May 9, 1997
Conventional Wisdom Says
Machines Cannot Think
By GEORGE JOHNSON
EW YORK -- Whether the machine or the man ultimately wins the rematch between Deep Blue and Garry Kasparov, it is probably just a matter of time before a computer prevails. What is far less certain is just what to make of such a victory.
How to define intelligence and decide who or what has it remains among science's unsolved, and possibly unsolvable, problems. Whether a machine like Deep Blue, combining lightning-fast search power with a growing database of chess knowledge, can be said to think depends on one's philosophical prejudices. And it can be surprising to learn who holds which view.
It would seem likely that Deep Blue's inventors would be touting their brainchild as a landmark in artificial intelligence, and that Kasparov would be dismissing Deep Blue as a mere automaton, triumphing when it does mostly because of an inhuman ability to consider 200 million chess positions in the time it takes to furrow an eyebrow.
Its ability to do that has enabled it to hold its own against Kasparov, and after four games they are deadlocked. Each has won once and two games were drawn. The last two games are Saturday and Sunday.
During all this, IBM scientists have taken pains to emphasize that Deep Blue is just a glorified calculator. On a special Web page put up for the occasion, IBM describes its chess expert like this: "Deep Blue is a machine that is incapable of feeling or intuition. ... Deep Blue is stunningly effective at solving chess problems, but it is less 'intelligent' than even the stupidest human."
Kasparov has been paying Deep Blue the compliment of describing it as though it were intelligent. After beating the computer last year, he said it exhibited the stirrings of genuine thought. "I believe signs of intelligence can be found in the net result, not in the way the result is achieved," he said before this week's rematch. "I don't care how the machine gets there. It feels like thinking."
Kasparov is taking what philosophers call the functionalist position. Intelligence is as intelligence does. From his point of view, the digital calculations taking place in Deep Blue's processors are as invisible as the firings of neurons in a human opponent. Biological or mechanical, the brain is a black box. All that matters is the outcome.
The IBM team, which knows the workings of the computer too well to be so impressed, is coming down on the side of those who argue that intelligence requires an ability to learn from mistakes, a talent still lacking in Deep Blue; and perhaps even emotions and the chimerical quality called consciousness.
Maybe when scientists learn more about human brains, they will be just as unimpressed by how Kasparov thinks through a game. If a brain is just a biological computer, as most neuroscientists assume, mysterious qualities like intuition should turn out to be a matter of calculation, searching a neurological database of possible solutions. Brains make up for their slowness by learning to recognize the most promising possibilities. Computers make up for their ignorance and poor skills at pattern recognition by exhaustively considering possibilities that a human wouldn't bother with.
But without programmed lessons in chess strategy, even a machine as powerful as Deep Blue would quickly become lost in the Borgesian labyrinth of possible chess games, searching and searching and never finding the right move.
The problem is what computer scientists call combinatorial explosion, and it happens even in a game like tick-tack-toe. The first player puts a mark in one of nine different cells. For each of those plays, the opponent has a choice of eight different counter moves. Then for each of those eight possibilities, the first player can respond with seven different plays. Altogether there are 9x8x7x6x5x4x3x2x1, or 362,880 ways to fill in the crisscross lines with Xs and O's. A computer that held this sprawling map of potentialities in its silicon brain would have the game down cold.
For chess, that kind of omniscience is impossible. In 1950, Claude Shannon, a Bell Laboratories mathematician and one of the inventors of computer science, calculated that possible paths through the maze of a typical chess game could number 10 to the 120th power. Even a computer exploring a billion of these variations a second would take more than 10 to the 100th power years to analyze a game completely. The universe is thought to be 10 to the 10th power of years old.
Since it is impossible to win by trying out everything, computers must limit the search to a fraction of the territory. The problem is like finding one's way in Manhattan without a map. Suppose you are plopped down somewhere and must try to get to City Hall. At the first intersection, you are faced with three choices: turn left, turn right or go forward. At the next intersection you will be faced with three more choices. Combinatorial explosion again.
A robot could systematically plumb the depths of every possible trajectory, eventually reaching the destination. But it is easier to follow some rules of thumb. City halls are generally in an area called downtown, which is usually where there is a cluster of big buildings. So head for the tallest skyscrapers.
You might waste time exploring every inch of midtown before heading toward the Battery. But this might be avoided by considering another rule: if you run into a stretch of streets that are numbered, then downtown often is in the direction of decreasing order. If you find yourself on 26th Street, look for 25th and then continue in that direction. This might at least get you to Greenwich Village, where all bets are off. But if you get stuck there, you can start looking for tall buildings again.
These rules are known in the programming trade as heuristics, and they are used by both people and machines. They don't guarantee success, but they help limit the search and tell you when you're getting warmer or colder.
In the 1950s, an IBM scientist named Arthur Samuel combined heuristics with rapid search to make an expert checker-playing program. Another IBM scientist, Alex Bernstein, worked on a machine that soon played a good beginner's game of chess.
In her book, "Machines Who Think" (Freeman, 1979), Pamela McCorduck tells how some IBM sales executives began to complain about the publicity. They were afraid that customers would be threatened by such intelligent machines. So an advertising campaign began to reassure potential buyers that computers were just mindless calculators. In playing down the philosophical implications of its electronic champion, the Deep Blue team is following in a long tradition.