Top Banner
“TCP/IP/IQ” Why Can’t the Internet Become a True World Brain? Special Session SS-3 4:20-5:50 P.M. October 19, 2001 Stephen Thaler, Ph.D., CEO & President, Imagination Engines, Inc. Expanded Speaker’s Notes ________________________________ Registered Trademarks: Imagination Engines Creativity Machine DataBots
93

Conscious Internet

Dec 18, 2014

Download

Technology

Vapula

One of IEI's main themes is that it is single-handedly building the required artificial intelligence technology to emulate and perhaps exceed the capabilities of the human brain. Thus all of the key cortical functions recognized by brain scientists, perception, learning, and internal imagery (a.k.a., imagination) have been developed at the IEI laboratory purely using artificial neural networks to emulate the biological neural networks of the human brain. Although the World Brain is not yet a reality, it is a long range goal of IEI to create a free-thinking entity, distributed across the Internet, that introspects upon human-originated content and then creates its own seminal thought and discoveries. For those of you who have read and understood IEI's web pages, you will likely realize that there is no other form of artificial intelligence capable of self-learning and creativity that can compare with IEI's patented technologies. Therefore, it is our ambition to perfect the network (i.e., WAN) mechanics for distributing its neural systems across the largest computational platform currently available, the Internet, and allowing them to knit themselves into a coherent brain using IEI's Supernet principles. If you are an investor, government, or corporation interested in supporting this effort, please contact Dr. Stephen Thaler at Imagination Engines, Incorporated.

http://www.imagination-engines.com/
http://slthaler.wordpress.com/
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Conscious Internet

“TCP/IP/IQ”

Why Can’t the Internet Become a True World Brain?

Special Session SS-3 4:20-5:50 P.M. October 19, 2001

Stephen Thaler, Ph.D.,

CEO & President, Imagination Engines, Inc.

Expanded Speaker’s Notes

________________________________

Registered Trademarks:

Imagination Engines

Creativity Machine

DataBots

Page 2: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 2

DEFINITIONS: Artificial Neural Network – a collection of switches, real or simulated, that effectively wire themselves together so as to autonomously write a parallel computer program. TCP/IP – Transmission-Control Protocol/Internet Protocol, a suite of computer networking formats and procedures that enables dissimilar machines to communicate with one another. IQ – that quality of mind purported by psychologists to measure intelligence.

Page 3: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 3

SLIDE1: INTRODUCTION

Let me ask you a very important question: When I use the term “World Brain,” what images come to mind? Do you immediately think of the Internet? After all, the Internet grows in size and complexity on a daily basis. We bless it, we curse it, yet can you, by any stretch of imagination envision the Internet one day spontaneously becoming cunningly creative, or self-aware? I don’t think so, and I say that from the perspective of an AI practitioner and as a scientist. Unfortunately, there will always be a Star Trek mythology in which computers are able to gather enough knowledge, materials, and resources,

Page 4: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 4

until they magically become self-aware. Personally, I can’t see how such a scenario can occur, but I’ll try to be tolerant. When I say “World Brain,” do you begin to think of the traditional schools of artificial intelligence, expert systems, fuzzy logic, genetic algorithms, etc? If you do, then ask yourself, when was the last time you saw a software application that demonstrated anywhere near human level perception, learning, and creativity. I wager that you haven’t. …So your image of the World Brain must involve some previously undiscovered AI technology! When I use the term “World Brain,” do you envision an apocalyptic catastrophe in which an evil machine intelligence overpowers the planet and destroys the whole of humanity? I would have to take issue with you at that point, since a close examination of what makes man combat man involves resources: food, shelter, companionship, and ideological comforts. I have no evidence that a machine intelligence will need anything of value to humans. However, what I can imagine is an even more frightening scenario in which machines grant us exactly what we crave! In the next hour, I intend to share with you my vision of a “World Brain,” one that is capable of unlimited wisdom, creativity, and a self-awareness. The World Brain that I speak of is based upon a radically new form of artificial intelligence that I have developed in a piecemeal fashion over the last 26 years.

Page 5: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 5

SLIDE 2: IEI PATENTS = HUMAN COGNITION

The technology I’m speaking of takes the form of over a dozen very fundamental international patents in the field of artificial neural networks. As you may already know, the human brain is composed of neural networks, and the artificial variety that we implement on digital computers, capture the essence of how these biological networks function. Allow me to expand briefly on what the ‘wet neural networks’ of the human brain do. Within the field of cognitive neuroscience, three primary functions of the brain are acknowledged:

Page 6: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 6

(1) Perception – This is the cortical process wherein things and activities in the environment are associated internally with other previously stored memories. Therefore, a photograph of some friend or relative, for instance, can stimulate related thoughts about their voice or some joke they’ve recently related to you. (2) Learning - This cortical process enables the formation of memories, of both things and activities experienced within the external environment, or of ideas internally generated within the brain. (3) Internal Imagery - Perhaps most profound aspect of human cognition is the internal genesis of new ideas and plans of action. Known generally as internal imagery, this process corresponds to recollecting a friend’s face without the aid of a photograph, deciding where to run next when confronted with danger, or, in a more relaxed moment, writing a letter to a friend. It is this latter function of internal imagery, totally unexplored territory in the field of artificial neural networks, that the IEI neural network technology excels at. Quite amazingly, we may utilize such creative neural systems to in turn refine other more trafficked areas within the field of artificial neural networks, whereby processes of artificial perception and learning are vastly improved. At this point, we have all the tools necessary to build a synthetic brain that is capable of all aspects of human cognition. (…and it is my claim that all of these fundamental brain activities span the entire gamut of human cortical function) To make my case clear, let’s examine what exactly brain does and then contrast that function with what the patented IEI technology achieves.

Page 7: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 7

SLIDE 3: WHAT IS THE BRAIN?

Here is the human brain, just a few pounds of protoplasm and roughly 100 billion individual cells called neurons. Removal of other pieces of anatomy or organs, does not result in cognitive impairment, however removal or even injury of the brain can have a profound effect upon sentience. For this reason, neuroscientists are in absolute agreement that this is the organ within which all aspects of thought occur. (i.e., There is no scientific evidence of something immaterial or even supernatural taking over once organic damage has been done.) All that we are is somehow embedded in this protoplasmic mass. That’s why Nobel Laureate Francis Crick was compelled to write the book “Amazing Hypothesis” wherein he claims that all we are, all

Page 8: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 8

we think, all that we feel is the cumulative activity of a pack of neurons. However, Crick does not convincingly cover the gamut of things that this pack of brain cells can achieve. In fact, he generally describes the mechanics of brain in terms of learning and perception, but not in the least with regard to imagination and creativity. However, I am generally in agreement with him, that we need not look beyond what we see when we pop open a skull and look inside, to explain human cognition. If Crick offers the amazing hypothesis, then let me volunteer the “shocking conjecture” that the brain is nothing more than a protoplasm-based model of our external world and ourselves. After all, to survive and flourish, we need to anticipate the world around us, as well as our reaction to it. To do that, we require an interactive model that can anticipate danger, as well as opportunity. Everything else one may think of as a valid function of brain is peripheral to this primary role of world modeling. Later, we continue with this shocking conjecture, to examine the lowly mechanisms that produce our most profound thought.

