Misguiding research on Artificial Intelligence.
AthensOriginally uploaded by hzenilc.
The field of strong AI, sometimes referred to as artificial general intelligence or AGI, is defined as intelligence that is capable of the same intellectual functions as a human being, including self-awareness and consciousness, I will call the assumed object of study of AGI just agi in lowercases, not only to avoid confusion with the current field of research AGI, but also to differentiate agi from what is known as AI, just as AGI is meant to identify a field apart from AI
I do think that it is techniques similar to those that led to the discovery of the Game of Life (see the previous post) that will eventually produce artificial general intelligence or agi. I think the study of the Game of Life is for the Game of Life what the fields of AI and AGI are for ai and agi respectively. In other words, there are no real fields of AI or AGI apart from adopting a suitable approach to the systems we already see around. Any simple systems such as CA are likely to achieve agi of the kind we expect without having to do anything else! From my point of view it is just a matter of technological sophistication, of providing the necessary elements and interfaces with the real world, and not of theoretical foundations or the invention of a mathematical or new conceptual framework. If the former is what AGI is focusing on, I think it is headed in the right direction, but to think of the AGI field as building agi from the bottom-up would be a mistake of the same kind as thinking of the Game of Life as designed or engineered. I find the terms usually used in the AGI field — “agi design” and “designing agi,”— discouraging.
Perhaps an engineering based approach will succeed first. Whatever the case, it is a very exciting prospect which I look forward to, and I hope to make a contribution in my own field. But that doesn’t change the fact that in the end, just as with the Game of Life, agi will have been engineered and successful because it was already there rather than because we were clever enough to reinvent intelligence, as we have reinvented flying by building airplanes. I think current AI does for intelligence what airplanes did for flying. We don’t mind how we fly but we do mind how machines think or don’t think (at least as a matter of research for AGI). So the common airplane analogy does not work anymore for this. If we haven’t been able to create intelligence of the kind seen in humans it is because we are still looking to build airplanes. While it is unobjectionable for airplanes to keep flying as they do, flying unlike birds fly, current approaches from AI and AGI to agi will just keep producing airplane-type solutions to a non-airplane problem. That does not imply that we have (as the negation of the airplane analogy suggests) to come up with agi by imitating human intelligence step by step; that’s another matter.
I don’t see why intelligence would need to be different in nature from the potential behavior of all the sophisticated systems that are around us, why it should differ from them as airplanes differ from birds. Unfortunately all current research on AI and AGI seems to have a stake in making just such a distinction. While research on complexity theory and complex systems is faring better, it still seems to me to have missed the real problem when it comes to agi.
Systems such as the Game of Life and Rule 110 (capable of universal computation) are potentially all the primary ingredients needed for agi. Disregarding matters of computational efficiency. I think it is just a matter of time before we are able just to provide the current systems with the potential power of general intelligence with the means to reach that power. Perhaps it would turn out to be an engineered approach, just as the Game of Life turned out to be capable of universal computation. But even when the authors of these approaches have the strong impression that they are designing agi, they will end up with just another powerful-enough general system capable of agi, but so ubiquitous that is more in the nature of a discovery than an invention.
–
“No one has tried to make a thinking machine. The bottom line is that we really haven’t progressed too far toward a truly intelligent machine. We have collections of dumb specialists in small domains; the true majesty of general intelligence still awaits our attack. We have got to get back to the deepest questions of AI and general intelligence and quit wasting time on little projects that don’t contribute to the main goal.”
— Marvin Minsky (interviewed in Hal’s Legacy, Edited by David Stork, 2000)