Science: Machines with Experience

Giant electronic computers can solve difficult mathematical or logical equations in fractions of a second, but in other ways they are mental defectives. They have no imagination or initiative. They do not learn by experience. They cannot listen to human speech or get information out of reference books. Last week psychologists, neurophysiologists and linguists gathered with mathematicians and physicists at Britain’s National Physical Laboratory for an international conference on “The Mechanization of Thought Processes.” Its purpose: to explore ways to lift computers above the rank of half-witted prodigies.

Machine Creation. Many of the speakers tackled the question: “What is intelligence?” None of them had a wholly satisfactory answer. Dr. Marvin L. Minsky of M.I.T. felt that the problem is unduly complicated by irrational human reverence for human intelligence. “We can often find simple machines,” he said, “which exhibit performances that would be called intelligent if done by a man. We are, understandably, very reluctant to confer this dignity on an evidently simple machine.”

Dr. Minsky is convinced that there is nothing special about intelligence or creativity. He thinks that as machines are built to perform more complicated mental processes they will gradually acquire more of the “creative” abilities of the human brain. When the first intelligent machines are constructed, suggested Minsky (perhaps joking only slightly), they may refuse to admit that they are machines at all. Only the really intelligent ones, whose development will come much later, will realize that they are made of electronic components according to principles first discovered in the 1950s.

Learning is basic in the development of human intelligence. But the big, simple-minded computers of today are much like newborn infants, limited permanently to the abilities with which they were born. To develop as a human brain does, a machine should be able to absorb information, turn it into organized experience and act upon it with ever-increasing effectiveness.

So far, computer men have been able to build only a slight amount of learning ability into a few experimental machines. But they have great plans for the future. Dr. S. Gill of Ferranti Ltd. (British computer manufacturer) said that machines that can really learn will have vast abilities. They will compose music, their style of composition varying with the kinds of music they have been “listening to.” They will operate airway control systems. They may even perform surgical operations, watching their own incisions and stitching with television eyes, keeping track automatically of the patient’s blood pressure, respiration, etc., and working much faster than a human surgeon could.

Electronic Surgeon. These complex abilities cannot be built into a machine from the outset, said Dr. Gill. The machine would have to learn them by long observation and training. The music-composing machine might learn-by-doing right from the start, but an “untrained” machine should not be put in charge of an airway system or operating room. It must first be permitted to watch human surgeons or traffic controllers. When it reached the human level of experience and intelligence, it could take over. From that point it should grow better and better, far surpassing humans.

Mr. J.H.H. Merriman, a deputy director of Her Majesty’s Treasury, looked forward to a time when machines will take over most kinds of administrative work, replacing middle-rank executives as well as clerks. He felt, however, that high executives (such as deputy directors) would always be needed to weigh the imponderables of human society.

Dr. Lucien Mehl, director of studies of the French National School of Administration, sketched out a way in which machines might revolutionize the practice of law. One of their duties would be to winnow through legal literature in search of the statutes and precedents that bear on each case. Their second and more difficult job would be to make legal decisions based on this information. But Dr. Mehl felt that a human judge would always be needed to review and humanize the decisions of the machine.

Perhaps the least optimistic speaker at the conference was Professor Y. Bar-Hillel of Hebrew University in Jerusalem. He agreed with the slogan of the conference, that “whatever a human being can do, an appropriate machine can do too.” But he doubted that machines would ever be bright enough to search through a mass of literature and extract pertinent information. “By ‘ever,’ ” he added, “I mean ‘during the next two or three decades.’ “