A blue-ribbon team of computer experts staged a classic intellectual confrontation here this week between homo sapiens and some mindless machines. The man-versus-machine contest took four hours. The (human) experts probably will spend the next four years trying to settle who won.
At a meeting of the Association for Computing Machinery -- an academic conference featuring a long series of reports with titles like "Compiler Optimization by Detecting Recursive Subprograms" -- a group of programmers decided to try the most famous test of so-called "artificial intelligence": The Turing Test.
The test was devised by Alan M. Turing, a British computer pioneer whose visionary and occasionally unorthodox way of thinking (he committed suicide by poisoning an apple with cyanide and eating it) played an important role in shaping the modern computer age.
In a seminal 1951 paper, Turing predicted that "normal" conversation between man and machine would be common "in about 50 years time." Film director Stanley Kubrick saw that paper, did the addition and made a movie about the year 2001.
Turing also proposed a test to determine whether a computer had "intellect." A human seated at a keyboard would carry on a typewritten conversation with an interlocutor hidden behind a curtain. After a few minutes of interrogation, the human would have to guess whether the curtain concealed another human or a computer programmed to answer like a human. A time would come, Turing predicted, when computers could regularly fool the human player into thinking he was talking to another person. Then people would have to admit that machines could "carry out something which ought to be described as thinking."
Turing proposed this famous test 35 years ago this month. This week's convention offered the group a perfect setting for a commemorative Turing test, because it featured the 16th Annual North American Computer Chess Championship among 10 different chess-playing programs -- on machines that range from a $2,000 Apple IIe to a $5 million Cray supercomputer -- in a three-day round-robin.
While some of these chess programs are good enough to rank in the top 1,000 or so chess players on earth, the computers don't know the first thing about chess. All they know are the instructions that human thinkers have written for them. Without these instructions -- some extending to 50,000 pre-programmed moves -- the machines couldn't tell a castled king from a passed pawn.
Instructing a computer to play good chess is an esoteric calling whose practitioners debate endlessly about such programming twists as the "alpha-beta pruner" and the "killer heuristic." Since many programmers try to make their machines mimic human players, the group reasoned that a chess-playing program should be a good entrant for the Turing test.
In the Turing test here, a human chess master played simultaneous games against eight opponents behind a curtain. Some of the opponents were computers, some human. At the end of the games, the human player and the audience were to guess which opponents were machines.
The human given the "honor" of taking on eight major-league opponents at once was polite, 17-year-old chess phenom Alex Fishbein, a computer science major at the University of Colorado. On first meeting, the slightly stocky Fishbein -- wearing wire-rimmed glasses, a two-button blazer with both buttons buttoned, a white shirt with its tail hanging down below the blazer, and a shapeless mop of dark brown hair -- looked as if he had showed up for a casting call for "Revenge of the Nerds."
But once the games began, the sheepish, soft-spoken teen-ager turned into a tiger. With a steely confidence that never flagged during the four hours of play, he marched resolutely along the long row of chess boards, making a move against one opponent and then striding quickly along to assess his strategy against the next.
Whether Fishbein would win the games was irrelevant to the Turing Test, of course. But the audience, some of whom used portable computers on their laps to track the action, soon began pulling for the tireless young master. When the mystery opponent on Board 3 conceded defeat after 150 minutes, a murmur of approval swept the room. When Fishbein stormed to his second win five minutes later on Board 7, the spectators were moved to applause.
When his last opponent finally resigned, Fishbein had compiled an enviable record: five wins, two losses, one draw. But now the time had come to take the Turing Test. When the human player tried to guess the nature of his eight opponents, he flunked.
Three computer programs were among the eight mystery players. Two of them convinced Fishbein that they were human. For the first time in history, the computer passed the Turing Test.
But not completely.
While Fishbein was playing his eight games, members of the audience took the Turing Test as well. For the most part, they were not fooled. A few spotted all three computers, and several spotted two of the three.
"Chess is something computers can do almost as well as humans," noted Garth E. Courtois Jr., a programmer who supervised the Turing Test. "And our audience was able to identify them even in a chess game. This was not a particularly bright day for the computer."