Chasing Intelligence
In May 1997, the world chess champion, Garry Kasparov, was beaten by IBM's Deep Blue. It was a massive media event, a massive victory for artificial intelligence, and a definite blow for the human ego, which had long perceived the complex game of chess as the touchstone of human intelligence. Dreyfus, writing in 1979 of the failure of the chess program MacHack to beat all but amateur players, noted smugly, "During the past two years, it [MacHack] lost all games in the tournaments in which it had been entered, and received no further publicity. We shall soon see that given the limitations of digital computers this is just what one would expect."(2) As Sheryl Hamilton put it, "Again, somewhere in this favoured land, monitors are still shining bright. Somewhere (computer) nerds laugh and somewhere (IBM) programmers shout (for joy). But there's no joy left in Mudville because mighty Kasparov struck out."(3)
But an intelligent computer still does not exist. One cannot carry on a conversation with Deep Blue, and Deep Blue does not pass the Turing Test - it cannot be mistaken for human. It can only play chess.
In his book What Computers Still Can't Do (a revised and updated version of the 1979 revision of What Computers Can't Do) Dreyfus attempts to categorize intelligent activities into four areas. Area I is "associationistic:" it is innate or learned by repetition and it is not affected by the situation. Examples are memory games or the word-by-word translators commonly available on the Internet, which will happily churn out sentences of nonsense Spanish if you ask them to. Area II is "Simple-Formal." It, too, is learned by rule, and applied in highly structured situations: examples are tic-tac-toe and mathematical proofs. Area III is "Complex-Formal", also learned by rule or practice but heavily dependent on the situation for the correct interpretation. Examples are chess (which computers can now play), Go (which computers cannot), and recognition of "complex patterns of noise", speech recognition, which programmers are currently struggling with. Area IV is "Nonformal," entirely situation-dependent, and learned by example and what can only be called intuition. Examples are riddles, effective translation, and sentence recognition. No computer has been built which can perform any type of Area IV activities.(4)
The error in judgment occurring here is the assumption that intelligence, as pure, rational problem-solving, can be programmed, an error which has occurred throughout the history of artificial intelligence, from HAL 9000 of 2001, Stanley Kubrick's famous computer, which beats its crew members at chess and eventually attempts to kill them in an act of pure logic, all the way to his real-life brother Deep Blue. It is this tradition which sparks Douglas Adams's ironic parody Deep Thought, the "stupendous super computer which was so amazingly intelligent that even before its data banks had been connected up it had started from I think therefore I am and got as far as deducing the existence of rice pudding and income tax before anyone managed to turn it off."(5) It is the notion of a disembodied and purely rational intelligence far superior to our own.
It is also mythological, and likely to continue so. As Antonio Damasio points out in his book Descartes' Error: Emotion, Reason, and the Human Brain, reason is not necessarily all that reasonable. His case studies are people with specific sorts of brain damage, people who pass all laboratory tests of intelligence and who show undiminished capacity for logic and reason but who are yet incapable of functioning in society. Their sole symptom of damage is an increased emotional detachment which in turn leads to an inability to distinguish what is important from what is not. They can clearly identify the important features of a hypothetical situation - they still have their "rules" - but cannot apply them to real-world situations, or, to use Dreyfus's terminology, they cannot perform Area IV activities. Of one such case, whom he calls "Elliot", Damasio writes:
Imagine a task involving reading and classifying documents of a given client. Elliot would read and fully understand the significance of the material, and he certainly knew how to sort out the documents according to the similarity or disparity of their content. The problem was that he was likely, all of a sudden, to turn from the sorting task he had initiated to reading one of those papers, carefully and intelligently, and to spend an entire day doing so. Or he might spend a whole afternoon deliberating on which principle of categorization should be applied: Should it be date, size of document, pertinence to the case, or another? The flow of work was stopped. One might say that the particular step of the task at which Elliot balked was actually being carried out too well, and at the expense of the overall purpose.(6)
The similarity to the prime problem in programming an artificial intelligence - too much information - is striking. Dreyfus notes that "In chess programs... it is beginning to be clear that adding more and more specific bits of chess knowledge to plausible move generators, finally bogs down in too many ad hoc subroutines.... What is needed is something which corresponds to the master's way of seeing the board as having promising and threatening areas."(7) Progress has been made in this area, as witnessed by Deep Blue and some rudimentary voice recognition software, but nothing near the capacity to organize and use information that a human being exhibits.
previous page
next page
Contents copyright © K. Feete, 2002. All rights reserved.