Building synthetic brains capable of human level discovery and invention...

Michigan Company Founded upon What Could Be History's Most Important Breakthrough in AI and Robotics

Summary - A Michigan-based platform company has been formed upon the foundation of a revolutionary new artificial intelligence patent that enables contemplative, synthetic brains to dynamically build themselves on a wide variety of computational platforms. The emphasis of this new business enterprise will be to stimulate the sagging Michigan economy through the creation of thousands of new jobs via subsidiary ventures spawned by these artificial minds.

What is this new and fundamental development in the area of artificial intelligence that makes such a potentially gargantuan enterprise possible? …Until now computer scientists somewhat mimicked human brain function by either painstakingly writing computer code or seeking the necessary data to train artificial neural networks. Nowhere to be seen was the kind of artificial intelligence anticipated by science fiction, the kind that didn't require human nursing, that could build itself, make significant conceptual leaps, or even become self-aware. Now, with the arrival of U.S. Patent 7,454,388, machine intelligence becomes self-assembling, cumulatively bootstrapping itself from a totally naive state to progressively higher levels of creative intelligence. No, this is not a Darwinian genetic algorithm requiring vast computational clusters, generating potential ideas through random mishap. This is a radically new form of synthetic consciousness that intelligently designs itself on the humblest of computational platforms, ultimately obviating the need for even computer scientists. The patent, authored by scientist/inventor Dr. Stephen Thaler, is appropriately called "Device for the Autonomous Bootstrapping of Useful Information" (DABUI) upstaging his prior landmark invention, "Device for the Autonomous Generation of Useful Information" (DAGUI, a.k.a., the "Creativity Machine") that issued in 1997. Using the methodology of this newest patent, vast numbers of artificial neural nets can autonomously interconnect themselves into vast brain-like structures called SuperNets™ that freely contemplate their world, think new thoughts, and develop their own sense of consciousness and self-esteem.

Based upon this impending patent a Michigan LLC, Imagitron, was formed on June 27, 2008. The intent of this company is to first create myriad projects based upon this newly issued patent that will in turn employ thousands in the state of Michigan . On September 25, 2008, the Michigan Senate approved a resolution recognizing the achievements of Dr. Stephen Thaler and his Michigan-based company Imagitron for "piercing the veil of technology and determining that Michigan will be home for his legacies." This resolution was sponsored by Senators Valde Garcia, Thomas George, Wayne Kuipers, John Pappageorge, Cameron Brown, Tony Stamas, Jim Barcia, Jason Allen, Roger Kahn, Gerald Van Woerkom, John Gleason, who all clearly understand the immense economic impact of an invention having such fundamental scientific and technical significance.

Heretofore, Thaler's lobbying success in Michigan has been attributed to his very affirmative response to queries about what the technology can do for the economy. To this question he posits numerous examples of products and services invented by this AI paradigm and currently sold by international corporations; that the technology has been at the core of a whole new generation of robots capable of Machiavellian improvisation on the battlefield; and that this paradigm has been central to the "connect-the-dots" problems in pre-911 data mining efforts by US intelligence agencies. In effect, it can do anything, stimulating the markets as it goes about its business.

When asked why this radically advanced artificial intelligence has not already commanded trillions of dollars in investment, Thaler invariably responds, "Imagine a genius walking into any culture, as a total stranger, and claiming he or she has the solution that others may have staked their careers and livelihoods searching for. It's a totally intimidating scenario to most, but one that must be dealt with in bringing this vast new artificial intelligence technology to Michigan ."

Queried about what this brilliant new development meant for the average person, his response was this. "To the person on the street, this patent represents the AI depicted in science fiction, truly intelligent and inventive supercomputers with a mind of their own, Machiavellian military robots, vehicles for life extension, and unlimited virtual pleasure. The difference between the scenario offered by the DABUI, and the future typically portrayed by the popular press, is that the synthetic brains behind all of these systems require neither the esoteric theories of academicians nor radically new hardware. It's here now primarily because it builds itself even on relatively mundane computational platforms.

In a sense, this patent represents a judgment day for the human race whose fate hinges upon the acceptance of these highly advanced AI paradigms.

The DABUI Simplified (This won't be easy!)

Summary - DABUIs are collections of untutored neural networks that spontaneously generate ideas among themselves, forming memories of the more memorable of these notions, and subsequently using these older epiphanies as foundations for even better ideas.

