Herbert Simon.
Courtesy of Carnegie Mellon University Department of Psychology

Two kinds of smarts
One of the most frustrating sagas in the history of computers concerns artificial intelligence (AI). Once touted as the last frontier of machines that were originally called "electronic brains," AI has been a troubled child.

Back in the "electronic brain" era, experts predicted that computers would soon be thinking like people and diagnosing diseases, designing cars, even doing psychotherapy. But while expert systems can now help with diagnosis, the more elaborate -- and human-like -- goals remain vaporware (defined). The reason? When AI scientists finally grasped the complexity of human thought, they had to scale back their focus to making expert systems that would help humans with complex tasks, not take them over entirely. Common sense, we know now, is not so common.

Isn't thinking hard?
Nowadays, computers are not designed to emulate us, but rather to do specific jobs better than we can, and the humbled field of AI is looking at more limited -- but still formidable -- goals, like interpreting human speech, making computers understand visual images, or directing robots in outer space.

Are machines like Deep Blue "intelligent?" That depends, says Herbert Simon, a Carnegie Mellon psychology professor who helped originate the fields of AI and computer chess in the 1950s. He notes that AI folks use two definitions for intelligence: "What are the tasks, which when done by humans, lead us to impute intelligence?" and "What are the processes humans use to act intelligently?"

Measured against the first definition, Simon says, Deep Blue "certainly is intelligent." And he says "it partly qualifies" according to the second. "It certainly did use an enormous amount of [computer] cycles, [a hallmark of brute force], but it also used a limited amount of rules."

Rules abound
And rules, Simon points out, are used in AI programs for things as diverse as finding oil and diagnosing diseases. These programs may contain thousands of "if-then" statements. This is a human-like strategy, Simon maintains, because "a large part of human knowledge is stored in this form." If our knowledge base could be measured, he expects it would come to "maybe 10,000 or 100,000 statements."

Thus a typical common-sense life rule might read, "If you are driving a car, do not aim at stationary objects like trees."

A typical chess rule might read, "If you are in check, then you must move your king, destroy the attacking player, or put something between the attacker and your king."

In carrying out these rules, Deep Blue uses about 6,000 "coefficients" for evaluating various moves, says Joseph Hoane, an IBM research scientist and member of the Deep Blue team. These coefficients are "a crystallization of what I learned from grandmasters" and have written into the code. "If you went over 100 chess books and extracted all the rules of thumb, that would have a large overlap with these rules."

But rules aren't perfect, even if you are smart enough to think of one for every possible situation. For one thing, Simon points out, rules can conflict. Thus our simple driving rule would conflict with a rule that says, "Ignore the rule about not aiming at stationary objects if the objects are being carried on a truck. Then use the rules about moving vehicles."

In chess, rules might conflict when both sides were conducting attacks. Then we'd need rules -- or coefficients -- to tell us which threat would materialize first.

Still, while Deep Blue's code contains thousands of such rules, the machine's behavior -- its "smarts" -- rests upon it's eerily unhuman ability to look deep into the future possibilities of the game. In other words, its intelligent behavior still rests on brute force, not human-like intelligence or intuition.

Brute force to the rescue
In an essay, no less an authority than Monty Newborn, chairman of the Association for Computing Machinery's Chess Committee, says "the brute force approach ... is apparently in vivid contrast with the search done by grandmasters." Newborn wrote about computer chess in "Kasparov versus Deep Blue" (see our bibliography).

But even if we're nit-picking here -- and we don't think we are -- Deep Blue's victory opens the door to defter approaches to machine chess, Simon says. "Now we can proceed to the really interesting task -- making a computer that uses really humanoid methods and speeds" to solve big-time chess problems.

Simon should know. It was his challenge, 40 years ago, that helped spur computer-whizzes to pit machines against grandmasters.