The Father Of Ai Says His Child Has Gone AstrayGary Mcwilliams
Marvin L. Minsky, one of the patriarchs of artificial intelligence, is in a feisty mood. The man who co-organized the 1956 conference that launched the quest for artificial intelligence and set off a search for the holy grail of "thinking" computers now sits on a couch in his Brookline (Mass.) living room and chastises those who would reduce his grand vision to a series of equations that solve a simple problem.
The idea of software that replicates human thinking still burns brightly in Minsky's brain. But progress toward his goal has slowed, he says, because AI "was taken over by people who tried to be too formal and mathematical about it."
Minsky, 64, has been frustrated about the course of artificial intelligence for years. At Massachusetts Institute of Technology's Artificial Intelligence Laboratory, which he co-founded with fellow AI patriarch John McCarthy in 1958, Minsky worked to understand how humans learn. The laboratory came up with Lisp, for list processing, the first programming language in the field. But a series of startups that tried to commercialize it eventually failed because the language was too cumbersome.
In 1969, alarmed at the tilt by other researchers to neural-network technology, Minsky co-authored a book called Perceptrons, debunking the idea that software could actually simulate the function of human neurons. His influence was still so great that almost overnight, funding for neural-network projects dried up--an outcome Minsky says he never intended.
CHILD'S PLAY. Neural networks aren't his only pet peeve. Minsky has been speaking out against just the type of relatively simple systems that fall under the heading of applied intelligence. "We have lots of expert systems that do specialized things, but there isn't any machine that understands the things a 6-year-old knows," Minsky says. For instance, most of today's intelligent software doesn't know what any child does--that you can pull a string but can't push one, for instance.
In answer to a narrowing concept of intelligent software, Minsky in 1986 put forth an overarching theory of intelligence. Presented in his book The Society of Mind, his theories cover everything from how language arises to his argument that machines need not operate purely on logic to have reasoning. The core idea is that intelligence involves the interactions of a series of otherwise "dumb" programs--like the one that lets you breathe without thinking about it.
But these days there are few researchers paying any mind to Minsky. In the past, graduate students flocked to the AI pioneer, elaborating on and trying to implement his ideas. Two examples are Danny Hillis, founder of Thinking Machines Corp., a maker of supercomputers, and Raymond Kurzweil, founder of three AI companies, including speech-recognition software maker Kurzweil Applied Intelligence Inc. Kurzweil began corresponding with Minsky in the 1960s, when he was a Queens (N.Y.) high school student. Now, he thinks of The Society of Mind as just a "philosophical poem."
TINKER TOYS. Meanwhile, the AI research community has taken up neural networks and other ideas Minsky opposed. Yet Minsky retains his childlike exuberance. Besides a mess of papers, his living room is crowded with two pianos and dozens of odd gadgets, such as lamps that are turned on by the sound of clapping but keep flickering off, inexplicably. Roger C. Schank, a Northwestern University AI researcher, makes fun of his longtime associate's gadgets. "Once those lamps are fixed," Schank says, "there's no possibility he would ever use them."
That's perhaps the main criticism of Minsky: He's far more interested in the ideas and theories of technologies than getting the stuff to work right. To this day, he has never been very specific about how his Society of Mind theory might work. Says William R. Swartout, a director at the University of Southern California's Information Sciences Institute: "Marvin feels his latest theory is very specific, but in areas, it is still a little vague." On the other hand, it's often grand but vague ideas that lead--eventually--to real breakthroughs.