To appreciate Thaler's newest patent, one needs to understand what an artificial neural network or ANN is. Put as succinctly as possible, an ANN is a collection of switches, either hardware-based or software-simulated, that interconnect themselves so as to autonomously write arbitrarily complex computer programs. Rather than enlist humans (i.e., computer programmers) to supply the underlying logic of such self-built programs, the network is shown representative examples of inputs and outputs, during which time any given connection between synthetic brain cells can strengthen or weaken. The collective effect of such weight adaptations is that the connections between switches (i.e., neurons) capture the intrinsic relationships between applied input and output patterns. In effect, this process of self-organizing connections is not only how our brains learn to recognize fragrances, tastes, or faces, but also how they form opinions about the world. That is why such simple neural architectures are called "perceptrons," since they map input patterns from the outer world (i.e., raw electromagnetic and acoustic signals) to internally generated neuronal firing patterns of the brain that are the feelings or opinions we have about such worldly inputs.

But if you think that self-writing input-output programs are the underlying model of all human cognition and consciousness, think again. For instance, how does the brain engage in thought even without any sensory input from the world outside our skulls? Obviously, an internal dialog must be going on between the biological neural networks of the brain. To some, this might be the end of the story, as they simply anthropomorphize these neural networks into a kind of society of mind. But drilling deeper, one must inevitably ask how such biological neural networks can carry on an intelligent conversation with one another without any input whatsoever from the external world? …Remember, these networks aren't little people or homunculi! They are simply switches and interconnections.

In short, this is the question that Thaler answered as early as 1975, when he allowed a perceptron, one of these brain-inspired input-output programs to build itself. Then, instead of supplying the input patterns such networks typically require, he supplied no inputs whatsoever and began 'tickling' the connections between neurons, progressively increasing the magnitude of disturbances in a process that was tantamount to the introduction of heat to a sensory-deprived biological brain.  At the first stages of simulated brain toasting, the network activated through a series of patterns it had 'seen' in the course of its learning process. In other words, it was slowly reliving its stored memories. Then, as the simulated heat was turned up even more, the turnover rate of such memories increased to a fevered pitch until a limit was reached. Just above this critical threshold, the perceptron failed to activate into true memories, instead generating degraded memories or "confabulations." 

But to the Thaler, the human onlooker, many of these confabulations were potentially good ideas. Realizing, however that the 'goodness' of these confabulations was a value judgment generated in the perceptrons of his brain, a vastly profound idea was born: Use a synthetic perceptron to perceive value or utility to the confabulations produced in the heat-driven net, and if necessary, allow the perceptron to take control of the heat-like fluctuations in the confabulating network, what he coined the "imagitron." Such brainstorming neural networks, imagitron-perceptron pairs, were called "Creativity Machines." As early as 1995, whole societies of such networks were carrying out brain-like feats such as juxtapositional invention, as well as inductive and deductive logic. In effect, the much touted "singularity" had arrived, but Thaler lacked the capital to both publicize and educate a naïve public awaiting human like capabilities to arise in machines simply because the algorithms got more complex and the hardware faster.

In 1998, with the invention of extremely large and fast neural networks by a Creativity Machine, it was now possible to build such discovery systems from networks that could learn from both their mistakes and successes. In this way, a totally untrained system of brainstorming neural networks could reinforce the memories of those confabulations promising utility or value, while weakening the memories of those that were less promising. In effect, the confabulatory pattern that was the brainstorm one moment became a relatively mundane memory the next. However, when new ideas were required, these older revelations could be hybridized automatically into even more novel and effective solutions.

The most striking aspect of this radically new AI technology is its simplicity and elegance. Adaptive neural networks are simply box cared together and stimulated via ubiquitous noise to devise valuable concepts and strategies. The immense irony is that none of the traditional hallmarks of great scientific discovery such as complex equations and exotic theories are required. Nevertheless, this simple neural architecture is capable of generating both, suggesting that the brain itself is built upon this relatively simplistic scheme.

Why is This Patent So Important to Robotics?

