The RICX Perceptual Simulation Technology™ Architecture The RICX Perceptual Simulation Technology™ Architecture
The following are exerpts from our 40 page white paper that describes the RICX Technology architecture in detail and how it works. Parts of the paper are not published here, as they are remain confidential.
Introduction and Context
This paper describes a new invention called a Reality Identification Cortex ™ (RICX) that was first conceived by the author in February 2004 and has been under research and development since that date. During the summer months of 2006, more details of the operation of this technology have been worked out, so it is now time to describe the technology in a confidential white paper. The author may also eventually expand this paper into Chapter 6 for reference 5 at a future date in order to explain in detail how RICX technology integrates with DLF TM Technology. (Reality Identification Cortex, RICX, and Digital Life-Form (DLF) are trademarks of Blue Oak Mountain Technologies, Inc. - See our About page for details.)
The research for RICX technology was motivated by DLF technology and the need to automate the latter's sensing and data input process. The DLF technology is a layered model architecture that can use computer systems to simulate life-forms and biological consciousness as processes that are similar to and emulate the processes exhibited by higher animals and people, at least to whatever degree that is possible with current technology. DLF technology substitutes computer technology that is specially programmed for this purpose, technology designed to emulate biology. DLF technology is designed to simulate consciousness to help robots perceive and identify objects in the world around them and to simulate the formation of human-like concepts of these objects and their relationships in the world. By doing so, robots running software based on RICX and DLF technologies will be able to use the human-like concepts in conjunction with a continuous stream of human-like sense perceptions to inductively produce and “understand” simple natural language sentences about objects in the world. In the patent application for DLF technology and the proof of concept DLF Program, most of the input that simulates sense perception had to be hand coded by a programmer. This was necessary due to the fact that no technology to automate this process was available at the time of the original DLF Technology design (see Reference 5). RICX technology will substantially improve the effectiveness of a DLF technology robot.
A DLF Technology robot could have much better performance if the sense perception process of detecting and identifying objects in the world was automated and worked more like it does in real life-forms, which is precisely what RICX technology is designed to do. So for the past years since 2004, the author-inventor has worked on perfecting RICX technology in order to make an improved simulation package that combines both DLF and RICX technologies.
Comparison to State of the Art Technology
It is well known that most state of the art pattern recognition technology that could be used to simulate sense perception works only in very narrow domains. Probably the best example of a somewhat promising new technology that will be able to work in certain domains is the HTM technology being developed by Numenta, Inc. (See references 7 and 8 below.) HTM technology holds promise due to its use of active memory and mathematical design. However, though both technologies are mathematical in their design, HTM technology is very different from RICX technology in its design details and function. HTM technology uses probability matrices, whereas RICX technology uses direct measurement of the world and mathematical techniques to identify objects and their relationships in their real world context , rather than by creating probability models.
In order to more easily grasp the essential differences between DLF with RICX technology and HTM technology, we need to compare some differences in the concepts, processes, and functionality between these two kinds of systems:
HTM Technology (Numenta)
DLF/RICX Technology (Blue Oak)
Concepts
Concepts
Causes: Objects in the world that have persistent structures
Causes: The identities (attributes or properties and measurements) of objects in the world in action.
Objects: No clear definition of objects outside of HTMs is provided other than the term “causes.”
Objects: Things that exist in the world independent of consciousness and are what humans identify them to be using sense perception.
Inference: The detection of novel input by a mechanistic process similar to pattern recognition, where new input is compared to a probability model of invariant representations of previously detected causes. The goal is to force the dynamic world into a static data set suitable for computer analysis.
None at the sensory levels: Inference, Prediction, Belief, Direct behavior, and Categories are all part of a higher level of processing called “conceptual consciousness” for DLF technology, a product of it, or controlled by other simulated consciousness processes, not part of sense perception.
Moreover, the goal of sense perception for DLF/RICX technology is NOT invariant representations, but rather identification of invariant relationships in the world , including the contextual information that is part of the world's dynamic nature.
Prediction: Combining memory of invariant representations of previous causes.
NOTES:
1) Consciousness is not part of HTMs, which are purely mechanistic automatons that do not simulate goal-directed behavior for their own sake.
2) See References 7 and 8 for all concept definitions in this column.)
Consciousness: A biological process that in humans consists of the perceptual and conceptual levels explained in Reference 2. Perception is the direct awareness (measurement) of the objects in the world using sensations and percepts that identify objects and implicit relationships by measuring differences and later using these data to form concepts. Concepts formed by this method are a new data type that relates mathematically abstracted relationships implicit within the data contained in percepts.