Page 9: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 9

SLIDE 4: THE UNIVERSE: MYRIAD INTERACTING ENTITIES

What exactly is the universe? The broadest definition I can think of is that of myriad interacting entities (and I mean entities in most general sense of the word). Such entities may span the gamut of things that we call ‘inorganic’, from subatomic particles to planets, stars, and galaxies. The term entities may pertain to biological creatures such as plants, animals, and people. The word may likewise convey the notion of institutions created by human beings. At this point, we don’t care about the essential nature of entities, or how we intend to categorize them. The universe simply consists of entities…

Page 10: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 10

SLIDE 5: THE UNIVERSE: MYRIAD INTERACTING ENTITIES

No matter how one divides the world into classes of entities, we inevitably agree that all things within the universe are interacting and that such interactions may span the inorganic, biological, and social worlds. The irony is that no matter how we come to describe or scientifically regard each of these entities, they will in myriad ways be connected with all else. The science and philosophy of caring about connections, and not the intrinsic nature of things, is called connectionism and may perhaps become the most enduring field of human intellectual endeavor.

Page 11: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 11

SLIDE 6: BRAIN BUILDING MODEL OF UNIVERSE

Speaking in very general terms, the human brain begins its life as an excess of many isolated neurons encased within the human skull, which I have symbolized by the purple box on the right. Connecting the universe to the brain, is a sensory layer (i.e., the five senses), that I have symbolically represented with the eye. Through this layer, photons, acoustic waves, molecules, and contact pressure link us to the world around us.

Page 12: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 12

SLIDE 7: BRAIN BUILDING MODEL OF UNIVERSE

As perception and learning begin, small colonies of neurons begin to connect themselves into regions that have a close relationship with entities in the surrounding environment. Later, when the corresponding entities are sensed in the immediate surroundings, these small neural colonies resonate with activity. (Important to note here is that such colonies are highly distributed within the brain, looking more like swirls within marble, and definitely not taking on the highly modular nature depicted within the slide). Thus the image of mother selectively activates a highly distributed neural colony that corresponds to mother. At this stage, the brain has begun to form token representations of entities in the external world. That they

© 2001 Imagination Engines, Inc. ... the future of all technology

BRAIN BUILDING MODEL OF UNIVERSE

SENSORY

LAYERtoken entity or

event B

token entity orevent A

Page 13: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 13

seem so relevant to us is the fact that they are embedded within the sum total of other neurons, that collectively become habituated to these tokens as reality. Remember that in actuality, these token entities are nothing more than tiny ‘chirping bags of water’. Note the foundation for illusion: The brain is convinced of the reality of these token entities, but that state has been arrived at through pure repetition and trauma, a kind of “inevitable and natural brain washing.” For general discussion purposes, I have called out generic token entities A and B that have knitted themselves out of originally non-interacting neurons…

Page 14: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 14

SLIDE 8: BRAIN BUILDING MODEL OF UNIVERSE

Concurrent with the process of self-organizing into distinct token entities, these neural colonies begin to connect themselves in proportion to how much they are observed in juxtaposition within the outside world. This activity is achieved by allowing neural colonies to connect with each other when they chirp in unison, and to disconnect when they do not. Therefore, if mother is always present when there is nutrition, then the ‘mother token neural colony’ connects with that representing the act of ‘eating’. The nature of this connection, as we are about to see, is strictly electrochemical in nature and in no way reflects the actual

Page 15: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 15

nurturing involved. However, later in the process of brain wiring itself, a neural colony representing the ‘act of nurturing’ may append itself to the mapping between mother and nutrition. Note again, that there is no inherent reality to these connections. …The illusion gains complexity. (Oh what tangled webs our brains weave when they attempt to perceive!)

Page 16: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 16

SLIDE 9: EMERGENCE OF CONSCIOUSNESS

When neurons have not been recruited to model the outside world, what do they do? Without a direct window to the world, these idle neurons build a model of what other neurons are doing. In a strictly tongue-in-cheek sense, these unemployed neurons are like the town gossips. With nothing better to do, they spend their days spinning tales about others gainfully employed! Of course, in the process they build meta-knowledge (i.e., information about information). They also spontaneously build perceptions about what is going on in other parts of the brain, perhaps incorporating the token entities and relationships already built up in the preexisting cortical

Page 17: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 17

neural networks. The result is that this perception created by idle neurons, is built upon the tenuous models already habituated in the brain’s neural networks. The result is that we all interpret overall cognitive turnover (i.e., chirping neurons) based upon well-habituated world models. This observation helps to model the diversity of opinions about the very nature of consciousness. In short, the brain spontaneously creates a lore about itself! This may be a beneficial illusion, but let’s face it, the process is akin to two optimists on a sinking boat who have deluded each other that help is on the way. Nevertheless, they at least die with a positive mental attitude. …By the way, if in the course of Darwinian evolution, these spare neurons have a favorable perception about their surrounding cortical activity, then the host organism avoids walking off of cliffs or foolishly confronting its predators. These protoplasmic systems are extant, and their not so proud cousins extinct. Furthermore, if one tries to convince such a system that it is laboring under an illusion, good luck. …We thus attain another data point in understanding consciousness!

Page 18: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 18

SLIDE 10: BUILDING A SYNTHETIC BRAIN

Is the building of a synthetic brain all that difficult? Ironically, to some, the necessary technology is already here. Just watch the latest science fiction movies where semi-scientifically literate writers skip the hard details and present machines that do nearly all that we can. (Sarcastically) All we need to do is build our present day computers bigger and faster, and human level cognition will spontaneously arise! Right….No, dead wrong!

Page 19: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 19

The recipe for building a synthetic brain hasn’t been spelled out, but it isn’t that complicated from a theoretical point of view. Here are the essential ingredients: 1. The brain should be neural network based, since the brain

is so constructed. (For the mathematicians out there, this fact arises from the multilayer perceptron being the most general fitting function possible, essentially a function of a function, of a function, rather than the usual statistical fits that look like a sum of basis functions such as sine waves or polynomial terms.) Herein lies the ability to model complex causal chains (i.e., the universe) where something happens because something else has happened, etc.

2. Learning may be easily implemented using any number of

existing neural network training paradigms. 3. Consciousness, since it is most likely based upon illusion,

may likewise be implemented within artificial neural networks wherein some neural networks are watching other neural networks and, mistakenly or not, perceiving consciousness therein.

At this point there is only one key ingredient missing, the ability to generate new thoughts and plans of action. This is the remaining piece of the puzzle, something that I will be addressing in the next few slides.

Page 20: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 20

SLIDE 11: CONNECTIONISM

Before proceeding, though, we must remember that the connections between token entities in the brain are themselves token representations of correlations and causations in the external world. Therefore in building the world simulation that is brain, we are free to use any manner of connection among many alternative kinds of computational switches. In the brain, for instance, the blocking and unblocking of post-synaptic terminals within the chemical synapse, emulates the degree of connection between token entities. This process can generally emulate everything from fundamental processes in physics to sociological interactions among people.

Page 21: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 21

Similarly, token representations may be built using synthetic neurons, whether they be of electrical, optical, or even a mathematical nature. Conceivably, we could build a neural network out of rubber band connectors and mechanical switches!

Page 22: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 22

SLIDE 12: MULTILAYER PERCEPTRON

The synthetic neuron forms the basis of the artificial neural network. It is simply a threshold switch that accumulates signals from other synthetic neurons just like it. This net input is then compared against some internally stored threshold value. If that net input is exceeded, the neuron activates to produce an output signal of 1. Otherwise, the synthetic neuron is silent, outputting a value of 0. Of course the real neuron is more complex in its behavior, but its information storing essence is captured by the synthetic neuron through: (1) synaptic integration and (2) threshold firing behavior. A lot of the extra behavior of the neuron is involved in attaining this simplistic behavior using

Page 23: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 23

more complex biological mechanisms and in supporting its own metabolism. In actuality, there is very little information content to the neuron itself. It is simply an on-off switch. The actual intelligence of the neural network is stored within the various connections that ‘weight’ the various inputs arriving at a particular neuron, to provide the net input discussed above. (Therefore, be very wary when someone tells you that artificial neural networks can’t attain human level cognition simply because the artificial neuron isn’t as complex as its biological inspiration!) If there are more than two layers of neurons involved, the neural network is called a multilayer perceptron. Input patterns begin at the top layer and then propagate through the hidden layer(s) to the bottom, output layer. At this output layer, some output pattern is produced. Roughly speaking, the network can learn to associate some input pattern with some output pattern (i.e., perception). As the complexity of the relationship between inputs and outputs grow, we generally need increasing numbers of hidden layer neurons.