Heretofore, there have only been two types of robotic artificial intelligence, what roboticists respectively call "reactive" and "deliberative control." In the former approach, sensors aboard the robot detect a situation in the external environment and then trigger some appropriate preprogrammed response, without any consideration as to which of several alternatives would be optimal. In a sense, this kind of algorithm is like a biological knee-jerk response or spinal chord reflex. In the latter, deliberative approach, the control algorithm receives sensor inputs that convey the robot's present situation, but then considers which of many alternative, canned responses is the most appropriate. In this way, there is some superficial similarity to the way the human mind considers alternatives and then commits to what it perceives as an appropriate course of action.

IEI has pioneered an entirely new school of what is called "creative" robotic AI, by building systems that deliberate appropriate responses to their challenges, not by enlisting previously coded responses, but by spontaneously inventing new behaviors that are appropriate to the demands on the robot. Prior to this patent, the first generation of creative robots were driven by DAGUIs consisting of pre-trained neural nets that were coerced into brainstorming sessions to invent the required behaviors. Now, with the DABUI, totally untrained neural networks could be combined into an ongoing cycle of experimentation and reinforcement learning. In effect, the robot carries out some tentative strategy that if it works, is reinforced as a memory in proportion to its success in meeting some overall objective. Later, when it is necessary to improvise some new approach to a problem, the self-bootstrapping Creativity Machine is able to hybridize the memories of semi-successful strategies into even more promising ones.

In addition to this general ability for a robot to bootstrap its own behaviors and cumulative wisdom, this patent teaches the following very fundamental robotic AI paradigms that enable a whole new generation of genuinely autonomous and ingenious robots:

1. How whole arrays of untrained neural networks (i.e., an uneducated synthetic brain) can collaborate in brainstorming sessions to generate extremely complex robotic plans of action that simple imagitron-perceptron pairs can't. (In other words, this patent is the foundation for self-assembling robotic brains consisting of vast contemplative confederacies of synthetic neural networks.)

2. How artificial neural networks may receive raw sensor inputs from a mobile robot to produce navigational fields (i.e., virtual reality highways) for the robot to follow. In other words, this patent teaches how such robotic brains can develop intuition about where they can and cannot go, in the same way we anticipate the pain of running into a wall when walking down a hallway.

3. How DABUIs may then be used to predict optimal paths through such environments using the spontaneously generated navigation fields generated by neural networks. (Put simply, this patent teaches how such robotic brains may contemplate and then select movement strategies based upon pre-generated navigation fields.)

4. How DABUIs may be harnessed to drive so-called "foveational" processes (i.e., the mechanism by which the brain scans a scene in search of some target or anomaly). (The patent teaches how these robots can autonomously choose where to point their sensors in search of some target, anomaly, or objective.)

5. How robots may build neural network models of some sought target without recourse to counterexamples (i.e., what a target looks like, and not everything else). (In other words, these robots are now capable of learning the essence of a whole genre of things, so that it can identify a particular kind of automobile, even though it may be painted or equipped differently from its training examples.)

The immense strength of this patent stems from the fact that artificial neural networks, in particular, the DAGUI-invented "algorithm-less" networks, are the only AI systems that can autonomously learn to recognize patterns on their own, without recourse to human programming, so there are no "scholarly" efforts prior to launching such inventive robots (i.e., genetic programming). Furthermore, through our pioneering efforts in DAGUI technology, we know that there is but only one way to stimulate the generation of useful behaviors for a robot to perform, the application of heat-like perturbations to a neural network based associative memory. And finally, such noisy associative memory devices (imagitrons), can readily absorb the essence of their successes and reject their failures, so as to progressively bootstrap themselves to higher levels of adaptive competence. In effect, there is no other way to arrive at self-bootstrapping, creative artificial intelligence, for robots or otherwise.

Some Key Events Leading to the DABUI

1975, The Foundational Experiment - Thaler discovers that internally perturbed neural networks can generate potentially valuable false memories and that such confabulations can then be promoted to the status of ideas by a separate neural net monitoring the progressive turnover of such confabulations and perceiving value therein.

1989, The True Creative Loop - Thaler discovers that the monitoring network can be used to control the perturbation level within the confabulatory net so as to optimize the turnover of potentially useful ideas. In so doing, he produces a working model of the brain's thalamo-cortical and cortical-cortical loops, the very foundation of cognition and consciousness. He also articulates the vastly important notion that ideas, both small and profound, are simply the result of biological neural networks producing false memories that are seized upon by attendant neural nets. (See below, A quantitative model of seminal cognition: the creativity machine paradigm, Proceedings of the Mind II Conference, Dublin, Ireland, 1997.)

