The Future of AI: of birds and airplanes
Asimov said once that he thought the relationship between true AI and the human brain would be like birds and airplanes: where one is an organic solution to the problem, the other is man-made. The important thing is not the flapping or the feathers, but that they exploit the same properties and principles of flight in order to fly. (others have made this comparison too, but I think he said it best)
This passage from Arguing AI echoes a similar sentiment, and takes it a step further:
Just as today’s researchers work in the shadow of HAL, Hayes says, Aviation pioneers like the Wright brothers had to find ways to break free of the pseudoscientific assumptions of their predecessors. […] “The prehistory of aeronautics, both popular and scholarly, dwelled on the idea of imitating bird flight, usually by somehow attaching flapping wings to a human body or to a framework worked by a single person.”
[…] “Aerial flight is one of the great class of problems with which man will never cope,” wrote American Astronomical Society president Simon Newcomb in 1903. “The construction of an aerial vehicle which could carry even a single man from place to place at pleasure, requires the discovery of some new metal or some new force.”
Hayes continues:
“Instead of treating intelligent machines as a looking glass to explore our own quirks and inconsistencies, scientists should dispense with the Turing model altogether and build machines that leverage existing human intelligence.” […] He dubs this approach cognitive prosthesis. “The most useful computer applications, including AI applications, are valuable exactly by the virtue of their lack of humanity. A truly human-like program would be just as useless as a truly pigeon-like aircraft.”
Which I think is a very useful comparison: researchers have concentrated very heavily on things computers don’t do well: pretending to talk like people, emphasis on teaching computers common sense, and many things besides that humans do well and computers don’t. To truly see the benefit of AI, perhaps they should be trying to develop and further things computers do well: sifting through massive amounts of information, incredibly complicated math, and aiding human productivity with intelligent, user-friendly programs.
Instead of trying to replace people- a goal that seems out of reach and far more difficult than anyone could have believed- maybe the real benefit would come from Hayes and others like him working on, as he calls it, computer prosthesis.
In addition, although I’m far from an expert, I also wonder whether modern AI is just not complicated enough to stand on it’s own: Minsky’s autonomous agents working together to mindlessly create intelligence may just be too premature if the programming cannot yet make a single agent. Or in other words, a single unit of an intelligent system, while small in the grand scheme of things, may itself be far more complicated than the AI field is capable of right now.
Maybe expecting a leap from the early decades of modern computers to independently thinking machines is just too wide- and maybe computer prothesis could be the necessary evolutionary intermediary that may one day lead to true thinking machines.