Direct behavior: Controlled and based on calculated predictions, previously modeled probabilities of when behaviors are activated.
. Perception: Simulated by a RICX as part of DLF technology. By doing so, this process mimics the sensors and neo-cortex of humans using simulated bio-automatic processes to measure (detect and differentiate) and identify objects in the world by producing simulated “sensations” and “percepts” (integrations of measurements), which are data types that store the identity of objects as property/value pairs that are derived from and connected to the original measurements that produced them. The goal is not to produce “invariant representations,” but rather to limit the variation to a detected measurement range (mathematical relationship) and thereby reduce the number of units to be processed in order to make object identification easier and more efficient.
Categories: Called “pooling,” the formation of categories for pattern recognition and other behaviors is performed to create “many to one” relationships between detected patterns and invariant representations used to make recognition comparisons.
Concept Formation: As part of DLF technology and using a quasi-mathematical procedure based on processes in Reference 3, concept formation methodically calculates categories that are “many to one” relationships using the data stored in simulated sensations and percepts to produce abstractions that are symbolized by words in a natural language such as English, as well as connected to the original measurements used to produce them.
Beliefs : The moment to moment distribution of possible causes of what is happening in the world represented as a Bayesian probability network.
Induction and deduction (logic): Using processes described in Reference 1, DLFs can encode, decode, and validate simple natural language sentences (premises or simulated thoughts) that are about causation and other specific and relational information of objects in the world and their dynamics.
Hierarchy: HTMs are designed to simulate the hierarchical structure of the human neo-cortex and the world.
Hierarchy: A relationship between concepts (abstractions from abstractions) at the conscious level, and one that is also simulated in a different way to produce percepts and mimic some of the operations of the human neo-cortex (see RICX above).
Process
Process
Sensors sense the world and their output is the input of an HTM system with a design based on the human neo-cortex. The HTM receives spatio-temporal sequences from sensors. It then infers causes in the world as a hierarchy of beliefs using a Bayesian Network of probabilities to produce a probability model of invariant representations of the inferred causes in the world. Such models can then be used to do pattern recognition of objects and direct other behaviors.
The energy (such as light) in the world is sensed by a simulated Digital Life-Form called a DLF. The DLF simulates perceptual consciousness as exhibited by higher animals and humans using a RICX technology system to simulate sense perception and the identification of objects in the world by detecting sensations and objects (implicit relationships) using mathematical comparison of differences. It does so by measuring sensations and objects, and calculating their property/value pairs. In vision, for example, the resulting identities of sensations and objects are stored in a new data types called IR-Pixels and O-Pixels, which generically known as simulated “sensations” and “percepts.” IR-Pixels are patches of color that are irregular groups of ordinary pixels that simulate retinal sensors. O-Pixels are groups of IR-Pixels that are grouped based on perspective, common motion, or the occulting edges of figure ground and are simulations of real objects in the world. O-Pixels are the invariant identities of sensed objects. Further, by using these invarient data, by comparing these base-level, perceptual identifications of objects, concepts (abstractions) are later calculated and validated by a higher level of processing in the DLF system. Concepts are symbolized in the system by natural language words provided by a human trainer or tutor. Concepts are then used in conjunction with additional sensations (IR-Pixels) and percepts (O-Pixels) and the process of logical induction to produce premises , which are simple natural language sentences about objects the system has sensed in the world. Logical deduction and reductions are also used as means to check the logic of the premises produced by induction. With a constant flow of new sensations and perceptions of objects in the world, calculating more concepts and inductions, and by interacting with humans, DLFs can produce simulated human-like knowledge which can be used to do object and relationship recognition, causal recognition, and control robots to direct many other kinds of behavior. (See Reference 5.)
Functionality
Functionality
When marketed, HTMs will be able to simulate some human-like conscious behaviors, but these behaviors will be based on a Bayesian probability model of the world, not direct measurements of the properties that identify its objects the way some animals and people do. This technology, like all others in the state of the art, is strictly a mechanistic human tool designed only for human purposes . It is not a close simulation of goal-directed, biological consciousness and other actions observed in life-forms, actions that the life-forms take for their own sake driven by their need to survive.
DLF and RICX technology together are designed and intended to simulate the biological, goal-directed behaviors of life-forms that possess consciousness. Though running on a mechanistic, computer-based platform that is substituted for the biological layer, the simulated DLFs are primarily designed to function as if they were goal-directed (teleological) biological systems. That is, as if each DLF was a self-generating, self-sustaining, self-regulating life-form that identifies the world around it for its own sake and its own survival. Though the computer system or robot running the DLF/RICX software is still ultimately a human tool, it is designed to mimic the goal-directed autonomy of a biological entity that lives for itself , for its own survival first and human goals secondarily, such as a dog or a horse trained for explosive sniffing. Identification (consciousness) is the form of sense perception for simulated life-forms that is designed to mimic that same function in real life-forms, where the process is survival driven . Consciousness detects differences in the world ( differentiation ) and groups that which has been detected in various ways to reduce processing units ( integration ). In this way, millions of pixels are reduced to a few hundreds of sensations and percepts of objects in a visual scene, for example. Then, additional knowledge is later inferred from perceptual identifications by higher level conceptual processes that form abstractions based on implicit relationships using perceptual identifications. Since this abstract information is based on and connected to direct measurements of objects in the world using conceptual chains, is itself real and certain , not merely probable.
Detailed Description of RICX Technology
Now that the essential differences between HTMs and RICXs are clear, we can move on to a detailed description of how RICX technology works. By way of introduction to this section, a few comments are necessary.
Both RICX and DLF technology (see reference 5 for specifics on the latter) are adaptations of processes based on the concepts and principles of the nature of biological consciousness works and how it works, as discovered and developed by Ayn Rand (as well as others who have further developed her ideas - see references 1 and 2). Based on Ayn Rand 's clear explanation of the nature of consciousness (“Existence IS Identity. Consciousness Is Identification.” – see reference 3), the ideas in this paper apply that explanation to the design of an entirely new technology for simulating consciousness using computer hardware and software that is specially programmed to simulate key biological functions. This paper takes the clear understanding of what consciousness is and how it works in people, and then explains how this understanding can be leveraged to simulate the processes of consciousness using computer systems, to whatever degree that is possible with current computer technology. It should be noted that Ayn Rand's ideas take sense perception as a given , a starting point for her work in philosophy and writing fiction. As far as is known, she never investigated how sense perception works in detail, other than to say that “a percept is a group of sensations automatically retained and integrated by the brain of a living organism.” (See reference 3.) Rand used sense perception defined in this way to develop her theory of concept formation (see reference 3), which is the theoretical foundation for the operating theory on which DLF technology is based. While DLF technology can operate in a limited way with pre-programmed sense perception software to enable it to identify objects in the world, to reach its full potential, DLF technology requires a means of doing so that more accurately simulates the automatic nature of biological sense perception. This paper is the result of the author's work to observe the operation of sense perception and generate logical inductions regarding its operation using Rand 's clear explanation as a guide. RICX technology is the result of that work and is intended, at a programming level, to provide sense perception “service” for DLF technology robots.
As noted earlier, commonly used pattern recognition technology and the HTM technology produced by the work of Jeff Hawkins is the closest the state of the art comes to providing tools that could be used for simulating sense perception. Jeff's book and white paper (see references 7 and 8) provide and excellent summary of what is known about the neurology of how the human brain operates and performs sense perception, such as how the eyes make saccades approximately three times per second and the layered processing of the human neo-cortex (see reference 7, Chapter 3). Other researchers have identified that even adult brains rewire themselves by growing new neurons, and that individual neurons are wired to detect specific objects (see reference 6).
After having worked out the basics of RICX technology, the author's research led to the work of James J. Gibson's “ecological approach to visual perception” (see reference 9). Gibson provides a basis for understanding the overall context of how sense perception processes data, the complete view as it were, as well as the integration of the various facets of sense perception, which not only cannot be disassociated from each other, but cannot be functionally disassociated from the actions of the sensing life-form as many researchers routinely do. Moreover, Gibson makes a powerful case that the human concepts of art, photography, and communication are not appropriate for formulating the premises in a theory about sense perception. Vision, touch, and hearing, for example, are not separate sensory “channels” like TV channels, and visual perceptions are not processed like an artist does to make a painting or a digital camera to make a picture. Naïve realism is a false theory, and there is no “little man” to receive the communications over the “channels” or view the “pictures” when they arrive inside the human brain.
All of the descriptions of the direct observations of the function of neurons, their explanations, and the summaries of other experimental results in cognitive psychology provided by Gibson have been extremely helpful to the author in the creating the design of RICX technology.
The RICX Architecture
Since this is the first preliminary and complete design description and explanation of a brand new technology, it is premature to specify some details, details that can only be worked out after many actual experimental operations have taken place. However, one has to start somewhere, so the following are the author's best estimates at this point of what the RICX design specifications need to be to perform their desired function. The details will be refined moving forward.
Preliminary Design Specifications
Sensors: Since the main purpose of RICX technology is to mimic human sense perception, the first requirement is sensors that approximately match the detection specifications for the human senses. Luckily, many good ones are made already and available for state of the art robots. It is important to keep in mind that the role of sensors used for the purpose of simulating consciousness is simply to transduce the energies that impinge on them and convert these energies to another form of energy or digits in the case of computer systems, all the while conserving the identity of whatever patterns each energy may carry. This conversion in kind of energy is not a disintegration of the information in the patterns of the energy from the plenum of reality, but rather it is a facet of the plenum that is the real world outside the sensing robot. Each sensor is like the facet of a diamond in that it provides a perspective on the plenum that is the stone itself. The information in these facets becomes the content of consciousness that needs to be conserved for future processing as it is transferred into the RICX system, just as such informatin is conserved by the sensory systems in animal perceptual systems.
Low Level Processing: Once transduced into the system, the various patterns in the energies must be identified by properties and measurements from differences detected in the raw digital data. Identification requires the comparison of sensor data from various times as will be explained below. Simulating the low level processing such as occurs in the retina of the eyes, the pressure sensors of the fingers, and the cochlea of the ears, and so on, may be possible with off-the-shelf products. Many good sensors are already sold for robotics and other uses, as well as the drivers that aim and focus them. In the case of vision and hearing, it is necessary to saccade these simulated “eyes” or turn the head to enhance hearing or odor detection by providing more comparison data for difference detection. Drivers to do this may or may not also be found for sale or adapted from existing state of the art sensor software already designed for robots. This remains to be investigated. Commonly used bitmaps (as currently designed) may or may not work for simulating vision and other senses because they are not designed to saccade and to maintain the relationships of various areas of the visual field like the human visual system does. Bitmaps were design as mechanistic human tools for various computer oriented purposes such as communication and picture processing, not as means to measure and identify information to be transferred to the visual cortex in a brain for survival purposes. But, it may be possible to use some of them anyway, or modify them so they can be used for this new purpose.
Neo-Cortex Simulation: The most difficult part of the RICX design specification at this point is that of the neo-cortex simulation. It is too soon to say whether it will be necessary to build a hierarchical structure with 6 layers as the human neo-cortex apparently has (see reference 7), or whether another design approach can be found that can effectively simulate how sense perception operates. The main thing that must be kept in mind during the design process is that whatever the details of the specifications are, they must support the operational efficiency of the perceptual process as described in the next section below, while being as true as possible to the way similar processes occur in biology in living animals.
Reality (the human scale world) is a plenum, and life-forms perceive it as such. There is evidence that the seat of the processing of perceptual consciousness is the brain stem in higher life-forms (see reference 10). Without implying naïve realism, the brainstem may be the locus where sensory input is integrated inside the brain, the neural plenum of reality so to speak. The brain stem may be the place inside the human brain where the control processing for action decisions interfaces with the faceted conscious perceptions from the senses occurs.
In order for action decision processing to happen using percepts (identified and measured invariant patterns and objects), the perceptual facets must be converted and integrated into the appropriate input by the lower sensory processing layers that process the raw digital data from the sensors. The result will be a system that “sees” roughly what you and I see when we look at the plenum of reality, though in a somewhat different form because humans are not digital. The form does not matter. What matters is that the identity of the implicit relationships is conserved as this processing takes place.
Having made the point about reality being a plenum that processed in the form of preserved relationships that are mathematically extracted (not disintegrated and then reintegrated), it is necessary to explain that that fact does not totally eliminate the need for integration to be performed at some points in the processing of sensory data. The transfer of data from receptors in biological eyes, for example, is a largely parallel process. The optic nerve is known to have millions of neural pathways. The computer technology that the RICX system will be using, however, is much more a serial process. In addition, computers process differently than neurons do, so the identification of say a patch of “blue” by a DLF robot using computer technology will require some calculations to integrate the sensor data, whereas a biological neural network may not need to do so. But this does not matter provide the implicit identity relationships are not lost.
In terms of those lower level processing layers, it is clear that probably at least 4 layers will be necessary for the first proto-type, and possibly more. As explained in the next section, percepts are not a simple data type, but are integrations of pixels into sensations and sensations into objects. The integration is done by mathematically extracting certain relationships implicit in the identities of the pixels. How many levels of processing will be needed to make even a simple demonstration system operate in a practical way can not yet be absolutely determined at this point.
It should be noted that Gibson might not agree with my assertions regarding integration and pixel processing (see reference 9, Chapter 15, pages 267-290). However, Gibson was talking about the fact that the human perceptual system (which is not digital) and does not process pictures for the sake of naïve realism's “little man.” Gibson specifically makes clear that the goals and objectives of picture and graphic art processing in general is not the same as sense perception, so using the concepts from those fields unmodified is a big mistake. It is probably safe to say this is one of those areas that will require much experimentation to make a workable perception system for DLF robots. And, while it is possible it may ultimately be necessary to invent an entirely new kind of analog system to properly mimic human sense perception for some advanced “Mr. Data” type robot, the author believes it is very likely possible to leverage extant computer picture processing technology in some ways to build a reasonable simulation of perceptual processes for DLF robots, so they can identify objects in reality. For example, heat seeking missiles use simple negative feedback control systems and are not even close in capability to a DLF robot, let alone dogs trained to track game for hunters. Yet, the author has seen video of one such missile make a U turn and chase after a jet fighter the same way a well trained game dog would chase a rabbit or a fox. In that one respect the missile is a reasonable simulation of the animal behavior, even though it works in an entirely different manner. If something similar can be done for sense perception using appropriately modified extant technology, a useful system will result. The main point here is that the reader needs to think of sensor input and its processing into simulated sensations and percepts not using the concepts of communication channels and picture or artistic processing, but using the concepts of consciousness (as defined by Ayn Rand) as they apply to various kinds of sensors as facets on the plenum of reality. And further, that the processing of sensor data is the identification of invariant patterns as these concepts of consciousness are understood in cognitive psychology experiments and the resulting observational data. That being said, there is no point in renaming well accepted terms such as “pixel” merely to maintain consistency with the views of Gibson. The RICX perceptions simulation system is digital and cannot work the same way a biological system does, nor does it have to.
DLF Technology Interface: Since one of the key reasons for developing RICX technology in the first place is to automatically generate percepts as data for the higher level processing of DLF technology that simulates human reason and simple natural language capabilities, the data requirements and control requirements for sensor direction and focus on the various facets of the plenum of reality must be taken into account from the outset. Therefore, the interfacing designs and relationships between these two technologies (RICX and DLF) must be planned and worked out carefully as well. The most important aspects of these relationships are that the RICX design must accommodate the identification of reality as a plenum by having as inputs information about its various facets, and output identity lists because the DLF processing layer design requires identity lists* of property/value pairs as its inputs. So to begin with, the entire plenum must be sensed as a piece, including its dynamic aspects as the sensing system moves through the plenum, and the dynamics that result from other actions of the sensing system, such as saccades, selective focus, head turning, locomotion, and so on. Motion causes both global and local changes in what information arrives at sensors. What Gibson calls the “optic array” flows around and past the sensing system. When observing specific objects in a scene, the various points of observation determine the information each eye senses (light travels in straight lines), as well as what parts of objects may be occluded from view by various edges of other objects. Human observers learn what such changes imply about objects as observers move and change their viewpoints.
* (It should be noted, however, that the above current working design specifications do not rule out other future designs not yet thought of and which DFL and RICX technology may implement, that do not use lists.)
The figures above from Gibson show what this relationship between the observer and the observed means graphically (see reference #9, Chapters 5 and 7 respectively). Once sensed by two or more sensors that transduce energies from two or more facets of the plenum of reality (e.g. vision, sound, touch), and the relational identity information thus gleaned must be processed so it can be identified. That is, so the invariant patches of color, sequences of sounds, areas of pressure, kinesthetic forces, odors, tastes, and so on can be differentiated from those that vary constantly, and from each other. Note that these various facets of the plenum do not have to be integrated because they are all integral parts of the plenum of reality already. They, in fact, need to be disintegrated . The invariants must be differentiated from the plenum (chunked into process-able pieces so to speak by the selective focus of consciousness) in order to be identified . Differentiation is one of the two key processes of consciousness, but it does not mean that the differentiated data is somehow isolated from it context forever, isolated from the plenum of reality from which it is derived. Differentiation simply means that some parts of reality have different measurements than other parts, and one of the capabilities of consciousness is to focus on this vs. that. To differentiate is simply to make a metadata note that one part of a scene is more like its immediate surroundings than another part of the scene, or that one object in a scene is bigger or smaller than another as part of the processing. Differentiation is part of the process of identification by comparative measurement and relationship extraction by abstraction . It is not a process of total disassociation or disconnection. To do the latter is to drop context and render the information just gleaned from reality useless.
Once the differentiation of invariants is done, objects can then be differentiated to produce a collection of identity lists of property and value pairs for each object at least partially visible. The identity lists are the output for higher level processes such as action control, concept formation, logical induction, logical deduction, reduction, integration, and natural language processing (see reference # 5, Chapter 5 for a complete explanation of how these higher processing layers work).
The question that may arise at this point is: Where is the representation of reality for the DLF robot to perceive?” The answer is that there is none . Reality is outside the RICX system (where it belongs), and it stays outside the system, integrated and as a constant reference whenever needed. If the DLF robot needs to, it can always look at the plenum of reality again, and yet again, as animals and humans do constantly. The output of sense perception system is the invariants and the identity information of objects, not a photographic or artistic representation for the “little man” of naïve realism to look at that has been integrated from several “sensory channels” as most cognitive theories assert.
While there is no “little man,” some research published in 2007 that was done with children born with little or no neo-cortex does, however, indicate that there may be a place in the brainstem where high level conscious processing apparently is localized. The following are a couple concluding comments from one of the papers: “After making his own observations of children with hydranencephaly and their families, Merker seconds that point. He notes that well-treated youngsters born with little or no cortex regularly display brief losses of consciousness due to absence epilepsy, a clear sign that at other times they're conscious. … Perhaps most intriguingly, kids with hydranencephaly demonstrate that the brain stem is not simply a reptilian relic stashed in the brain's basement. ‘The human brain stem is specifically human,' Merker says. ‘These children smile and laugh in the specifically human manner, which is different from that of our closest relatives among the apes.'” (See reference 10.) It is the opinion of the author that this area of the brainstem is probably the interface between perception and action, the area that provides control in the form of action decisions based on inputs from the neo-cortex and the rest of the human nervous system. As such, simulating it will be an important part of any robot system that attempts to similar perceptual consciousness.
RICX Process Operation Basis and Overview Explanation
The following process describes the overall action of simulated sense perception for RICX technology. The focus is on vision because it is one of the most complex senses and where the author has chosen to begin this process description. However, the reader needs to keep in mind that vision is only one facet of the plenum of reality, and that simultaneous and parallel process are running in a DLF robot sense and process the other facets of hearing, touch, and so on. But it is the same processing system (simulated consciousness) that will eventually identify and output the additional property and value pairs and implicit relationships that are available from the sensor output from other facets. This will occur as a part of an expansion of the same processing stream and internal data structures used to identify the properties for vision input. The other facets are not processed in separate pathways or “channels” as occurs in state of the art systems. ( Note : Similar and equally detailed explanations for the facets of the other senses will be written by the author at a future date as this work progresses. These will be added by the author to what follows as a systematic expansion of what follows when time permits.)
This entire identification process of what is in the sensory data is based on the “some but any” principle that Ayn Rand identified as part of the concept formation process, and Drs. Leonard Peikoff and David Harriman later extended to the process of logical induction or premise formation (see references 1, 2, and 3). The author has now identified how this principle also operates to integrate sensor outputs into sensations (IR-Pixels when simulated) and sensations into percepts of objects (O-Pixels when simulated). The “some but any” principle is the very basis for extracting invariants from constantly changing data. The invariants are the measurement ranges that various property and value pairs typically fall into as a result of the natural identities and implicit relationships found in reality. As both Aristotle and Ayn Rand have said: “A is A.” In other words, a thing is itself, its identity and it is what it is. That fact is self-evident, implicit, and it is axiomatic. (The author is doubtful he would ever have recognized this new application of the “some but any” principle had it not been for the two previous applications identified by Rand , Peikoff, and Harriman - see references 1, 2, and 3.)
The basic idea of the “some but any” principle has two main components as follows:
First , the identities of objects in the world consist of one or more characteristics, attributes, or properties (the author prefers the word “property”). Every property is quantify-able and must have a unique measurement value (number) associated with it. This is the case because to exist at all every “thing” must be a “something.” Or, as Ayn Rand put it: “Existence IS Identity.” (See reference 3) In other words, to exist at all an object must have at least one property , and a property must exist in “some” specific amount , but it may exist in “any” amount that is the typical measurement range for that property. The physical properties that exist independently in various arrangements in reality are the source of the information that becomes the content of consciousness when sensed and processed in a specific manner.
This fact is supported by the laws of science, such as physics or chemistry. For example, an earth bound land animal can be no smaller than chemical processes allow and probably not much bigger than the biggest dinosaur was. Or, it is known that the size of insects, which do not have lungs, is limited by the percent of oxygen in the air and its ability to defuse to internal cells in their bodies. Or, for another example, a stone that had no property of “size” would not be a stone; it simply would not exist . An endless number of similar examples are easy to think of.
Second , that while the identity of every object is therefore unique, due to its specific list of properties and their measurement values, in any group of similar objects, the measurements of all of their commensurable properties will necessarily fall into a range of values . If you think about it, this is a natural consequence of the first component of the “some but any principle.” Some things will be bigger or smaller than others due to natural variations of the materials of which they are made or the requirements of the various environments in which they exist, and these differences will fall within the ranges that the underlying physical laws cause. Moreover, throughout biological evolution, it should be no surprise that life-forms have taken advantage of and naturally selected for these facts in the way they use their consciousness to identify objects in the world in order to improve their chances of survival.
Ayn Rand discovered that the “some but any” idea was the underlying principle that enables the methodical concept formation process she discovered in human consciousness to establish a “one to many” relationship (between a concept and its many instances or “units” as she called them) based on the observation and comparison of two or more similar objects, as opposed to many other different objects. She discovered that the “some but any” principle is the basis for including and integrating an unlimited number of specific objects or “units” into a group of two or more similar members based on the observed measurement ranges of their properties (“Consciousness IS Identification.” - see reference 3). It is the implicit mathematical relationship of some objects being in certain relative measurement ranges (as opposed to other objects which are not), that is the “universal” that Plato and many other philosophers could never find. The “universal” is not in the objects as Aristotle thought, nor is it in heaven or arbitrary as Plato and many modern philosophers have thought. Rather, the “universal” is a mathematical relationship that is implicit in the identities of the observed objects and the cognitive processing needs of the human consciousness that is identifying it. That is the monumental discovery of Ayn Rand.
Drs. Peikoff and Harriman later recognized that this principle also enabled the formation of one or more valid inductive generalizations based on the observation of just one instance of causal action , provided the acting objects were already instances of valid concepts that were part of a system of valid concepts . Peikoff and Harriman recognized that the conscious process of logical induction effectively leverages the “some but any” relationship that already exists in previously formed and validated concepts (such as “push,” “cause,” “ball,” and “roll”) in the examples they themselves used to understand logical induction as applied in the study of physics (see reference 1). So for example, if one had the previously validated concepts just indicated above and observed the experience of pushing a ball only one time, one could then induce the fact that: “Pushing causes balls to roll.” Why? It is due to the fact that the “some but any” principle was used to form all the individual concepts in that premise and the rest of the conceptual system in the first place, and so the induction is valid because the concepts are valid (concepts connected to observations directly or through chains of other valid concepts, and not because of probabilities), and because the entire system has already been validated. The premise is valid because the premise's meaning is the union of those concepts . It is not only perceptual observation that validates the process of logical induction, though that is part of it, but as stated, every valid induction is supported by the validity of the entire human conceptual system and by all of the observations of reality and thinking that were necessary to build that system in the first place . This is the power of the “some but any” principle. (See reference 1)
DLF technology uses an adapted version of the “some but any” principle to enable computers especially programmed for this purpose to simulate the goal-directed action and the conscious processes of a life-form to calculate a special data-type that simulates human concepts. Simulated concept formation is a level of simulated consciousness that forms and processes abstractions, abstractions that are formed methodically on the basis of the simulated conscious identification of one-to-many relationships. The one-to-many relationships are in turn based on the observation of simulated sense perceptions of real objects in the real world. In the DLF technology system, ontologies are not arbitrary, but methodically calculated based on observation, measurement, and the “cognitive needs” of simulated consciousness.
During an investigation attempting to see if the state of the art HTM technology could be adapted to RICX technology and would work for automatically generating percepts (it will not), it occurred to the author that a percept of an object or scene is also an example of a “one to many” relationship : One percept relates some number of sensations (the integrated output from individual sensors), except that sensations and percepts are formed by automatic biological functions in the brain, as oppose to concepts and inductions, which are formed by conscious choice. Or, as quoted from Ayn Rand in the introduction to this section: “a percept is a group of sensations automatically retained and integrated by the brain of a living organism.” At that point, it became clear that sense perception, like concept formation, is about the reduction of processing units through the use of one to many relationships . So the questions for the author became: How are sensations integrated from pixels when trying to simulate consciousness using computer technology? How are the invariant aspects of percepts of objects and scenes integrated from biological sensations? What are the components integrated and how could these processes be digitally simulated? Are some of the invariants that are integrated “patches of color,” and if so, how are the integrations formed? What processes do the retina and neo-cortex perform to integrate and identify relationships in the energies that have been transduced from light waves into neuro-chemical sensations to make percepts of objects for people? How can computer technology be adapted to simulate those processes, to take constantly changing digital output from sensors and extract the stable invariant relationships that are contained in such data?
(Note: A large section of this paper detailing the specifics of the identification process in detail and its operational mathematics is still confidential, so it is not included here.)
Epistemological Status of RICX Technology Output
The epistemological status of the knowledge produced by the human senses has been the subject of controversy for thousands of years. While outside the scope of this paper, it is important to note that this controversy has been resolved as far as the author is concerned (see reference 2, Chapters 1 and 2). This issue is relevant here only because it is likely to be raised by some people as they try to understand RICX technology, people who are unaware there is an objective resolution to the problem. So it is important to briefly cover the issue here.
Other technologies, such as HTM and state of the art image processing technology, produce probability models or mechanistically calculated invariants based on what their sensors sense in the world as anticipated by human programmers for whatever purposes the employers of the programmers had in mind, but they do not identify objects the way conscious life-forms do by implicit relationships. These systems will undoubtedly be useful for a wide range of purposes, but they are not suitable for the simulation of consciousness in the author's opinion. Biological consciousness is a means of survival for life-forms. As such and to effectively serve that purpose, the knowledge it produces must be goal oriented as well as certain and real . Dead life-forms cannot act. Likewise, for simulated consciousness to effectively mimic its biological counter-part (to whatever degree that is possible with current technology), the “knowledge” simulated consciousness perceives must also be real and certain too. Even simulated survival depends on the accurate identification of the very real objects outside a simulated life-form (DLF robot), and probability models cannot be as effective as direct measurement and relationship identification for that purpose. For example, the survival of a robot operating unguided by humans on Mars is in just as much “danger” approaching a cliff as a real life-form. RICX technology uses no probabilities anywhere in its core process, but rather direct measurements that depend on the law of cause and effect and ordinary mathematical calculations based on those measurements. (Though this is not to say probabilities may not be useful as a secondary means to knowledge when there is too little data to calculate with certainty.)
The identities of objects and the identity of energies out in the world interact in causal relationships that change the identity of light, in the case of simulated vision for example. The light travels to the RICX sensor array system where, due to its design, causality is able to transduce the energy to a new type while conserving the identity and relationship information the energy implicitly carries through appropriate processing. The RICX sensors repeatedly measure various properties of the light and the input from other facets. These sensor measurements are mathematically compared to differentiate measurement ranges, the “some but any” principle and the mathematics of measurement ranges are used to group and integrate the differentiated measurement regions in the sensor array, and finally to integrate them into new data types called simulated sensations and percepts (IR-Pixels and O-Pixels). These data-types are a new form of invariant that does not exist in state of the art systems. The RICX system then uses these new data-types (simulated sensations and percepts) as processing units, and as data units themselves, for other processing. The new data-types thereby provide both processing economy and certainty as a primary means to knowledge of the world for DLF robots to use in a goal-directed way so the robots can better simulate human behavior, as long as sufficient data is available. Processing economy is provided because each sensation and percept is used like a variable in algebra, and is thereby equivalent to all the individual sensor measurements or regions of them in its area of the visual field. This is much more efficient than re-processing all the individual sensor measurements or regions of sensor measurements every few seconds or for every action decision. The way that sensations and percepts are calculated makes them unique and timeless for whatever duration their source energy or objects persist in the world.
Simulated sensations and percepts are the objects out in the world that the RICX technology system perceives in the sense that they are the invariant identities of those objects in property, measurement, and calculated form. This form is simply the form, the way in which the simulation system sees and identifies the world. The simulated sense perceptions that RICX technology produces are certain and as real as the laws of cause and effect and mathematics, just like yours are.
Usefulness to Science and Industry
There are many uses for RICX technology. In addition to automating the process of sense perception for DLF Technology, some other examples are yet to be developed areas in artificial intelligence, artificial life, robotics, medical sensing systems for disease and drug detection, identification systems for security and other purposes, micro detection and perception systems for genetics and nano-technology, providing the data for building ontologies for the so-called “Semantic Web,” and numerous military applications---to name a few.
The author is hopeful that the value of both DLF and RICX technology will be someday be recognized by science and industry.
References
(References 1-4 are available at http://www.aynrandbookstore.com ).
< Back to DLF Simulation Technology page
© Copyright 1998 - 2010, Blue Oak Mountain Technologies, Inc. All Rights Reserved
Welcome Page | About Us | Testimonials | FAQs