May 18, 1997
Machine vs. Machine: Deep, Deeper, Deepest Blue
or those who worry that losing a chess match to a machine somehow diminishes humanity, Deep Blue's victory over Garry Kasparov last week was easy to explain away. Deep Blue is a powerful machine, and the outcome of the match was no more surprising than if Mr. Kasparov had tried to outrun a train. But if you think again, the whole thing is eerie. One kind of computer, made of neurons, lost to another kind of computer, made of microchips.
chess is just calculation, right? The rules of the game, the various strategies and algorithms are all coded into computer language. Feed a string of binary digits into one end of the machine -- the sequence of 1's and 0's describing Mr. Kasparov's latest move -- and information bounces around in the computer like steel balls in a pachinko game. Then out comes another string of 1's and 0's, the computer's countermove.
Another Pachinko Game
But if the reigning model of the cognitive scientists is right, the brain is also a computer, a shuffler of information. Deep Blue's latest move falls like a shadow on Mr. Kasparov's retinas -- a knight where there was once an empty square. Then the image is converted into a sequence of electrochemical pulses that are carried up the optic nerve and into the brain. There the pulses -- like pachinko balls again -- are sifted through the sieve of neurons, circuitry fine-tuned over the years to calculate the best moves in the game.
When the neurological ricocheting is done, another string of pulses emerges from the brain and causes motor nerves to direct the muscles in Mr. Kasparov's hand to pick up a chess piece and move it to a new position on the board.
It's not a matter of man versus machine but machine versus machine. At this point, Mr. Kasparov is still a much more interesting computer than Deep Blue. He knows when he makes a bad move and displays behavior that other humans interpret as regret. In tense moments, signals are sent to his stomach causing it to secrete acid. Sensory nerves detect the corrosive fluid and send danger signals to his brain. The pain motivates him to focus his attention, make better moves and relieve the suffering.
Evolution programmed him that way.
Deep Blue doesn't have a stomach, and it devotes every bit of its resources to winning the game. It doesn't need some jury-rigged contraption like regret to focus its energy. But someday there will be a Deeper Blue. Like Mr. Kasparov's brain, it might scrutinize its opponents' physical behavior for signs of nervousness and scan its own past moves to see which could have been better.
To make sure Deeper Blue doesn't fritter away too much time on these ancillary activities, programmers might give it the digital equivalent of a stomach -- an internal gauge whose reading (call it anxiety) increases when it senses that its opponent pulls ahead. When the anxiety level is low, Deeper Blue might allow itself the luxury of contemplating long-range strategies or surprise moves. When the anxiety reading is high, it would focus on the immediate game.
Artificial anxiety might be the key to making Deeper Blue want to win. Its rules about chess could be supplemented with rules about life: that the goal is not just to make the best possible play each turn, but to keep on winning and to gain fame. What is "fame"? Define it as the number of times that one's name appears in television broadcasts, newspaper and magazine stories and on the Internet. If your opponent is winning the recognition game, the anxiety meter would start rising. To relieve the tension, the computer would draw more attention to itself. Call a press conference or throw a fit -- fire up the random number generator to induce erratic and newsworthy behavior.
For computers and brains to be aware of something they must have an internal model of it -- a representation, either digital or neurological. In the recent match, Mr. Kasparov kept honing his mental model of Deep Blue, developing a theory of how the machine worked. But Deep Blue could not learn on its own. The programmers had to do the fine-tuning. In last week's match, the evolving model of Mr. Kasparov was distributed among the programmers, their chess consultant and Deep Blue itself. But there is no reason why the entire representation of Mr. Kasparov could not eventually be contained inside Deeper Blue's memory.
And if Deeper Blue can have a model of Mr. Kasparov, then Deepest Blue probably could also have a model of itself.
In some books that would count as consciousness. At this point, Mr. Kasparov or his great granddaughter might have a sporting chance again, for they could wage psychological warfare against the computer, just as they would with a human opponent. And since the machine would now have stooped to its opponents' level, it would only be fair to give the human player a computer to do fast chess calculations. The game would become interesting again.