by Alice Boatwright
In 1950, British mathematician Alan Turing predicted that by the end of this century computers would probably be able to match human performance in a variety of intellectual tasks, and the field of artificial intelligence, known by the shorthand AI, was born.
In addition to inventing the idea of the general purpose computer, Turing helped launch an ongoing debate over what constitutes intelligence, the relationship between intelligence and consciousness, and whether a machine can truly be described as having either quality.
This spring some of AI's foremost researchers and harshest critics will gather on campus to participate in a Berkeley Extension course, "Turing's Prophecy: Where Is AI After 50 Years?"
Over nine evenings, from March 4 to April 29, they will discuss the high hopes and unsubstantiated claims, remarkable successes and devastating failures, brilliant insights and stubborn questions that have characterized artificial intelligence's first 50 years.
"We thought this was a good time to review where things stand," says Edward Theil, the course organizer. Theil, who calls himself an "enthusiastic amateur" in the field of AI, is principal and founder of the Berkeley Technology Group and former head of the Human Genome Informatics Group at the Lawrence Berkeley National Laboratory.
"Most of Turing's predictions have not come true -- or they haven't come true yet," explains Theil, "but many of them still seem possible -- perhaps in the next 50 years. A lot of the problems encountered with AI have come from it being oversold in the beginning," says Theil.
According to Stuart Russell, a professor of computer science at Berkeley who will lecture in the course, "One of the main things we've learned is that the problems are much harder than was anticipated in the 1950s.
"We're much more impressed by human intelligence as a result."
Nonetheless, AI has made substantial scientific progress, to the point where today's expert systems successfully diagnose diseases, neural networks make millions on the stock market and planning software can map out manufacturing processes with millions of steps.
Despite all the progress, whether or not the computer "really thinks" is a question still unanswered.
Russell suggests that this is a little like arguing over whether airplanes fly or submarines swim.
"Artificial intelligence is about getting computers to do things considered intelligent."
"Whether the field is viewed as successful or not has depended in part on how that is defined," he says.
Over the years, three major strands of AI have developed.
The first uses the computer to simulate, and thereby shed light on, specific aspects of human intelligence.
The second is aimed at making computers more useful, and the third tries to understand intelligence as an abstract quality that appears in some form in human beings and could also occur in machines.
Berkeley's Mills Professor of Philosophy, John Searle, who will speak on "Computers and the Mind," says "It's very important to distinguish between simulated thinking and the real thing."
"I don't deny you can build a mechanical brain," says Searle, "but computation is insufficient to suggest consciousness."
According to Russell, "no one has ever written a serious paper that would shed any light on how one would go about building a conscious machine, or on how one would tell that it was conscious once it was built."
But, he suggests, "If you focus on behavior, you absolve AI of the responsibility for producing consciousness. It's still an interesting question though -- what if your son or daughter wanted to marry a computer because it was a lot more interesting, witty and sympathetic than a classmate?"
Russell's own research focuses on the development of what he calls "rational agents" -- software that can choose the best possible action in a given situation.
In 1995 he was cowinner of the Computers and Thought Award, the top international award in AI, for his work in this field.
According to Theil, AI is now moving away from the past controversies towards an over-arching theoretical framework.
The technology, adds Russell, is also becoming more robust and useable.