Think Piece

Nearly 5,000 members of the artificial intelligentsia were camped at UCLA last week, the largest academic conference ever held in Westwood. Their meeting is devoted to the subject of machine intelligence, a goal that has tantalized computer researchers since Alan Turing first dreamed up those clever devices.

For years there have been endless articles asserting that scientists are on the verge of achieving artificial intelligence, that it is just around the corner. The truth is that it may be just around the corner, but they haven’t yet found the right block.

Artificial intelligence aims to build machines that can think. One immediate problem is to define thought, which is harder than you might think. The artificial intelligentsia complains, with some justification, that anything that their machines do is dismissed as not being thought. For example, computers can now play very, very good chess. They can’t beat the greatest players in the world, but they can beat just about anybody else. If a human being played chess at this level, he or she would certainly be considered smart. Why not a machine? The answer is that the machine doesn’t do anything clever in playing chess. It uses its blinding speed to do a brute-force search of all possible moves for several moves ahead, evaluates the outcomes and picks the best. Humans don’t play chess that way. They see patterns, which computers don’t.

This wooden approach to thought characterizes machine intelligence. Computers have no judgment, no flexibility, no common sense. So-called expert systems, one of the hottest areas in artificial intelligence, aim to mimic the reasoning processes of human experts in a limited field, such as medical diagnosis or weather forecasting. There may be limited commercial applications for this sort of thing, but there is no way that the process can be generalized to make a machine that can think about anything under the sun, which a teen-ager can do.

The hallmark of artificial intelligence to date is that if a problem is severely restricted, a machine can achieve limited success. But when the problem is expanded to realistic proportions, computers fall flat on their display screens. For example, machines can understand a few words spoken individually by a speaker that they have been trained to hear. They cannot understand continuous speech using an unlimited vocabulary spoken by just any speaker.

The artificial intelligentsia correctly argues that it is unfair to ask them to solve all problems at once. Advances in knowledge occur incrementally, and they have been at it only a short time. But one cannot learn to fly to the moon by climbing trees, which is what they have been doing up to now. As many of them concede, a new breakthrough is required, some new approach as yet unknown and unimagined.

Their effort is fascinating. It pushes knowledge about machines and knowledge about thought. But so far, as noted here before, artificial intelligence is real stupidity.