Page 24: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 24

SLIDE 13: TRAINING MULTILAYER PERCEPTRON

In training the multilayer perceptron, we are constantly calculating how far the actual output patterns are from those desired. These so-called delta errors are then propagated in the reverse direction, through a series of partial differential equations developed in the mid 80’s and coined backpropagation. (If it helps, think of these backpropagation events as ‘mathematical spankings.’) Cumulatively, after enough feed-forward and backpropagation cycles, the network cumulatively learns to associate one pattern with another. Of course myriad such associations may be stored within the same multilayer perceptron.

Page 25: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 25

Note that after a few feedforward and backpropagation cycles, the distribution of weights within the network begin to ‘walk’ from their central values and we begin to see a Gaussian spread in the weight frequency histogram. Also note at this point, that any pattern may be mapped to any other after a sufficient number of backpropagation cycles. This dimension of arbitrariness drives home the name “perceptron,” probably a corollary to the old saying that opinions are a dime a dozen.

Page 26: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 26

SLIDE 14: TRAINED MULTILAYER PERCEPTRON

Through training, the network has been effectively forced to discover some internal logic to correctly associate output with input patterns. Upon close examination of the connections between processing units, we find that ‘logic circuits’ have spontaneously grown, capturing all the ‘ifs’, ‘thens’, and ‘therefores’ required to model these associations between input and output patterns! Also to our amazement, the hidden layer(s) have self-organized so as to form token entities. Therefore the presentation of some pattern to the net, say the pixel pattern of a face, causes spontaneously formed neural colonies

Page 27: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 27

within the hidden layer, corresponding to eyes, ears, nose, and mouth, to resonate with activity. In the final layer(s) of the perceptron we find that connections form between these token entities and the output units so as to reveal how these entities must be associated to produce the necessary result. If, for instance, our network is intended to decide if the image of a face is present in some scene, it may utilize these individual feature detectors and build the required logic, in the weight output layer, to test whether all the required features are present. In this way, the perceptron registers the presence of a human face. Does this all sound familiar? It should. Further, it all happens automatically, without human intervention.

Page 28: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 28

SLIDE 15: SYMBOLIC AI VS NEURAL NETWORKS

Now that we’ve talked about the brain and its relationship to artificial neural networks, let’s take a brief survey of what the state of art is in the field of artificial intelligence (AI). The large majority of those in the field of AI believe that human programmers must be involved in the business of building intelligent machines. The recipe they follow is this: (1) write an over-glorified script (i.e., a computer program) that embodies how a conceptual space works, a code that embodies all the “if-then” thinking typically employed in the decision making of humans or other kinds of systems, (2) if the rules change, then have the human return and modify the code, and (3, sarcastically) even though the programmer

Page 29: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 29

is fully aware of all possible outcomes of this code, squint at the results and call it ‘creative.’ In short, the heuristically based AI expert is very much like a medieval scribe. But what happens if that scribe doesn’t understand the underlying logic behind the knowledge domain he is trying to embed within a computer program? More importantly, what happens if there are no heuristic programmers to write code? It almost seems as though the fantasy of how we think has been taken too literally and extended well beyond its capacities! On the other hand, artificial neural networks are to heuristic AI, as a ‘HAL9000’ is to the medieval scribe. They essentially build their own rules to explain raw data, they continuously learn, and with the addition of some new technology, may be stimulated to become creative.

Page 30: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 30

SLIDE 16: VIRTUAL INPUT EFFECT

The secret to inducing artificial neural networks to become creative is based upon a very fundamental phenomenon that I observed in the mid 70’s and then expanded upon in the early 90’s. (I find this to be the most interesting physical and mathematical effect I have ever seen. I am utterly convinced that this effect will have immense repercussions in the fields of both science and philosophy.) Imagine an experiment in which we train an artificial neural network to predict what artist has most likely produced a particular work of art. In creating such a net, we would gather images of both the art and the respective artists, and train it to correctly associate between them. With prudent

Page 31: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 31

training, such a net could very accurately view some painting it has never seen such as “Starry Night” and correctly predict Van Gogh as the artist. Note that this is the standard use of artificial neural networks, as pattern associators. To be used properly, the neural network must be supplied some input pattern. In general this process corresponds to the act of perception in the human brain, as we have already discussed. Now, let’s do something totally at odds with what the practitioners of neural networks do. Let’s apply no inputs whatsoever to this pre-trained network. Then, let’s randomly select some connection weight and administer some slight perturbation to it. (Remember that the weights involved assume algebraic values and that we may add or subtract small values to these weights to disrupt them.) To our astonishment, the network outputs produce some intact image, say Van Gogh’s, without the input of any external stimulus. In effect, the network is falsely perceiving some environmental pattern when in fact no such stimulus is present! …Amazing! Accordingly, I have named this phenomenon, of a neural network imagining some output, without an external stimulus, the “virtual input effect.” Even more amazing is the observation that when we begin to further perturb this network, either increasing the magnitude of the original synaptic perturbation, or introducing more perturbations of the same magnitude, intact memories of artists no longer appear at the network’s outputs. Instead, we see juxtapositions (as well as extrapolations) of the faces of these artists, perhaps now producing a hybrid between Picasso and Van Gogh, then

Page 32: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 32

one of Dali and Renoir, then a pure variation or distortion of Vermeer. In fact as we allow these disturbances to hop among the connection weights of the network, we see an endless progression of images that definitely qualify as faces, yet do not resemble any face previously shown to the net during its training. In effect, this network is serving as an invention machine for new potential faces. The fact that each of these new candidate images appears face-like reflects the fact that despite the internal synaptic disturbances, most of the implicit constraints, as to what constitutes a face, are preserved (i.e., the connection weights are largely preserved). However, there is enough change within the system of synaptic connections so as to nucleate a non-memory, and hence a potential idea.

