One year ago today, May 11, 1997, marked a change in mankind’s view of the universe as profound as the Copernican revolution. In the final game of a six-game match, Gary Kasparov finally met his better across the chessboard. However, it was not a human, but a computer that soundly trounced the reigning world chess champion.
I have to pity Kasparov. Though largely recognized as the greatest chess player who ever lived, he reigned supreme during an era in which computers are becoming our mental equals. Fifty years ago, Kasparov’s prowess would have written a chapter in the annals of chess history for all time. Now his name will always bear a footnote: “Lost non-title match to Deep Blue.”
Deep Blue did not win the match through superior chess play. Rather, it was able to use its strengths and exploit Kasparov’s human weaknesses. As a machine, Deep Blue never became tired, flustered, distracted or confused. Kasparov, on the other hand, admitted that the pressure of the first five games had taken their toll. He had been worn down by the machine; beaten before the last game even started. “For me, the match was over yesterday,” he said. “I had no real strength left to fight. And today’s win by Deep Blue was justified.”
At the seventh move of the final game, Kasparov made a foolish, all too human mistake. The contestants had been following a standard opening until Kasparov transposed two moves in his mind, playing them in reverse order. He might as well have tried pulling the rip cord on a parachute after landing. Deep Blue, incapable of such human errors itself, capitalized on the mistake, and accepted Kasparov’s resignation at move 19. It “played like God,” Kasparov said after the match.
It was just a matter of time until the computers started to triumph over mankind’s best in what traditionally has been considered an intellectual battleground. Desktop computers have been beating average players for years. Sooner or later the fastest computers were destined to topple our chess kings. In the eyes of computing pioneers, it turned out to be later.
Arthur Samuel figured it would only take him a few months to program a computer to play championship level checkers in 1947. Those few months turned into 15 years. Working on his pet project nights at IBM, Samuel developed a program that learned from its own mistakes.
IBM in no way supported this work, unlike its devotion to the Deep Blue team, becoming upset when Samuel joked that the IBM 701 series, which he helped design, would be very useful in modeling the checkerboard. Yet despite his persistence, Samuel’s program remained incapable of beating humanity’s best, though it managed to top its own creator regularly.
By the ’70s, however, the computer community abandoned checkers because of its simplicity. Chess provided a far greater challenge, and became the game of choice.
What makes chess such a good test of intelligence for computers is the sheer complexity that arises from a finite set of rules and pieces. During the middle of an average game, there are around 35 possible moves a player can make. This results in around 1,000 possible configurations of the pieces after merely one move by each player. Allow both players another move and there are a million possibilities, then a billion and so on. Taking all the available moves from the first into account, experts estimate that there are 10120 unique games of chess waiting to be played.
That is a really big number. Deep Blue, with its 512 processors working in tandem, can analyze about 200 million moves per second. If you left it running around the clock, the computer would be able to analyze all of the games in about 10100 years. Of course the universe itself is only 1010 years old, leaving you waiting a long time for the result.
Because of this insane complexity, chess appears to be, for the time being, an intractable game. Tic-tac-toe, we all know, is simple to analyze. If both players make their best possible moves at each step, the player that goes first will always win. Checkers is not as obvious, but the general consensus is that again the first player to move should always win. But in chess, we have no idea. Maybe the game is inherently a draw, maybe white will always win, maybe black.
What we do know is that computers have become better than the best that mankind can throw against them.
But are they really exhibiting intelligence?
Some have criticized Deep Blue for the manner in which it chooses its next move. They say that the machine is not intelligent because it does not think like we do.
When Kasparov plays chess, he can consider two or three possible moves in a second. In order to play as well as he does, he must be very selective in which lines of play to analyze. To narrow his options, he is able to use human intuition and pattern recognition, calling on years of study and practice, to eliminate the obviously bad moves.
Deep Blue, on the other hand, does not have any intuition. It must think about all possible moves, hoping to finding the best one before its time runs out. This “brute force” approach to chess playing cannot be done by humans, but is ideal for number-crunching computers. They look deeper and deeper into the future of the game, ten moves for Deep Blue on average, eventually deciding which move now will lead to the best position down the road.
But that’s all it is, say Deep Blue’s detractors, number crunching. There is no insight, no understanding, no creativity and no real intelligence.
This human-centric bias is no better than judging someone based on any other superficial feature. Sure, the computer is not thinking the same way that we do, but does that mean it is not thinking at all? There are many ways to think about any problem, and chess is no different. Deep Blue has a unique personal approach to the game like any other player. It may be different from the way any human thinks about kings, queens and pawns, but it is no less valid.
We must consider the performance of the machine, not the inner workings. After all, do we have any other basis for determining the intelligence of other humans than the way they act? We look around and find that people behave much as we ourselves do. Then by analogy we assume that they have minds like ours, though we can never experience these minds directly. A computer might exhibit the same human-like behavior, but because we do have access to its internal workings, we are tempted to trivialize them and deny them any claim to intelligence. But our understanding of how they work makes them no less intelligent.
Someday neuro-physiology might reveal the precise relation between the brain and thought. Will we suddenly cease to be intelligent because we understand the science behind thought? Deep Blue might only be a savant, expert at a single task, but it still thinks its way through the problems in its restricted world.
If Deep Blue is the first of a new kind of intelligent computer, are we facing mankind’s end? Last year, headlines called the Deep Blue-Kasparov match “humanity’s last stand.” Many would have us believe that we are on the edge of a precipice, ready to fall into an age of computer mastery. However, these science fiction fueled fantasies are only paranoid delusions. Mankind is still standing strong in the grand scheme of things.
Deep Blue has not instituted an insidious plan for world domination. It only plays chess.
Eventually we will have computers that perform well in wider ranging areas. Their internal workings will be different from ours, but at that point they will be intelligent. Computers will always be our creations, but should they always be our servants? When computers behave just like people, perhaps better in some areas and worse in others, we will be forced to recognize them as thinking beings, no matter how their insides operate.
Chris Trejbal’s column appears on alternate Mondays. He welcomes comments by e-mail to [email protected]