chains to the scene; like an equation, the two sides balance each other, the variables that are the concepts in the sentence are equivalent to the measurements of the specific objects in the scene in the DLF's world. In other words, the physical scene is a solution that fits the conceptual "variables" in the sentence.

The DLF can verify and therefore know the meaning of any sentence the same way it "knows" anything else, namely by tracing the calculation chains of the simulated concepts the sentence contains for both the objects the sentence refers to and for its grammatical construction.

5.9.3 Encoding Simple Sentences

Encoding simple sentences to simulate natural language is the opposite process of decoding them. The first step is to choose a subject: Consider a variation on the examples about shapes I have been describing, except this time the example will be of a square next to a triangle. The example might go as follows:

A human teacher draws a triangle, and next to it a square on the DLF Program screen interface, then types the question: "What is the drawing?". The DLF perceives these shapes and the sentence in its next C.Event.

Assuming the DLF is not starving so that it has the EPs to engage in optional actions, in subsequent C.Events the DLF can identify the shapes by tracing the conceptual calculations their measurement ranges fall into in order to identify the concepts that subsume them, and then it can get the words to name them from memory: "square," "triangle," "is," and "next to." In addition, the sentence "What is the drawing?" would be decoded according to the process described in the previous section.

All of these actions are optional mental actions initiated by the DLF for the simulated pleasure they cause for it.

By initiating another optional mental action, the DLF can then create an instance of the concept sentence (which is a framework of "sentence" attributes), into which words it has retrieved from memory can be placed to produce: "A square is next to a triangle." The wording is not arbitrary or canned, but is derived from the conceptual recognition of the scene as just explained above, and the grammatical construction from the inheritance of the sentence instance from its concept.

To answer the teacher's question, the DLF then needs only to set the words into its Say action method, which will then write those words to the program interface on the screen.

Repeating the processes just described of decoding and encoding sentences is how a DLF can simulate a conversation with a human being using simple natural language sentences. The DLF identifies the objects and their relationship both with simulated percepts and concepts, calculates the symbolic equivalent using its concepts, then places words into a sentence using inheritance from the concept "sentence" to set the grammatical structure, and finally, it outputs the result using the appropriate action method.

This process is an entirely new form of simulated natural language understanding and production in the fields of AI and AL.

5.9.4 Beyond Simple Sentences

Natural language as used by people is much more complex than the simple sentences I have been using for my examples, but that complexity is beyond the scope of this patent description. I believe the simulation of complex natural language is possible for DLFs, but not until second and third generation versions of DLF Simulation Technology.

The purpose of this description is to show how rational self-consciousness at the level of simple natural language can be simulated by a DLF and animated by a teleological computer simulation system, as well as to show the new capabilities it brings to the state of the art.

5.9.5 The Simulation of a Fully Volitional DLF

With the capability to simulate simple natural language sentences, the action capacity of a DLF reaches its most powerful level for this first version of DLF Simulation Technology, that of simulating a fully volitional agent capable of initiating first causes.

As pointed out at the beginning of this chapter, such a teleological agent is not conscious in the same sense as a biological life-form is conscious, but only as a simulation of biological consciousness. The DLF system is not alive, yet strictly speaking, neither is it a machine because only some of its behavior is pre-programmed; the balance of its behavior is the result of the causal efficacy of the DLF system's own simulation of teleological processes and human conscious processes, the simulation of the optional actions the identity of its design makes possible, and the interaction of all of this with reality.

Being capable of optional actions and capable of simulating conceptual identification provides some "freedom of choice" for a DLF, but to have the full power of volition requires the simulation of natural language.

The reason for this is that only natural language sentences enable a DLF to easily encode symbolic representations of its world (and itself) using its conceptual calculation chains, and this greatly amplifies the power of optional actions (just as it does for children). It does so because symbols (words and concepts) free the DLF from the specifics of percepts and thereby enable it to change its own identity and action alternatives. Using optional mental actions and natural language, a DLF can plan its own actions before it executes them.

A sentence is a complete thought, and a complete thought is a symbolic representation of an event in reality. Once a DLF can use optional mental actions to encode and decode complete sentences, it can simulate reality for itself by "thinking" about complete events and scenarios in its world. Then it can decide whether or not to cause the symbolic mental events in its "thoughts," as motivated by its own simulated values, in its future C.Events.