1995, How Robotic AI Will Inevitably Parallel Nerobiology - Thaler proposes how an organism and its species can develop new behaviors through permanent and transient, damage-induced confabulation generation. (See Death of a gedanken creature, Journal of Near-Death Studies, 13(3), Spring 1995.)

1996, How Advanced Robots and AI Will Carry Out Deductive and Inductive Discovery - Thaler proposes how compound Creativity Machines can invent and discover by carrying out both inductive and deductive reasoning by chaining together both memories and confabulations within multiple noise-stimulated neural nets into associative chains (i.e., Terry bought ammonium nitrate, ammonium nitrate is an oxidizing agent, oxidizing agents are used in explosives, explosives are used by terrorists…and then a monitoring network discovers, "Aha, look out for Terry, who is a staunch critic of the US government!"). (See A Proposed Symbolism for Network-Implemented Discovery Processes, In Proceedings of the World Congress on Neural Networks, (WCNN'96), Lawrence Erlbaum, Mawah, NJ., 1996.)

1996, The Gamut of Cognition and Consciousness in Both Biological and Synthetic Brains - Thaler proposes that the gamut of all cognitive and conscious activity in the brain is the result of either transient reversible, or long term irreversible damage to biological neural networks, and that the more intense of these perturbation modes produces false memories that qualify as potential ideas. (See The death dream and near-death darwinism, Journal of Near-Death Studies, 15(1), Fall 1996.)

1996, DAGUIs Learn How to Learn - Behind closed doors, a DAGUI uses its generic feedback loop not to deliver misinformation to connections within a neural network, but information in the form of intelligent weight updates. This discovery sets the stage for the "Self-Training Artificial Neural Network Object" or "STANNO." This particular kind of artificial neural network greatly differs from the preceding genre in that the ten or so pages of human-conceived training algorithms are replaced by the self-organized and implicit training algorithm spontaneously developed by another neural network. This highly efficient adaptive neural net is intentionally developed as the building block for the DABUI, allowing a DAGUI to bootstrap itself based upon its successes and failures.

1997, A Sweeping Model of Cognition, Creative or Not - Thaler presents his paper, A quantitative model of seminal cognition: the creativity machine paradigm," to the Mind II Conference in Dublin , Ireland . To quote from Thaler's 1997 paper, "… much of human creativity may be attributed to the failure of cortical networks to activate into known memories as these networks perform vector completion upon their own internal disturbances. In lieu of intact memory activation, the networks produce a stream of degraded memories, now constituting what we commonly refer to as "ideas," that are filtered for utility and interest by attendant cortical networks." This paper concludes with the following: "…Placed on the same continuum of perturbation, all cognition may be viewed as acts of creativity. Even at the lowest levels of synaptic disruption one idea is supplanted by another in a display of low-level originality, as in everyday stream of consciousness, conversation, or movement planning. The noblest invention, scientific discovery, or artistic inspiration lies at the opposite extreme of this spectrum, where the Creativity Machine model implicates large localized perturbations as the nucleating events. Within either of these regimes, consciousness itself may be no more than the spontaneous invention of significance by associative cortical networks to the endless noise-driven activations of their brethren…"

1997, The Future of AI and Robotics Memorialized - On the day when the fictional SkyNet supposedly became conscious, Thaler receives a patent for this synthetic thalamo-cortical loop. The official title of this patent is "Device for the Autonomous Generation of Useful Information (DAGUI)." The patent is aptly described as a model of both human creativity and consciousness. (See US Patent 5,659,666). In the preceding three years, DAGUIs have revolutionized the fields of materials discovery, natural language generation/comprehension, control theory/practice, entertainment, cyber-warfare, and personal hygiene products. They have even devised their own patents.

1997, The First DABUI-Based Control System - Thaler builds a highly experimental, spreadsheet-based control system for a chemical reactor. The underlying and then proprietary DABUI cumulatively learns the reactor's behavior and then improvises strategies to cope with any drift from normalcy.

1998, Theoretical Chemistry as a Proving Ground for the DAGUI - Thaler introduces the notion of confabulation-based discovery in materials science. See Predicting ultra-hard binary compounds via cascaded auto- and hetero-associative neural networks, Journal of Alloys and Compounds Volume 279, Issue 1, 4 September 1998, Pages 47-59.)

