INTELLIGENCE

Uncork the champagne and ring the bells!

Charter AtE subscriber John McCarthy, professor of computer science at Stanford, has been awarded the Kyoto Prize, a Japanese award second only to the Nobel Prize in prestige, and equal to it in amount ($350,000).

John McCarthy is the father of Artificial Intelligence (AI), and among his numerous achievements is the invention of a computer language called LISP (LISt Processing), used by most workers in the field, and specifically recognized in the award as possibly his "most outstanding work."

McCarthy founded an Artificial Intelligence Lab at the MIT in 1958, became professor of computer science at Stanford in 1962, and from 1965 to 1980 organized and directed the Stanford Al Lab. Trained as a mathematician, he is also a remarkable engineer:

it was he, for example, who proposed the concept of computer time sharing and was involved in the production of the first time sharing systems.

Beyond that, John McCarthy is one of the very few scientists who feel a genuine responsibility for preserving science and pro-tecting it from the charlatans who would use it as a cover for their ideological and emotional needs, and from the Luddites who would destroy it altogether. He and I met in 1973 to discuss what could be done about their onslaught. I had just begun to publish this newsletter, and he became involved in SE2 (Scientists and En-gineers for Secure Energy). We still haven't won, but we are not giving up, either.

Now for AI. Intelligence is one of those things that is not easily defined, but clear to everybody anyway.

Or is it? It is very unclear to many who sit on committees to test the suitability of applicants for post-graduate studies. They often ask questions that test memory and erudition, but not intelligence, and sometimes they disqualify an applicant for being unable to solve a problem that he came back to school to learn about. A question probing for intelligence would be "If you found yourself in ancient Rome with only their technology and materials at your disposal, how would you prove that you come from a technologi-cally more advanced age?"

The question (yes, I sometimes asked it) most often turned out quite humiliating but that was not its intent. There were good tries and bad tries, which enabled the committee to judge the candi-date's ability to adapt to unforeseen circumstances, and that is how intelligence is most often defined. Erudition and memory clearly promote that ability, but they are no substitute by themselves.

For example, lower monkeys (macaques) can be caught live by a hollow coconut chained to the ground, filled with rice and pro-vided with a hole big enough to let the monkey's open hand in. I saw a film where he desperately pulls at the nut, to which he is se-cured by his own fistful of rice, but he lacks the intelligence (the ability to adapt. . . ) to open his hand and let go of the food. To the contrary, a chimpanzee in a cage, given a stick and some food placed in front of the cage, but surrounded by a rectangular bar-rier that has an opening on the far side, will (after some temper tantrums) use the stick to push the food away from himself through the opening first, and then manipulate it round the rectangle until he can reach it with his hands.

The chimpanzee, incidentally, might not have done this straight-away: he was confronted with more and more difficult configura-tions, suggesting that intelligence can to some extent be learned.

Here it may get a little confusing, so let me slightly redefine in-telligence as the ability to draw one's own conclusions from a body of knowledge to match the observed facts in achieving an object. That fits the intelligence of an Einstein or McCarthy, as well as the dim wits of the parrots ("No way to dispose of nuclear waste. . . ," "No longer an evil empire... "Run America on renewable energy. . . ")

It also fits a computer. Of course a computer is a dumb mass of matter; without a human programmer it can't even fall off the table. But there is a difference between a dumb program where the pro-grammer has built in all foreseeable decisions, and an intelligent program that gives the computer a number of strategic principles and procedures to be followed, provides a method of making deci-sions when confronted with options, and keeps improving the method by experience (i.e., has the capacity to learn).

In the latter case, the programmer provides the broad principles, but the computer makes its own detailed decisions.