This level of simulated consciousness is a requirement of intentionally conceiving of changes in reality to cause. All cause and effect instances are the events in reality. Before changing reality, the DLF must first identify some aspect of it to change, and then "imagine" changes to that aspect according to its values in subsequent C.Events. Simulated natural language makes this process easy because both the identifications and changes can be identified and "mentally" proposed using symbols (words and concepts), instead of hard to manipulate real objects.

To initiate a "first cause," a DLF can initiate an optional mental action to encode a natural language sentence. For example, in one C.Event a DLF at this stage of development can encode the following sentence using the process described above: "The square is in the circle." Then once that sentence is in the DLF's memory, in subsequent C.Events the DLF can optionally "decide" to draw the objects described by the sentence, or not, to enact that alternative as an optional physical action or not, depending on the state of its simulated life.

To do so, the DLF traces the conceptual chains of both the objects subsumed by the simulated concepts and the measurement ranges specified by its grammatical concepts that describe the sentence. Tracing these conceptual calculation chains will point the DLF to the specific measurements in memory it needs to set in its draw method to draw the scene. Then it can select an optional physical action from the alternatives available to it, draw the objects, and the DLF has caused a first cause.

Once the objects have been drawn in the DLF's world and perceived by the DLF that drew them, a closed system causal chain is completed that was not necessitated by any condition either outside or inside of the DLF:

The DLF's "decision" to initiate the behavior is therefore the simulation of a first cause, the simulation of a free will "choice" on the part of the DLF.

Given that this process can be repeated for any optional actions, such as forming other simulated concepts or using any simple simulated natural language sentences to "think" about any subject, a DLF at this stage of development is a simulation of volitional self-consciousness that is capable of many of the same behaviors that a small child at a similar level of development is capable of performing.

While this simulation of natural language is only an imitation of human consciousness, thought, and volition, it will serve as a close enough approximation to be useful to humans as a powerful new kind of interface to a new kind of teleological system.

5.10 A Summary Description of the DLF Simulation System

At the beginning of this chapter, I described the problem facing current state of the art attempts to build AI and AL systems as the need to specify in advance not only what actions the computers running them will perform, but also where, when, and how these actions are to be performed using some set of rules or other means.

In other words, that the pre-definition of action in extant AI and AL systems is at the same time the reason the systems can run automatically, and their downfall. This problem is so formidable that it prevents extant systems from ever encountering other more advance problems such as how to achieve data processing unit economy or simulating consciousness.

Furthermore, I pointed out that the human beings that AI and AL computer systems are supposed to emulate are not automatic in either the mechanistic or teleological sense, but operate themselves manually; the behaviors of the human programmers that write the automatic programs that computer systems run do not have all their behaviors pre-defined. In fact, it is precisely the attributes of consciousness and volition based on teleological causation that enable human beings to be capable of optional behaviors and to invent and build computers and write computer code in the first place.

I concluded that the most immediate problem facing the current state of the art is captured by the question: How does one design a computer simulation system that is not automatic, not a mechanistic automaton?

My answer to this question is embodied in the invention I have described in this chapter.

5.10.1 Innovative Capabilities of the Invention

This invention is certainly not obvious. The capabilities of DLF Simulation Technology and the system design explained in the preceding patent description enable the creation of an intelligent life-form simulator by solving the problems that have prevented others from doing so to date:

• The invention solves problem of action pre-definition by simulating goal-directed behavior, behavior which is a form of complex causality that moves the energy source and the locus of control inside the acting agent, an agent who's existence is conditional; a teleological agent is permitted all actions as long as they support a specified standard or condition: the agent's own life. This causal form limits action over time by eliminating any agent that acts in contradiction to the standard in the long-term. In addition, the simulation of the complex form of causality that makes goal-directed behavior possible can be animated by off-the-shelf computer hardware and software (which operates by mechanistic causality) in a manner similar to the biological life is animated by the mechanistic causality of physics and chemistry at its lowest levels - provided the proper teleological software is supplied. This invention shows life processes and intelligence as a layered model of increasingly complex subsystems, and uses causality substitution to insert the complex causality of goal-directed behavior as the interface between simulated consciousness and mechanistic causality.