2001, Connecting the Dots - A now controversial personality working in the intelligence community approaches Thaler to apply a primitive DABUI to the problem of monitoring the Internet for activities of certain personalities in a major Northeastern U. S. city. Early in 2002, he licenses the non-adaptive version of this neural architecture.

2001, AI's Best Bet - NASA's chief visionary, Dr. Dennis Bushnell proclaims the Creativity Machine Paradigm an AI best bet in creating human to trans-human level intelligence in machines.

2001, SkyNet? - Thaler builds a DAGUI-based control system for a constellation of DoD satellites, working through a major military defense contractor. (That's all we can say.)

2002, Homer Can Walk! - A complex hexapod robot learns how to walk in a virtual reality simulation using a DABUI. In order to convey the power of this technique, using a single adaptive neural network as a critic, a large portion of the space containing potential joint positions would need to be explored in order to train that net, requiring on the order of two millennia to acquire the necessary training exemplars. Instead, the DABUI-based cyber-roach mastered not only simple forward gaits, but how to move on demand, in any direction through complex servo sequences within its legs. Later, the cyber-roach develops upright bipedal gaits to attain speedy getaways.  …In effect, this experiment was tantamount to throwing a limp Homer Simpson into virtual reality and the cartoon character autonomously developing standing and walking behavior that are consistent with his structure and the physics of the simulation. …In other words, no artistic license would be required to make him look like he was walking. He would walk through his environment because of his own self-developed leg motion!

2002, Encore Walking Performance in Hardware  - Using a DABUI architecture, a hardware-based insectoid robot learns how to walk without any prior training or programming.

2002, Self Generating Visual and Sensory Pathways for Robots - A DABUI architecture builds itself into a complex, brain-like architecture to perform demanding machine vision applications. The technique is immediately applied to bomb impact analysis for the military.

2003, Will It Dream? - Using a DABUI architecture, a complex, hardware-based hexapod robot learns to walk, and thereafter improvises other complex motions, such as backing up, turning, sidle motion, and bipedal motion. It utilizes preprocessing occurring within its own self-organized synthetic brain pathways to understand the scene in front of it and to improvise approaches for moving through its sensed environment, in this case deeply buried underground facilities. The robot simulation first learns simple strategies such as how to negotiate a doorway or ventilation shaft. Thereafter, it masters climbing stairs, upside-down crawling along antenna masts, and generally devising clever ways of invading a deeply buried facility. Later on in this virtual reality simulation a complex SuperNets self-organize to perform million dimensional sensor integration that in turn feed DABUI architectures to invent the necessary behaviors, on demand. (See Thaler, S. L. (2004). Overall, the IEI researchers develop a technique wherein the robot uses its dreams in virtual reality, during its down time, to rehearse potentially novel scenarios in reality and to thereby devise the necessary behaviors to cope with them. (Creative robots to defeat deeply buried targets, Final report for period July 2003 - April 2004, Contract No. F08630-03-C-0138, April, 2004.)

2003, Borg-Like Neuro-intelligence - A collective swarm intelligence based upon the DABUI architecture develops the capability to efficiently map and destroy deeply buried underground bunkers. Interestingly, mutually self-protective behaviors among robotic swarm members spontaneously emerge. Key to the success of this swarm was a central "Borg-like" self-assembling neuro-intelligence that automatically optimizes the collective behavior of the swarm so that any facility may be explored with minimal overlapping effort, and if necessary, plan the most Machiavellian strategy for incapacitating that facility.

2004, Unexpected, but Welcome Confirmation from Peers Working in Robotics - Helmut Mayer of the University of Salzburg demonstrates the superiority of the self-bootstrapping Creativity Machine over other control schemes in relatively simplistic robotic problems requiring creative adaptation. (See A modular neurocontroller for creative mobile autonomous robots learning by temporal difference, Mayer, H.A., Systems, Man and Cybernetics, 2004 IEEE International Conference on Volume 6, Issue , 10-13 Oct. 2004 Page(s): 5742 - 5747 vol.6.)

2005, How Science Actually Works - A US West Coast professor proclaims the notion of confabulation-based cognition the greatest idea in history! His work is based upon now ancient, pre-computer technology called Bayesian networks and absolutely no attribution is given to Thaler's many previous publications and patents in this area that involve not the archaic Bayesian systems, but truly brain-like and trainable neural networks, multilayer perceptrons.