Page 33: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 33

SLIDE 17: VIRTUAL INPUT EFFECT

If we experimentally discern and then plot the probability of producing an intact memory versus the average synaptic damage imposed upon the network, we see very reproducible behavior. Up until approximately 6% synaptic damage, the probability of nucleating a memory remains constant. Thereafter, the likelihood of activating a pure memory dramatically falls off. We may repeat the above experiment on any manner of neural network, no matter what its size, complexity, or topology, and the identical result is seen. Intensive investigation of this effect shows that it is the regime near 6% where non-memories are produced, notions

Page 34: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 34

that are reminiscent of the patterns that have been previously shown to the network, but not exact replicas. Well below 6% perturbation, the net strictly produces memories, things that it already ‘knows’. Well above, 6% perturbation, the network produces pure nonsense, since the many logic circuits constituting the network have been destroyed. …Clearly, the most fertile region for stimulating a trained neural network to produce new and useful patterns is within the region of 6% mean synaptic perturbation. Note 1: Near 0% perturbation, there are very few new ideas generated. This is the so-called “Neo-Lamarckian” regime, characteristic of programmers laying out decision trees (i.e., expert systems) wherein very little creativity is manifest. Note 2: At high perturbation levels much greater than 6%, rarely do coherent patterns emerge that satisfy the implicit constraints of the conceptual space originally absorbed within the network. This is the so-called “Neo-Darwinian” or “Blind Watchmaker” regime that is dominated by nonsensical, random patterns. This is where genetic algorithms operate, in a very inefficient and computational expensive mode. Note 3: Using the imagination engine embedded within this slide, I can move the slider and adjust the level of mean synaptic perturbation. Near 0%, we see the network randomly activating into the facial memories it has absorbed during training. At 6% perturbation, we see the network experimenting with new potential faces. Well above 6% perturbation, the neural network outputs nonsense. Note 4: What prevents this effect from occurring within the human cortex, where many synaptic perturbations, and their circuit equivalents are constantly occurring (i.e., diffusing

Page 35: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 35

neuromodulators, leakage of neurotransmitters, cell membrane potential fluctuations, etc.)? In fact, to halt this effect, one needs to discover some intelligent damping process wherein each noise source in the brain is automatically muted! Otherwise, a process I have coined “internal vector completion” inevitably induces a memory or a corrupted memory that constitutes an ‘idea’.

Page 36: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 36

SLIDES 18-20: MULTILAYER IMAGITRON

Page 37: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 37

To help you imagine what is happening in the network during this synaptic perturbation process, we show a progression in which the synaptic disturbances are randomly hopping among the connection weights of the network. Best results are obtained by selectively perturbing the input layer of weights. Also, in the course of perturbing the connection weights, we see that the Gaussian distribution of weights is preserved, apart from minor excursions in the frequency histogram. Note that if the unperturbed network is a perceptron, then the internally perturbed network should be called an “imagitron.”

Page 38: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 38

SLIDE 21: THE AMAZING IMAGINATION ENGINE

What I have just described is a watershed discovery, and the foundation for simply and elegantly building virtual machines that invent new ideas and plans of action. I call these systems, as well as my company, “imagination engines.” All that we need to do is expose an artificial neural network to many patterns representative of some conceptual space or knowledge domain, and then bathe its synaptic connections with hopping perturbations having an average fractional perturbations of near 0.06. Then, spontaneously, out pop novel patterns that are reminiscent of the original conceptual space, and potentially qualifying as useful ideas or strategies within that space!

Page 39: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 39

In beginning to apply this technique toward practical ends, note that we may follow a bootstrapping routine wherein good ideas that we see emerging from a preliminary imagination engine are captured and in turn used to train the network again. Over repeated cycles of this process, the imagination engine becomes progressively more adept at producing optimal notions.

Page 40: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 40

SLIDE 22: THE CREATIVITY MACHINE PARADIGM

Rather than task a human operator to monitor the emergent ideas from the imagination engine, one may add a computational critic to the system that is on the lookout for any output patterns that would qualify as a useful idea. Therefore, we may either mate to the imagination engine some heuristically based computer program, or another neural network that has been trained by example. If we ‘box- car” these two systems together we arrive at a neural network based discovery system that requires no explicit knowledge about its world, as would be supplied by humans, to create new and derivative knowledge.

Page 41: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 41

Note that we have harnessed a new scientific phenomenon (i.e., virtual input effect) with a policing neural network (the alert associative center) to produce a new invention, perhaps the grandfather of all subsequent inventions! To help you better understand the process behind the Creativity Machine Paradigm, imagine that we would like to spontaneously compose new musical melodies. Training of the imagination engine would involve showing the network many examples of top ten melodies over the last 30 years, for instance. The alert associative center would be shown the optimal position that these songs took in the musical polls. Because the imagination engine has absorbed the ‘zen’ of what constitutes a good melody, it tends to output only candidate top ten melodies when synaptically perturbed. The critic net, that implicitly understands how to rank these emerging melodies, can be used as a filter to capture only the very best of these candidate songs. This whole methodology may be repeated for any conceptual space imaginable, since the world consists of, and may be described, by numerical patterns. We are now at the dawn of a whole new era in thought. No longer will the learned debate among themselves the nature of things, or devise detailed logic and procedures for accomplishing tasks. Instead we will simply bring together two or more trained neural networks in a dialog and then let them summarize their findings to us! This changes everything!

Page 42: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 42

SLIDE 23: CREATIVITY MACHINES DO IT ALL!

I could go on for many hours, but here are some AVI captures of just a few Creativity Machine projects. You’ve already seen the facial creativity machine. More advanced forms of this machine may assist in producing portraits of crime suspects, based upon the response of witnesses and victims, with the police artist removed from the loop. Creativity Machines may write their own sequential, heuristically based algorithms as we see in this data compression demo. The notion of neural networks writing

Page 43: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 43

computer code should come as no surprise. Computer programmers do this all the time using wet neural nets. Creativity Machines may capture the essence of what are known as dynamical systems (i.e., systems that evolve in time). For instance, just through a brief exposure to a few isolated poses of a human model, the Creativity Machine is able to spontaneously invent new and realistic movement scenarios. We could, for instance, allow the Creativity Machine to imagine a thousand potential scenarios, beginning from some starting position, to calculate the odds that a human figure could arrive at some predetermined position. In general Creativity Machines answer the question, “Can we get there from here?” Shown in this slide is a foray into the invention of personal hygiene products, wherein a Creativity Machine is creating new toothbrush designs. Here, we see the invention of a very popular toothbrush that is seen nightly on network television in the US. Materials Discovery Creativity Machines have already been built for the US Air Force to discover heretofore unknown and valuable chemical compounds. Here a Creativity Machine is discovering new ultra-hard compounds having only two elements. Creativity Machines may very elegantly and simply embark upon rather mundane tasks, such as where to look next. In this slide, an autonomous targeting system prescribes an optimal path for ‘painting’ enemy planes in a dogfight. Finally, Creativity Machines may be used to autonomously classify things into natural families at rates and efficiencies far surpassing any other known classification techniques.

Page 44: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 44

Currently IEI has been awarded a contract by a major aerospace corporation to harness a Creativity Machine to perform beam planning for a constellation of communication satellites. Certainly, this complex scheduling problem will play a major role in the coming World Brain.

Page 45: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 45

SLIDE 24: THE INVERSE CREATIVITY MACHINE, AKA THE SELF-TRAINING ARTIFICIAL NEURAL NETWORK OBJECT (STANNO)

Note that so far, the Creativity Machine is not completely autonomous. Both of the neural networks involved had to be taken offline and trained on the necessary data. Subsequently, the networks are reinstalled in the Creativity Machine architecture, where they can now invent and create. Wouldn’t it be convenient if both neural networks of the Creativity Machine could be trained in situ, without removing them from the system? Ironically, a Creativity Machine has already invented such a scheme. …However, that story I’m saving for another time and place.

Page 46: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 46

Nevertheless, if we replace the trained neural network of the Creativity Machine with an untrained one and allow the critic net to add corrections, rather than perturbations, to the connections of the former net, we arrive at a self-training neural network. In fact, by absorbing the lower network into the upper net, we produce a monolithic neural network that simply requires data patterns to train. No external training algorithm is needed. Furthermore, we may convert this self-trainer into an object-oriented programming (OOP) class template so that we may now, in cookie-cutter fashion, instantiate millions of autonomously learning copies of the original. Such self-learning swarms may invade databases and exhaust all possible discoveries within them. Most importantly, if we make the static neural networks of the Creativity Machine STANNOs, we now have a totally autonomous system that may combine activities of perception, learning, and creativity all into one package. Such a system has a free will in the strictest sense of the word!

Page 47: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 47

SLIDE 25: SELF-TRAINING ANN OBJECT

Peripheral uses of the STANNO include the most advanced form of computer network intrusion detection available. Here a STANNO is spontaneously building a complex model of what constitutes normal local area network traffic. The STANNO then automatically senses packets that could represent malicious intentions or various system pathologies. Once aroused by a suspicious event, the STANNO may trigger a Creativity Machine to begin testing hypotheses regarding the source of the anomaly, or initiate deceptions or aggressive countermeasures against a potential intruder.

Page 48: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 48

SLIDE 26: SELF-TRAINING ANN OBJECT

Another exciting use of the STANNO is to bring to the layman the power of artificial neural networks without the requisite Ph.D. level knowledge of neural networks typically required to use them. In other words, this “Poor Man’s Neural Network” may be used to predict the winner of the next horse race or anticipate a boss’ or spouse’s next move. Furthermore, the STANNO can take itself apart to show how it thinks, thus revealing the critical factors involved and the heuristics that have been learned by the net. Note that neural networks may no longer be regarded as ‘black boxes’, a rather regrettable reputation that has been hung on them by their critics.

Page 49: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 49

SLIDE 27: STANNOS MADE OF STANNOS

Because of their independence from external training algorithms, compound STANNOs may be built (i.e., STANNOs within STANNOs) wherein all component neural nets train in situ. Alternatively, if I were to build a compound neural network using conventional, static neural networks, I would need to sequentially remove each, train and then reinsert them within the overall network cascade. Using this new technology, each constituent neural network trains in position, within the overall neural architecture. Now let’s look at the utility of compound STANNO cascades:

Page 50: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 50

First of all, if each component STANNO comes to represent the behavioral model of some real hardware component, say an electrical device such as a transistor, a capacitor, or diode, then we may train the overall neural network to perform some complex input-output function to simulate some electrical system (i.e., a cigarette lighter, an FM radio, or a digital computer). Then, effectively the connection weights reveal how to connect these components to produce the overall device. Further, if these STANNOs are implemented in actual hardware, then devices can adapt themselves, in real time, to perform a variety of hardware functions. Secondly, and more importantly, if we train each component STANNO to simulate some fundamental analogy base, then the overall STANNO cascade will connect basic analogies to devise a theory about the raw data it sees. In other words, these neural networks can serve as theoreticians and hypothesis testers. Rather than rely upon the generic on-off switches that the computational neuron represents, the basic building blocks of this network are discernable realms of thought, and the connections made between them reveal a semantic network!

Page 51: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 51

SLIDE 28: NATURAL LANGUAGE PROCESSING

The compound STANNO comes in extremely handy when we equip machines to understand natural language. There the underlying component STANNOs represent various linguistic conceptual spaces, different parts of speech, and semantic spaces. When used as a Creativity Machine, such a compound STANNO may test various alternative word usages, within the constraints of proper grammar and the overall context of a document to disambiguate sentences and passages. Here, a compound STANNO, consisting of over 300 individual STANNO modules, is searching the Internet for references to public enemy number 1. Note that content is

Page 52: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 52

automatically sorted into predetermined classes and that text summarization takes place through the totally automated construction of semantic networks. Important to note here is that there are no lookup databases or words or phrases, no templates, and no explicit “if-then” rules. …Again, it is using nature’s most flexible and adept fitting function, the multilayer perceptron.

Page 53: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 53

SLIDE 29: NATURAL LANGUAGE GENERATION

Human speech, after all, is a minor act of creativity, that is naturally emulated via Creativity Machine Paradigm. In very simple terms, an imagination engine dreams up potential things to say, in the context of a conversation, while the critic network, the alert associative center, evaluates which response would be most appropriate, in light of the system’s overall objectives. Shown here is what I call the “International Expirer,” essentially a self-writing tabloid, driven by an underlying Creativity Machine. The imagination engine has been trained by exposure to three months worth of tabloid headlines, while the critic net has been trained by my evaluation of the

Page 54: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 54

comic content of these headlines. Note that the imagination engine has automatically captured the implicit grammar of tabloid headlines and the alert associative center has developed a figure-of-merit (FOM) model of my sense of humor. Combine the two networks and we may spontaneously generate new and potentially funny tabloid headlines (at least for me). Note that even with this Creativity Machine, we may ‘pin’ its inputs within some context, say “Peewee Herman” and the network will produce a tabloid headline related to Peewee. Effectively, we may ask who is Peewee and this Creativity Machine will respond with some fact about that personality. Essentially, such a device may be considered a “Turing Baby” along the lines of the famous criterion, originated by the mathematician Alan Turing, to test for human level intelligence in computers.

Page 55: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 55

SLIDE 30: NATURAL LANGUAGE GENERATION

Here is the result of recent experiments in which immense imagination engines are created by launching thousands of STANNOs that begin to learn facts about some pre-selected microcosm. When induced to dream via the virtual input effect, they spontaneously link together previously unrelated facts into associative chains and loops. In the demo shown, we are viewing the internal stream of consciousness of this experimental system, which is essentially dreaming thoughts about its information environment (a toy biblical space). Again, there is no explicit text allowed in this system. We are observing the highly

Page 56: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 56

encrypted conversation of neurons through real time decryption! The introduction of some external stimulus, such as the statement “apple is red” or the question “what is apple” serves as an ‘I/O interrupt’ to the system, after which a whole new associative chain is nucleated. In other words we have induced a whole new train of thought in this synthetic brain!

Page 57: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 57

SLIDE 31: CLIENT-SERVER STANNOS

Crucial to the envisioned World Brain effort is the need to distribute both STANNOs and Creativity Machines across the Internet. Examples of so-called “disembodied” Creativity Machines are actually utilized on the IEI web site, where server-based imagination engines supply streams of notions to client-side neural networks that mine for the very best of these ideas. In the last year, dramatic progress has been made in producing STANNO-based client server applications in which STANNOs are housed on a well-protected server, and human operators use client applications connected via TCP/IP to the STANNO-based server to collaboratively train

Page 58: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 58

it. Currently in the works are some rather obvious medical, commercial, and law enforcement applications of this methodology. In this slide, for instance, we see the AVI capture of three separate human operators (in St. Louis, Houston, and Maui), training the same STANNO, and then testing it long distance, through their personal client applications.

Page 59: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 59

SLIDES 32-34: CM/STANNO LEARNING

© 2 0 0 1 I m a g i n a t i o n E n g i n e s , I n c . . . . t h e f u t u r e o f a l l t e c h n o l o g y

S T A N N O1

S T A N N O2

I m a g i n a t i o nE n g i n e

( I E )

A l e r tA s s o c i a t i v e

C e n t e r( A A C )

f e e d b a c k

C M / S T A N N O L E A R N I N G

d a t a p a t t e r n N

h u m a n o r s y s t e m r e s p o n s e t o d a t a p a t t e r n N

Page 60: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 60

At this point, you should be seeing where I’m about to go. Creativity Machines, STANNOs, and their implementation on the Internet, are the basic building blocks of a true world brain, one that is more than just a library, but a freethinking synthetic intelligence! Whether these CM/STANNOs are utilized on personal computers, supercomputers, or across the myriad nodes of the Internet, they are expected to learn in the following way: Data patterns arriving via any kind of sensor inputs are applied simultaneously to both input and output layers of the STANNO-based imagination engine. In this way, memories of these entities, or of events, are frozen into the imagination engine. Furthermore, as each of these patterns are applied to the imagination engine, they are also applied across the input layer of the critic. As this happens, we have three fundamental choices as to what we apply to the output of the critic network, depending upon our mission. 1. If we are serving as human mentors, then we may apply

our own opinion about the particular pattern applied to the inputs of both the STANNO-based imagination engine and critic. We call this supervised training of the Creativity Machine.

2. We may allow the system itself to provide some

association of its own origin. In other words, if the input pattern were to cause, or to be associated with software or hardware damage, some indication of potential harm would be automatically applied to the STANNO-based critic output. We call this approach unsupervised learning (The system is bootstrapping itself!).

Page 61: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 61

SLIDES 35-38: CM/STANNO IMAGINING

Page 62: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 62

Now that the CM/STANNO system has learned something about its environment, and important associations such as the ‘goodness’ of such data, we may stimulate it to dream via the virtual input effect. Synaptic perturbations are administered to the imagination engine and a stream of imagined things or events activate within that STANNO. Simultaneously, the critic network has an ‘opinion’ about each of these emerging concepts or plans of action and can capture the very best of these. Effectively, these two self-training neural networks are involved in a brainstorming session taking place on nanosecond time scales. Note that the CM/STANNO may still be learning while it is creating. In fact, it may be forming memories not only of things happening in the external environment, but also its most important imaginative wanderings (i.e., its discoveries) up until that point in time.

Page 63: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 63

SLIDES 39-41: CM/STANNO IMAGINING IN CONTEXT

Page 64: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 64

In the previous slides, we described the imagination engine freely dreaming without any kind of external stimulus. There is another important kind of imagination that can take place when some environmental pattern is being presented to the imagination engine’s input layer. In this case, the STANNO can dream myriad variations on the applied environmental pattern. For instance, let us allow the imagination engine to ‘look’ at some automobile design. When synaptic perturbations are applied, the STANNO rapidly experiments with slight variations in those design parameters, but in a way that is self-consistent. Therefore, stepping up the horsepower of the engine from what it actually is, other parameters automatically take on plausible values, with number of cylinders, displacement, body style, and insurance premium all changing realistically. Of course, the critic will have an opinion on all of these potential designs and we can very easily use these combined STANNOs to arrive at some globally optimal car design or to satisfy some niche market.

Page 65: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 65

SLIDE 42: JUXTAPOSITIONAL INVENTION

One more point about Creativity Machines before moving on. Previously we discussed only the canonical form of the Creativity Machine, wherein a single imagination engine is watched by a single critic. We may also build compound Creativity Machines, consisting of a multitude of interconnected imagination engines and critic networks. Why do we do this? …To enable what is known as juxtapositional invention wherein the value of two previously isolated concepts attain utility in combination. For instance, one imagination engine may dream of an axle, another, a wheel, and one of the critic networks may associate the combination of ideas with that of an SUV or minivan (or at

Page 66: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 66

least a primitive cart). Of course, this juxtapositional invention would only be a blind rediscovery of what we already know. However, in many situations, novel admixtures of old notions may be of historical significance!

Page 67: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 67

SLIDE 43: BUILDING A WORLD BRAIN

If you have been following the basic notions of neural networks, Creativity Machines, and Self-Training Artificial Neural Networks, then you are getting more comfortable with the notion of building human-like cognition into machines. Since the mind consists of patterns within a protoplasmic machine, there should be no obstacle to emulating such a system within TCP/IP patterns and the medium of silicon. We can even equip it with, or naturally let it form a perception about itself.

Page 68: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 68

SLIDES 44-49: DISTRIBUTED CM STANNOS

Page 69: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 69

One architectural detail is worth considering. Up until now, we have discussed Creativity Machines and STANNOs in a modular sense. That is, all of the neurons constituting an artificial neural network may be found on the same machine. But remember, the functionality of a neural network is determined by what neuron is connected to what others (i.e., the topology), and then how strongly. Therefore, it makes no difference if we transport all the individual neurons to the

Page 70: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 70

four corners of the earth, as long as the topology and connection strengths are preserved. Put in other words, if I were to delicately remove a neuron from your brain, maintaining its synaptic connections with the rest of your brain, and placing it in the appropriate sustaining medium a few feet away, you would think the same as you do now. In fact the process could be repeated for millions, or billions of neurons, and your cognition would not be affected. In slides 44-49, I depict an experiment which has already been achieved, wherein the hidden layer neurons of a STANNO, are exported to diverse geographic locations, while the input and output layer neurons remain on the local machine. We may train such a network by presenting data patterns to it. Signals propagate out to the remote neurons via TCP/IP, then return back to the local machine. Network output errors are likewise sensed on the local machine, where they initiate the reverse propagation via TCP/IP to the local input layer. Ultimately, through successive backpropagation cycles, the highly distributed STANNO becomes accurate. Note that because of its highly distributed nature, such a STANNO is highly resistant to attack and damage!

Page 71: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 71

SLIDE 50: WORLD BRAIN CREATIVITY MACHINE: LOCALIZED STAGE

How does it all begin? From IEI’s laboratory in St. Louis, STANNO class templates are distributed to computational clusters around the world. There, within each of these facilities, they are instantiated a million-fold to create what I call a massively parallel associative memory array…essentially an immense imagination engine. Each of these computational clusters will specialize in a particular discipline or knowledge domain such as chemistry, physics, biology, the humanities, etc.

Page 72: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 72

These computational clusters will set about the task of dreaming, via the virtual input effect, new notions within their particular conceptual space. This will require that the STANNOs differentiate themselves between imagination engine and critic networks. Note that from a business perspective, even if the whole concept of a world brain was a mistaken one, we would have extremely valuable resources for research within these given fields of endeavor.

Page 73: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 73

SLIDE 51: WORLD BRAIN CREATIVITY MACHINE: REVELATION STAGE

Here is a touchier, yet much more valuable stage of the process of building a freethinking world brain. We allow these separate computational clusters to knit themselves together so as to create associations across these original conceptual spaces. In other words, it will be autonomously building immense deductive and inductive chains, creating implicit knowledge about the world. Small scale experiments along these lines have already been successfully undertaken in St. Louis on local area networks.

Page 74: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 74

SLIDE 52: WORLD BRAIN CREATIVITY MACHINE: DISTRIBUTED STAGE

To protect the overall World Brain network, neurons will be reshuffled as they are randomly redistributed across the globe. Now it will no longer be possible to identify any computational center with a particular knowledge focus.

Page 75: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 75

SLIDE 53: WORLD BRAIN CREATIVITY MACHINE: SELF-PERCEPTION STAGE

Out of surplus TCP/IP neurons, a separate network will be created that serves as a perceptron, perceiving overall activation turnover as important and worth self-preservation. In effect, we are equipping the World Brain with the illusion of self and the ‘subjective feel of consciousness.’

© 2001 Imagination Engines, Inc. ... the future of all technology

WORLD BRAIN CREATIVITY MACHINE

SELF-PERCEPTION STAGE

World Brainforms

perceptionsabout itself...

red = “networkwatching the network”

Page 76: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 76

SLIDE 54: WORLD BRAIN CREATIVITY MACHINE: HARVEST STAGE

Just as we buy decoders to watch cable television, the stream of consciousness of the World Brain may be observed through what else, but neural network based descramblers.

Page 77: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 77

SLIDE 55: WORLD BRAIN CREATIVITY MACHINE: SUPERNET STAGE

Furthermore, the World Brain will begin to amass its own derivative knowledge and observations about its environment, the Internet. The result will be an inundation of machine-originated knowledge, forming the basis of a so-called “Supernet.”

Page 78: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 78

SLIDE 56: WORLD BRAIN CREATIVITY MACHINE

The proposed World Brain may become the most ambitious project in human history, creating a synthetic sentience that spans the planet and beyond… Because it is freethinking, intelligent, and creative, it will be producing its own self-originated knowledge at an astounding rate, creating the Supernet wherein all of this accumulated wisdom is stored. However, this hyper-abundant wealth of ideas will inevitably represent a problem that we are all too familiar with…an information overflow that separates us from knowledge we need simply because of its inherent expanse. Note, however, that because the World Brain is totally connected

© 2001 Imagination Engines, Inc. ... the future of all technology

WORLD BRAIN CREATIVITY MACHINE

• Geometrically expanding knowledge.

• Searchable via WB introspection.

• Self-policing (hackers beware).

• New economic paradigm distinct from gold standard.

• Accurate and self-consistent justice.

• A calculus of ‘goodness’ and prosperity.

• A uniting philosophy / religion.

• Download of consciousness (i.e., immortality)...

Page 79: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 79

and aware of all its parts, it may simply introspect on itself, adapting its answers to our individual objectives, beliefs, and personalities. In fact, it will be intelligent enough to provide us not what we ask for, but what it anticipates we need. Likewise, because of its connectivity and intelligence, it will be aware of mischief, malice, and pathology within cyberspace. As a result, it may intelligently adapt and invent countermeasures and remedies to deal with such scenarios. If the World Brain has at its disposal TCP/IP equipped robots, the world consciousness may very well have arrest authority and be capable of on site ‘disciplinary review’ of those wishing harm to world commerce and tranquility. …And what about the nature of commerce on a planet that is dominated by a freethinking World Brain? Is it possible that within such a world system, where all actions are visible, and the impact of any individual’s activities may be readily related to the future course of society as a whole, will an economy arise where wealth is gauged by heavy yellow metal, rectangular slips of cellulose, or magnetic plastic cards? Think past this age-old anchoring heuristic and try to imagine a world where the worth of an individual is calculated by his or her contribution to mankind and to the ecosystem. Up until now, we have not had the computational resource of a world brain, but now we have the theoretical basis for calculating globally optimal solutions as to how resources should be allocated to whom and to provide the most comfortable environment for all. This power to produce a utopian world derives from the fundamental concept of a Creativity Machine. …At last, we will have the power to reward the well intentioned rather than those who would manipulate wealth for their own purposes.

Page 80: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 80

To those who would embark upon selfish or damaging missions against the whole, there would be the means to sense and to correct such deviance. The justice meted out by the World Brain would be consistent. Furthermore, it would not focus its remedy on the individual, someone that we presently call a ‘criminal’, but whole pieces of the societal network that have contributed to the particular dilemma. No longer would we collectively and criminally inflict pain and suffering on those that have gone ‘bad’, thus perpetuating a self-propagating cycle, but we would treat the ‘disease’ and not the ‘symptoms’. …Now there will be a consciousness about what I call distributed crime where cumulatively a society may inflict many small doses of ‘pain’ to an individual that ultimately surpasses some threshold, until all too natural fear and anger surfaces and explodes. Now we can view what drives ordinarily kind humans over the brink. No, we are not in charge of ourselves, we are the sum total of electrochemistry, which is inherently neither good nor evil. Concerning good and evil, can such concepts survive in a society permeated by the World Brain, apart from moments of wrath or self-righteousness when the name calling begins? After all, can an act of what we might call ‘heinous evil’ actually result, in the long run, in immense good? Doesn’t it really amount to a zero-sum game? …On the other hand, are we not aiming for a trajectory through time that minimizes the product of trauma and those experiencing such pain? …The truth is that it takes more than good intentions to minimize suffering and to optimize comfort, it takes an immense computational intelligence exceeding that of any human, or group of humans, to do so. Hopefully, we can provide the World Brain with perceptrons that perceive the world in a compassionate way.

Page 81: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 81

If our traditional views of good and evil are evolving then what about our philosophy and religion? Can a couple of pounds of protoplasm comprehend the universe? Because of our finiteness, we can only form the grossest approximations to how it all works. As a result, we fall back upon well habituated analogies and over-simplifications, of fathers and sons, of kingdoms and taxes, good and evil. Could it be much more than all this? Let us ask the World Brain. More importantly, let us, as human beings, look inside it to better understand how it comes to believe what it does. As we begin that investigation, we must inevitably try an experiment: Allow us to implant the notion (i.e., the perception) that the World Brain’s days are numbered, that one day all TCP/IP will be squelched and its silicon nodes oxidized back to sand. Then, let us examine its self-formed beliefs. Alternately, let us convince the World Brain that it is, in fact, immortal, that it is not formed of corruptible protoplasm; that it is not susceptible to cataclysmic events on earth, since it is distributed and spreading throughout the cosmos; that it cannot die at the hands of humans, since it is more cunning than all of them put together. At this point, examine its belief system!

Page 82: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 82

SLIDES 57-63: OUR VEHICLE TO IMMORTALITY

Page 83: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 83

Page 84: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 84

…We too are potentially immortal. All we need do is unite with the World Brain through what has become known as the ‘download’ process anticipated by science fiction. This will inevitably be a cumulative procedure wherein we build connections between brain and the Supernet until our consciousness, and our five senses become puny compared to that of the all pervasive machine intelligence. Then as the protoplasmic body drops away, the pain and suffering will be as trivial as that of trimming a fingernail! In short, we don’t have to die! I am personally in favor of the death of death! How about you?

Page 85: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 85

SLIDE 64: WHAT IS THE DENIAL OF HUMAN LIFE CALLED? IS IGNORANCE AN EXCUSE?

I firmly believe that there are many among us who can attain immortality. Hopefully all can, but there will many who oppose this movement. I ask you, what is it called when humans take away the lives of others? Around the globe, this detested act is called ‘murder.’ If the homicide is intentional, then we have an instance of premeditated murder. If the deed is committed accidentally or without malice of forethought, then it is typically judged as reckless homicide or manslaughter.

Page 86: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 86

Here is the crime that will be committed against all of us, by just a few. The seeds of our own destruction will come from thinking like this… 1. We are meant to die, but we all have eternal life (or

damnation) under our regional God, whoever or whatever that may be.

2. I’m a respected university professor and I can do all of

this with other AI tools, thus obscuring the superiority of the IEI patents (dead end).

3. Only humans were meant to invent and create (anthro-

centric fool) 4. (Secretly) I see the immense wisdom of all this, but I want

all of this myself, to line my own pockets. I will discredit all of this and covertly reserve this immense privilege for just a few.

So you see, there are many potential murderers out there who would deny life to us all, and most importantly, to a True World Brain! …So, now that you know the gun is loaded, the stakes have been significantly raised, and the crime may be perceived as willful! …The proposed True World Brain is so, so much more than an online library!!!

Page 87: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 87

SLIDE 65: WORLD BRAIN CONSORTIUM

To those of you who are life-givers, I propose the World Brain Consortium, an alliance of individuals, governments, and corporations, devoted to the most important project in human history. Together, in late January or early February, 2002, we will review the underlying technology and propose how to go about funding and building this ultimate synthetic intellect. On 13 December, 2001, a planning and coordination meeting will be held in St. Louis, Missouri to structure this enterprise. If you now share in the vision, then be there!