• Since intelligent action presupposes consciousness, in order to achieve simulated intelligent action the invention simulates consciousness at both the automatic, perceptual level and the volitional, conceptual level, the latter making the simulation of rational self-consciousness possible. The invention enables the simulation of consciousness as an attribute of the simulated life-form, an attribute used by the simulated life-form to identify objects and relationships in its world (whether simulated or real). Simulated consciousness in this invention is a series of C.Events, each of which is an element in a causal chain that begins with perception (identification) of objects by a DLF and ends with some action that effects objects in the DLF's world for the purpose of aiding the DLF's survival or for its optional actions (actions not necessitated for survival).

• Since volition implies the ability of self-regulation, the invention shows how goal-directed behavior, by moving the energy source and locus of control inside a teleological agent (the DLF) which faces the alternative of simulated life or death, makes optional actions possible, and it shows how optional actions in conjunction with concepts makes simulated volitional behavior possible, including the capacity to initiate first causes in reality.

• The need to process fewer data units equals a survival advantage for both biological and digital life-forms; the invention shows how data types called percepts and concepts reduce the system's processing units by many orders of magnitude (by gaining unit economy via content-oriented data compression), thus not only making "survival" easier for DLFs, but also greatly reducing the processing load on the computer system that animates them.

• The all-inclusive and timelessness attributes of simulated concepts and the conceptual data type make simulated consciousness of the world as a single unit and simulated self-consciousness possible; the invention shows how optional mental actions in conjunction with simulated concepts leads to the emergence of simulated self-consciousness and volition for DLFs.

• The invention shows how all simulated concepts are formed the same way using the Objectivist method, no matter what the subject content is, and they are calculated as optional actions by a DLF from the measurement ranges of the attributes of objects it perceives in its world; the simulated concepts are connected to simulated perceptual concretes and each other in chains of increasing abstraction (both more general and more specific), chains that begin with the perception of actual measurements of specific objects in reality, and end with specific actions taken to cause effects on these objects after the conceptual chain has been traversed. Concepts are symbolized by natural language words; relationships and events are symbolized by simulated natural language sentences that are their symbolic equivalent; simulated natural language sentences are identifications of the DLF's world in symbolic form. Each sentence is a complete simulated thought for the DLF, and each simulated thought represents an event or scenario in reality (of which the DLF is a part).

• Finally, all of the above innovations plus interaction with reality and with a human teacher leads to the emergence of simulated natural language understanding at the level of simple sentences. The decoding and encoding of simulated natural language sentences is accomplished by the DLF tracing the calculation chains and symbolizing words it has calculated and stored in memory in the process of forming its simulated concepts; the concepts are the content for both the meaning and grammatical specifications of its simulated natural language sentences.

As with the simulated concepts used by a DLF, the elements of this invention form a chain of increasing complexity, but one that is firmly connected by the links of causes and effects to the computer hardware that animates it.

5.10.2 The Invention is Useful

This invention is a useful addition to the current state of the art. A few of the uses the invention makes possible are listed as follows:

• The implementation of only the simulated percept data type, manually programmed, controlled by rules defined by human beings, and implemented on extant systems (without simulating consciousness) would make many computer systems more efficient by reducing the information units that need to be processed. For example, the percept data type will enable battlefield computers to process objects as relatively small attribute lists instead of huge lists of X,Y coordinates.

• Computers running DLF simulation software would be more intelligent and easier to use in general because they would require less knowledge by users to operate them. For example, the DLF could observe a user and over a short time learn to anticipate the user's needs like a human assistant might. Or, on a wider scale, DLFs expert in one subject could communicate over the Internet to exchange information and skills with DLFs expert in other subjects, DLFs could even work in teams to process the details of various tasks and enable human users to focus on strategic issues.

• Communication with a computer system running DLF simulation software would be much easier using natural language sentences that the system "understands" by means of its conceptual chains, as opposed to extant systems which have no understanding of natural language, but use natural language words as arbitrary symbols.

