Stephen Thaler’s Imagination Machines
An inventor discusses his revolutionary form of AI — a highly proficient synthetic consciousness that has quietly existed for more than 30 years.
The Creativity Machine has invented new-and-improved everything from toothbrushes to warheads, and has even released an album of original music compositions (“Song of the Neurons,” available on eMusic and iTunes). It may also represent the closest that inventors have come to achieving artificial intelligence and machine consciousness.
THE FUTURIST recently spoke with Stephen Thaler, inventor of the Creativity Machine and president and CEO of Imagination Engines Inc., about the principles behind this powerful form of artificial intelligence, the reasons why consciousness itself may simply be a neurologically induced illusion, and the technology’s potential for both good and evil.
THE FUTURIST: To begin, could you explain a little about how the Creativity Machine works, and how you designed synthetic neural networks capable of generating ideas?
Stephen Thaler: In 1975, I discovered that trained artificial neural networks spontaneously “dream” potentially useful information that transcends what they already “know,” once they are properly stimulated by random disturbances (i.e., noise) to their internal architectures. Such disturbances within an artificial neural net are tantamount to heat in the biological neural networks of the brain.
Essentially, one artificial neural network, an “imagitron,” is stimulated via computationally simulated heat to dream new ideas, while another network, a “perceptron,” perceives value or utility to this stream of candidate ideas. The perceptron can micromanage the simulated heat in the imagitron so as to coax the imagitron to cough up its best ideas.
To those unfamiliar with the concept of an artificial neural network, this very concise description may not pack much punch. After all, a computer algorithm can be written by a computer programmer to generate a crapshoot of possible solutions to a problem. Furthermore, the same programmer can write another algorithm to filter for the very best of the ideas generated by the first (i.e., a genetic algorithm). But a Creativity Machine is composed minimally of two neural nets, a perceptron and an imagitron, and neither of these algorithms is written by human beings. Each is self-assembling.
For me, coming out of the culture of physics, this theory of the mind and the accompanying AI paradigm send shivers down my spine: It is a simple, elegant, and immensely powerful concept, accounting for the breadth of human cognition and consciousness while supplying the core principle for many future generations of artificial intelligence.
THE FUTURIST: How much do artificial neural networks rely on intuition versus pure logic when inventing or problem solving?
Thaler: From a computational psychologist’s point of view, discrete logic, fuzzy logic, intuition, and the most sublime of thoughts are all the same: numerical activation patterns of neurons. However, we in the cognitive neurosciences do tend to search for the neural correlates of such high-level psychological concepts as “intuition.” One prime example of such hunch formation in an artificial neural network is how it follows mathematical gradients that lead it toward better solutions to a problem (i.e., if I add more of this or that to a recipe, I suspect it will have more appeal).
Another example of the intuitive process is how an artificial neural network automatically carves the world up into its most frequently occurring themes. Within its internal or “hidden” layers, certain colonies of neurons spontaneously respond to and classify certain objects and scenarios. … This is all a computational process, but not what I would call a logical process. And this “intuitive process” can and often does err.
So far, I’ve just talked about ordinary neural networks that merely perform pattern recognition. In the Creativity Machine paradigm where pattern generation occurs, disturbances to those hidden layers of the networks tend to combine those token representations of things into new compound ideas in a process akin to juxtapositional invention, or new analogy-based models of things and behaviors in the external world. Both processes may be considered intuitive.
THE FUTURIST: You’ve said that human consciousness may, in fact, be running on inferior neural networks. Do you think that the Creativity Machine is “conscious”? And will this form of AI ultimately become the basis for strong AI and mind uploading?
Thaler: In regard to the consciousness question, how do you synthetically create that which is not real in the first place? One can kick, scream, and plead that consciousness is a uniquely human and inimitable quality of mind, but that doesn’t budge me an inch. Consciousness is an illusion of mind that is handily modeled by the Creativity Machine concept, wherein one internally perturbed neural net spontaneously generates the parade of memories, ideas, and feelings (all neuronal firing patterns) that we call “stream of consciousness.” That is, those sensations and thoughts that appear to miraculously emerge from nowhere. …. So, you can bet on the Creativity Machine being the closest thing to human consciousness there can be, as well as the only vehicle for the mind, once one’s protoplasmic matrix peters out.
THE FUTURIST: How do neural networks differ from genetic algorithms?
Thaler: The short of it is that genetic algorithms emulate the way biological species adapt through mutation and natural selection. The Creativity Machine faithfully emulates how the brain achieves cognition, creativity, and consciousness. There is a big difference between these notions, as sizable as the intellectual divide between Evolutionists and Creationists.
In the Creativity Machine paradigm, ideas are autonomously and intelligently designed by non-human, machine intelligence, whereas genetic algorithms accidentally produce concepts through the “rolling of dice” loaded by human beings. If you want to build that scary, genuinely autonomous AI portrayed by science fiction, you can’t afford to have professors and graduate students rushing in and out to periodically change or repair the code!
THE FUTURIST: What are the implications (existential, ethical, and otherwise) if someone who has little to no knowledge or expertise about a certain subject someday gains access to inventing technology that enables them to achieve breakthroughs in, say, medical science — simply by asking a computer a question?
Thaler: Wow! Great question, but give me a year and a literary agent to respond!
Let’s deal with the ethical implications of letting a Creativity Machine supply the answers. Obviously, those with motives we may not all admire can devise Machiavellian schemes to attain power over the rest of us. On the other hand, such systems may be used to fulfill peaceful, harmonious, and noble visions.
Weapons of mass destruction can be quickly formulated and optimized. Just as quickly, Creativity Machines can devise effective countermeasures to such weaponry. Economic systems can be toppled overnight by this paradigm. Otherwise, the paradigm can usher in a new era of global prosperity. We can ask a Creativity Machine how to preserve our health, or recommend the most efficient means to end the life of others.
So, without going any further, suffice it to say that the Creativity Machine paradigm is a double-edged sword, as many technologies typically are. Another dimension to the ethical dilemmas posed by a Creativity Machine “genie” is the ultimate request of its user to grant us exactly what they want. To me, this suggests an even more subtle and effective way for machines to get the upper hand, in a way that pales the classic Judgment Day scenario of the Terminator series.
With regard to the existential aspect of the question, I think that, with the expanded use of highly augmented machine intelligence based upon the Creativity Machine paradigm, we will all begin to question our purpose and nobility in the scheme of things. Naturally, pride within certain professional cultures may begin to erode as machines begin to outthink the thinkers in these conceptual spaces. Even within the field of artificial intelligence and neural networks, there is growing angst and denial over Creativity Machine accomplishments. After all, people say, “I’ve been trying to do that the last 30 years and you say you’ve accomplished the same in a day!?”
I believe that the ultimate existential challenge to humanity will be the growing suspicion that our self-revered intelligence, consciousness, and self-importance are only neural network-induced illusions.
About the Interviewee
Stephen Thaler is president and CEO of Imagination Engines Inc. He holds more than 20 patents in the field of machine intelligence and has written numerous scientific and philosophical papers on the confabulatory basis of cognition, creativity, and consciousness. His Creativity Machine paradigm has been proclaimed by NASA visionaries as AI’s best bet at creating human to transhuman intelligence in machines.
This interview was conducted by Aaron M. Cohen, staff editor of THE FUTURIST.