2006, NASA Can Save Taxpayer Dollars - A DABUI architecture builds and tailors itself in a NASA demonstration to successfully achieve autonomous rendezvous and docking of a levitated space craft simulator. The crowning irony here is that a $50 webcam becomes the only sensor aboard this 1800 pound robot, essentially replacing a $10M laser guidance system.

2006, The True Singularity - A DABUI generates a musical album relying only upon feedback from Thaler's facial expressions. This is a million dimensional problem that is far beyond the capacity of a genetic algorithm, or any other form of AI. …What Thaler is trying to convey to the world is that it doesn't matter whether the problem is creative robots, machine authored music, or management of a constellation of DoD communications satellites, it's all the same problem - novel pattern generation, recognition of the value or utility of such novel patterns, and the reinforcement learning of such novel and utilitarian patterns. Apart from the non-recognition of myriad wet neural networks, it is intrinsically the core plan for all future AI, and hence the so-called "Singularity."

2007, The Dingularity - Thaler coins the name "dingularity" as the pejorative of the well known "singularity" predicted by a popular science writer. The former event, the dingularity, is used in the sense that this author will continue to "ding" the general public through sales of the book that prophesizes this momentous and yet-to-be experienced event. To commemorate this new term, Thaler produces a music video, entitled "The Dingularity," wherein the primary damping mechanism of said singularity is human naivety and inattention to what is to be the crucial development in artificial intelligence, the DABUI. Ironically, the melody is autonomously generated by the DABUI architecture, thus stimulating many future discussions over the immense underlying irony portrayed by the music video.

2007 The Asymmetric Threat from Within - A major Government R&D agency solicits the academic system for self-learning systems that emulate the thalamo-cortical loop of the brain. To quote this dark agency:  "There is growing evidence (e.g., Hawkins, 2004, Granger, 2006) that much of specialized human knowledge and even specialized structures of human brain could actually be constructed from a very small set of thalamic-cortical algorithms, and that the acquisition of these structures is self-directed over a large period of experiential learning to reflect the structure inherent in the world as presented by our senses.  If such a universal algorithm exists, many variants likely also exist, some of which may be much more appropriate for implementation in silicon…"

…But such an algorithm exists, has been vividly portrayed in peer-reviewed journals, and patented many times over! Evidently, Hawkins and Granger have re-originated this idea once again and issued the challenge to build it. Evidently, taxpayer dollars will be wasted on developing a technology that already exists.

2007, Parallel Onboard Intelligence for IEI Robotic AI - In collaboration with AFRL and NASA, IEI was able to parallelize its DAGUIs and STANNOs on graphical processing units using the 128 processors therein. Not only did IEI attain a speedy embedded target for its neural network paradigms, it accelerated processing by at least a factor of ten.

2008, Autonomous Trailering - Earlier this year, IEI developed a totally autonomous mobile robot that could rapidly identify target vehicles amid the typical clutter of landscapes and then hitch and tow these passive vehicles to pre-specified locations. This exercise was carried out with NASA's Surface Mobility Group in preparation for various robotic off-world exercises. The underlying DABUI had to be very clever in originating strategies for physically coupling with the target vehicle which presented itself to the tow robot in a wide range of orientations and elevations.

FAQs

Isn't this just a genetic algorithm?

Genetic algorithms (GA) generate whole populations of concepts or plans of action and then compete them against one another for "robustness." Unless the problem is trivial or toy-like, there inevitably is much human involvement devising what mathematicians call constraint relationships (i.e., the inherent coupling between the genes or attributes of the problem, in the form of explicit equations or logic). As a result, problem dimensionalities are intentionally maintained as small as possible.

But even with dimensionalities as low as 5-10, the GA runs the risk of combinatorial explosion. That's why you read about GAs solving antenna design problems involving bending a wire in five locations, but requiring whole clusters or supercomputers to do so. Such PR sounds impressive, since such sophisticated hardware is involved, but university publicists spin this deficiency into sci-fi like capacities for the unwitting public to be amazed with.