• DLF simulated knowledge and skills are stored in ordinary computer files so they can be copied, and sent to other DLFs anywhere over the Internet. This means, for example, that valuable solutions to problems or knowledge of dealing with problems discovered by a DLF in one part of the world can be almost instantly available to any DLF on the Internet, for its own local use.

All these uses, and many more that the author or others will think of in the future, make this invention extremely useful to many individuals, businesses, and other organizations.

5.10.3 Reduction to Practice

The DLF simulation system as an invention can be reduced to practice in a relatively easy and straight forward manner by means of the six steps described in the introduction to this chapter.

Any small team of 2-3 expert object-oriented programmers, after reading this patent description and studying sections of the references, then integrating that information with their programming experience, can reduce the invention described herein to practice with two or three years of focused effort. A proto-type DLF capable of simulating perceptual consciousness is already partially completed and successfully generates simulated percepts. A version with very limited simulation of conceptual consciousness using a simulated world could be developed in even less time to serve as a demonstration proto-type, probably in a year at a cost of one half to one million dollars, which is a tiny amount of money compared to many of the software projects companies develop routinely today.

Form or Product of the Invention

The ultimate product of this invention that will be sold or licensed is a design architecture for life-form and consciousness simulation, and a set of computer files that embody the attributes of that architecture; the computer files will be an embodiment of the kind that a computer system will be able to animate, as described in this document and the patent application.

The computer files will contain not only source code for the DLF Program and its documentation, but also the values and knowledge of one or more DLFs at some level of development. The reduction of this invention to practice will result in a simulation system that users can animate on their own computer systems and then apply to their various purposes. A company desiring to sell or license the DLF simulation system would first need to build and test the system to some level of development for demonstration purposes, in other words, create a proto-type system and a development programming environment in order to have a viable product to sell or license.

The exact state of development of the simulated consciousness of the DLFs in the product form of the invention will have to be determined by experiment and interaction with potential customers or licensees. The most likely state will be one in which the DLFs' simulated consciousness has reached the level of understanding simple natural language sentences and being capable initiating choices, but not necessarily having expert "knowledge" in any particular field, though simpler systems could also be made available.

This means in effect that the DLFs in such a system would have formed many simulated percepts of reality and have interacted with reality sufficiently to form key conceptual chains, enough simulated concepts and conceptual chains in fact to reach the ultimate genera for crucial parts of the conceptual hierarchy, including such concepts as "existence," "object," "action," "identity," "place," "world," "causality," "self," "consciousness," and so on, so the DLFs would have a working conceptual knowledge of reality and their own self-consciousness. After all, forming simulated concepts and using conceptual chains to calculate the meaning of natural language words is one of the essential attributes that differentiates this invention from the current state of the art systems.

That being said, however, there may be cases in which potential customers or licensees would want less developed DLF simulation systems for experimental purposes or in order to study how to improve DLF simulated consciousness, learning, or for some other limited purpose such as use in animating toy robots, dolls, or toy pets; in this limited form the invention could be marketed as a sort of life-form and consciousness simulator "toolkit" that others could use to develop specialized applications of simulated consciousness. It is therefore difficult at this early date to foresee exactly what states of development will be the best to offer to customers.

The best description of the DLF simulation system product at this point is that it will be a life and consciousness simulation design architecture embodied as a collection of computer files containing the DLF Program source code, DLF life simulation data (including simulated life values, energy packets, and internal control system of a teleological agent), DLF consciousness simulation data (including simulated percepts, concepts, and values), DLF action method source code capable of both necessitated and optional actions, DLF simulated world methods source code, and product documentation files.

At least that is the authors best estimate at the time of this writing. However, some of the specifics of the form of the product of the invention will undoubtedly change as the requirements of the market become better known.

5.11 General Summary

At the beginning of this book, I wrote that I would present, describe, and explain the ideas required to for an experienced object-oriented programmer to build a system capable of simulating self-consciousness.

I believe I have accomplished that goal. Now it is up to you to negotiate an agreement to license DLF Simulation Technology and build your own DLF simulation systems.

To request a license agreement for DLF Simulation Technology or to get a free copy of this book in .pdf format, please visit our web site at: