Deep Blue's stunning ability to see further into the future than any chess computer in history allowed the development team to focus on other problems, Hoane says. In fact, they essentially quit improving the search structure two years ago, he says, in favor of making the machine smarter. For one thing, they taught it the accumulated wisdom of grandmasters Big Blue hired to tutor Deep Blue. And it enlarged its database of openings and end games so the machine could rely on the experience of its human predecessors.
A product of evolution
It's not surprising that Deep Blue relies on brute force. After all, computers originated as giant number crunchers, notes David Waltz, president-elect of the American Association for Artificial Intelligence. Computers -- called "electronic brains" -- were modeled on the 1950s understanding of human thought processes.
Because the human brain supposedly had separate memory and processing centers, that pattern was emulated in the first computers. They had separate hardware for processing and memory, taking data from memory, processing it in a central processing unit, and returning it to memory.
Since then, it's become clear that the human brain is less a series of departments than an infinitely complex labyrinth of interlocked neurons with processing and memory distributed throughout the system. "Memory is not a separate box that you send signals to," Waltz says. "When you remember a scene, you may use parts of the visual cortex that you used to see the scene in the first place... The brain is clearly a parallel processor, and we can't do anything like the brain in a computer without understanding it better."
Given the new understanding of the brain, Herbert Simon, the Carnegie Mellon professor who flung the chess challenge before computerdom in 1957, says he would like to see "a chess program that did less mathematical analysis and more human thinking."
To emulate how humans approach the board, a smart computer would concentrate its efforts on the most promising moves. It might use psychology: If it "knew" that certain opponents favored certain openings, it would study those openings. If it suspected that bold, aggressive moves might unnerve an opponent into protecting a king that was not really under threat, it might make those chancy moves.
Still, given the inherent complexity of chess, it's likely that even smarter programs will favor supercomputers, Waltz says. "Some believe you don't need a machine anywhere near as powerful as Deep Blue, but I'm not sure. I think people are quite powerful processors, and I'm not at all sure you could do this on a PC, even with better software."
Not a threat?
For now, IBM is downplaying any resemblance between Deep Blue and humanity, saying "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." Instead of unsettling humanity with the idea that it had just been displaced, the IBM publicity campaign stressed that Deep Blue is really a machine designed to analyze weather, cleanse polluted groundwater or design drugs.
Harnessing the machine's awesome parallel processing could be handy, says Hoane. "In the 1960s, when writing software for the 64k of RAM found in mainframes, programmers would spend a week trying to get 20 bytes out of a program." Now, he says, the situation has reversed itself: Programmers can seek creative ways to use the machine's awesome power. "If you understand drug design and have all this computing power, you could have the doctor design a drug for your cancer in the office. You wouldn't have to wait 15 years for a drug to be designed for somebody else's cancer."
That it looks very far into the cloudy future of a chess game. It commonly searches 10 plies (five moves) ahead, according to IBM research scientist Joseph Hoane, who adds that this level of searching occupies only 1/25 of the machine's processing power. Most of the work concerns the really deep searches -- as far as 10 moves ahead.