DAGUIs in no way resemble a Darwinian competition. Instead of generating a whole population of 'fish' trying to eat one another, a single such fish continually morphs within the imagitron until the supervisory component, the perceptron, 'likes' what it 'sees.' As this notion transforms, it takes the form of plausible and semi-plausible possibilities that obey the mathematical  constraints that are soaked up by the imagitron in a matter of seconds through training. That is why a DAGUI requires no programming and does not suffer from dimensional restrictions. Instead of solving five dimensional problems, it readily solves multi-million dimensional ones on common PCs. Housed on supercomputers, DAGUIs can solve multi-billion or trillion attribute problems, essentially, the greatest sociological, economic, and political problems that now confront us.

DABUIs take their automation a step beyond DAGUIs. There is no human involvement in conducting off-line training of imagitrons and perceptrons. Instead, DABUIs train themselves, given a data feed to the outside world. Using the foveational mechanisms taught by this patent, they may pick and choose what they tend to study and learn in the external world.

In a sense, the DABUI, as manifest in neurobiology, may reflect a process wherein evolution learned to accelerate (or perhaps destroy) itself through a bootstrapping brain storming session between neural networks of the brain.

Isn't this just simulated annealing?

Simulated annealing is a process wherein very simplistic neural nets having one or two layers relax into memories of what they have previously seen, after being initially 'kicked.' They generate such memories by means of recurrent connections by which a network's outputs are repeatedly routed back to its inputs. In order to select the best memory, a professor or graduate student needs to write a program called an "objective function" that usually takes the form of superficially complex and exotic looking summations. Furthermore, such systems do not take advantage of the latest neural network training techniques devised in the 80s,that allow crucially needed three or more layers to function creatively.

DAGUIs do not generate memories. Internal and/or external disturbances meted out to the imagitron drive subsequent layers of the network to activate into false memories or confabulations. So, there is no relaxation of the network. Perturbations prior to the hidden layers, made possible by modern training algorithms, produce new classes of things, the network has already been exposed to. Further, perturbations within the output layers soften or break the cumulatively known rules absorbed by the net. New, derivative classes of things, and rule violation, are necessary conditions for new discoveries and creations.

DAGUIs, as well as their newest incarnation, DABUIs, do not require recurrent nets to function. The only necessary feedback is the perceptron's ability to fine tune the disturbances driving idea formation in the imagitron.

Finally, simulated annealing schemes do not involve tandem neural networks. That quality is unique to the DAGUI/DABUI family of creative neural architectures. (Try to find a text book, paper or patent that incorporates the fundamental notion of at least two artificial neural networks engaged in a brainstorming session with one another.)

Isn't this just a West Coast Professor's so-called "confabulation" theory?

In short, what this professor calls "confabulation" is a process used to select the expectancy of a concept that follows a particular context. This new lexicon has nothing to do with Thaler's more than 30 year old theory in which false memories (i.e., confabulations) generated by some neural networks in the brain are perceived valuable by other neural nets.

The very notion that a theory of any kind needs to be developed to judge expectancy of a concept indicates the lack of autonomy in the underlying system and the need for human programming. In this professor's system, centuries old Bayesian networks (not neural networks) are being used. The more modern neural networks spontaneously develop their own expectancy of a concept, through the well known process called "vector completion," whereby our "wet neural nets" fill in the blank in the sentence "Mary had a little _______."

This theory-less process of vector completion is part of the immense success of both the DAGUI and DABUI paradigms, since critic neural networks automatically carry out the juxtapositional inventions called inductive and deductive logic/discovery.

This academic then uses his misnamed theory to account for the entire gamut of cognition, as Thaler already has using self-organizing neural, and not human contrived Bayesian networks (See above key events leading to the DABUI).

Doesn't another sci ence author describe this process in his book about intelligence?

This author sets up the importance of thalamo-cortical algorithms, effectively issues the challenge to build them, and then coincidentally forms a company to produce them from ancient and human-contrived Bayesian networks. The agony is that such thalamo-cortical algorithms have already been patented for more than a decade (DAGUIs); numerous refereed papers have been written about them; and one may view products and services conceived by them on national television. …Isn't it amazing what money and its accompanying influence can buy?

But It Requires Information to Run!

If there is no information, then there is no need for intelligence of any kind.

Further, there is no basic, human-conceived rule that generates all subsequent knowledge

Additional Resources Related to DABUI

Crying in Vacuum | In Its Image | Song of the Neurons