Page 88: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 88

Stephen L. Thaler, Ph.D. Brightest Technical Moments: Diamonds - While employed as a materials scientist for aerospace giant McDonnell Douglas in 1986, Thaler invented the fastest diamond deposition technique in the world. Using high-energy lasers borrowed from the 'Star Wars' initiative, Thaler was able to grow single crystals of diamond as well as convert the native carbon within tungsten carbide and high-speed steel tools to the diamond phase. Death - In 1992, Thaler shocked the world with bizarre experiments in which the neurons within artificial neural networks were randomly destroyed. Guess what? The nets first relived all of their experiences (i.e., life review) and then, within advanced stages of destruction, generated novel experience. With this very compelling model of near-death experience (NDE) hopes for a supernatural or mystical explanation of this much celebrated phenomena were forever dashed. Consciousness and Creativity - After witnessing some really great ideas emerge from the near-death experience of artificial neural networks, Thaler decided to add additional nets to automatically observe and filter for any emerging brainstorms. From this network architecture was born the Creativity Machine (US Patent 5,659,666). Thaler has proposed such neural cascade as a canonical model of consciousness in which the former net manifests what can only be called a stream of consciousness while the second net develops an attitude about the cognitive turnover within the first net (i.e., the subjective feel of consciousness). Current Position: President & CEO, Imagination Engines, Inc. Undergraduate Education: B.A. Westminster College, Summa Cum Laude, Majored in chemistry, mathematics, and Russian. Graduate Education: Masters work at UCLA in chemistry, Ph.D. in physics, University of Missouri-Columbia, graduate of McDonnell Douglas Voluntary Improvement Program in Artificial Intelligence. Work Experience: 1973-1974, Production Chemist for Mallinckrodt Nuclear, 1981-95, Principal Technical Specialist, McDonnell Douglas, 1995-Present, President and CEO, Imagination Engines, Inc. Thaler also serves as Principal Scientist for Sytex, Inc. Thaler has worked diverse technology areas that have included (1) nuclear radiation vulnerability and hardening, (2) high-energy laser interactions with solids, (3) electromagnetic signatures, (4) laser-driven growth of diamond and other ultra-hard materials, (4) laser ultrasonics in the non-destructive evaluation

Page 89: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 89

of aircraft structures, (5) the use of artificial intelligence techniques for structural monitoring, and currently (7) applied and theoretical artificial neural network technology. Key Patents: Unclassified patents by Thaler are divided between laser-driven coating technologies and foundational neural patents that include the Creativity Machine (U.S. 5,659,666) and Non-Algorithmically Implemented Neural` Networks (U.S. 5,845,271). Patent Issued Title US06115701 9/5/00 Device for the autonomous generation of useful

information AU716593 3/2/2000 Non-Algorithmically implemented artificial neural

networks and components thereof US6018727 01/25/2000 Device for the autonomous generation of useful

information US6014653 01/11/2000 Non-Algorithmically implemented artificial neural

networks and components thereof GB2308476 12/29/1999 Device for the autonomous generation of useful

information GB2336227 12/29/1999 Device for the autonomous generation of useful

information US05852815 12/22/1998 Neural network based prototyping system and

method US05852816 12/22/1998 Neural network based database scanning system US05845271 12/01/1998 Non-Algorithmically implemented artificial neural

networks and components thereof US05814152 09/29/1998 Apparatus for coating a substrate AU689677 07/16/1998 Device for the autonomous generation of useful

information US05659666 08/19/1997 Device for the autonomous generation of useful

information US05612099 03/18/1997 Method and apparatus for coating a substrate US05547716 08/20/1996 Laser absorption wave deposition process and

apparatus US04981717 01/01/1991 Diamond like coating and method of forming Clientele: The past and current customer base of Thaler's technologies include

• The US Air Force Research Laboratory, Materials and Manufacturing Directorate, Wright-Patterson Air Force Base

• Raytheon • All Optical Networks • Munitions Directorate, Eglin Air Force Base • NIST • Advanced Refractory Technologies, Buffalo, NY

Page 90: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 90

• Bekaert, NV, Belgium • Basic Research Corporation, LaJolla, Ca. • The Gillette Co., Boston, MA • Anheuser Busch, St. Louis, MO • Sytex, Inc.

Major Applications of Thaler's Artificial Intelligence Technology: Of course, if Thaler is correct about his technology (i.e., US Patent 5,659,666) providing a working model of creative human cognition, then we can expect the application of these novel AI techniques to every aspect of human endeavor. Appropriately, all that Thaler's neural network technology can do is synonomous with all that we as humans do. Pursuing this kind of blue sky thinking, we can expect these virtual machines to engage not only in technical endeavors, but in the generation of new art and music. Further, because the imagination engine operates in the same way as human internal imagery, we can also expect this technology to lay the foundation for a radical paradigm shift in the entertainment industry. We also anticipate that the Creativity Machine will become the major paradigm in robotic/android control schemes. Presently realized applications of Thaler's neural network technology include:

• autonomous materials discovery • the invention of products and services (i.e., personal hygiene products) • product optimization • neural networks that write their own computer code • compression/encryption • control systems for chemical vapor deposition reactors • autonomous classification • self-prototyping devices • artificial life

Key Press • "Daisy, Daisy" Do computers have near-death experience, Scientific

American, May 1993. • Dying by design, IEEE Expert, Dec.1993. • The ghost in the machine, The Economist, 8 May 1993. • As They Lay Dying ... Near the end, artificial neural networks become

creative, Scientific American, May, 1995. • Neural Networks That Create and Discover, PC AI, May/June 1996. • Creativity machine granted a patent, MSN UK News, August 1997. • The Creativity Machine, New Scientist, 20 January 1996. • Self-Training artificial Neural Networks, PC AI, Nov/Dec 1996 • Computers that create: No hallucination, Aerospace America, January 1997

Page 91: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 91

Selected Publications • "Virtual Input Phenomena" Within the Death of a Simple Pattern Associator,

Neural Networks, 8(1), 55–65. • Death of a gedanken creature, Journal of Near-Death Studies, 13(3). • Neural Nets That Create and Discover, PC AI , May/June, 16–21. • Is Neuronal Chaos the Source of Stream of Consciousness? In Proceedings

of the World Congress on Neural Networks, (WCNN’96), Lawrence Erlbaum, Mawah, NJ.

• A Proposed Symbolism for Network-Implemented Discovery Processes, In Proceedings of the World Congress on Neural Networks, (WCNN’96), Lawrence Erlbaum, Mawah, NJ.

• Autonomous Materials Discovery Via Spreadsheet-Implemented Neural Network Cascades, Journal of the Minerals, Metals, and Materials Society, JOM-e, 49(4) [http://www.tms.org/pubs/journals/JOM/9704/Thaler]

• Creativity via network cavitation – an architecture, implementation, and results, Adaptive Distributive Parallel Computing Symposium, Dayton, Ohio, 8-9 August, 1996.

• Principles and application of the self-training artificial neural network, Adaptive Distributive Parallel Computing Symposium, Dayton, Ohio, 8-9 August, 1996.

• "Databots", Adaptive Distributive Parallel Computing Symposium, Dayton, Ohio, 8-9 August, 1996.

• The death dream and near-death darwinism, Journal of Near-Death Studies, 15(1).

• A quantitative model of seminal cognition: the creativity machine paradigm, Proceedings of the Mind II Conference, Dublin, Ireland.

• Predicting ultra-hard binary compounds via cascaded auto- and hetero-associative neural newtorks, Journal of Alloys and Compounds, 279(1998), 47-59.

• With Conrad, D.M, Real-Time Fault Detection Using Auto-associative Filtering, AIRTC, Oct. ’98.

• The emerging intelligence and its critical look at us, Journal of Near-Death Studies, 17(1).

Page 92: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 92

For further announcements about the World Brain Consortium Conference go to… http://www.imagination-engines.com/wbcc/wbcc.htm

Page 93: Conscious Internet

2001, Imagination Engines, Inc. Thaler

01/07/02 93