Eliza quickly became a hit at MIT. Weizenbaum, a trickster at heart, was amused by all the people calling for appointments to talk with his computer, and by incidents like the time someone logged on to Eliza by mistake, thinking they were chatting online with him. But he began to wonder the day his secretary asked him to leave the room so she could talk privately with Eliza about personal matters. When his partner Colby suggested the use of computer therapists to help people with real problems, Weizenbaum stopped smiling.

The computer wasn’t really thinking, he kept saying, it was all just sleight of hand. But no one was listening. Writing a decade later, in his book Computer Power and Human Reason, he still hadn’t gotten over how “extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.” But normal people were only a part of his concern. That those outside his field might be taken in was at least conceivable. But how to explain why top computer scientists were doing much the same thing?

Weizenbaum’s doubts just caused exasperation among his AI colleagues. But support for his critique came from an unexpected quarter—philosophy. If Western philosophers had labored twenty-five hundred years without producing a final definition of what thinking was, they could at least say with some clarity what it was not. Hubert Dreyfus became the first to join the fray. A physicist turned philosopher, he had expressed strong doubts about AI as early as 1961. Three years later he was hired by the Rand Corporation, the military think tank, to evaluate AI and came back with a scathing denunciation titled Alchemy and Artificial Intelligence.

In that report and later writings Dreyfus maintained that the approach then being used for AI would never work—that, like climbing a mountain to reach the moon, it was based on a naive theory. In the empty symbols being shuffled around in linear-logic processors, there was no larger awareness, he argued, of the complex background knowledge that confers meaning to the symbols used in human thought. There was also no tolerance for ambiguity and no appreciation for the kind of “likenesses” that give power to metaphor or help us identify members of the same family. Dreyfus condemned the failure to recognize such oversights as inexcusable and argued that AI was the least self-critical field in science.

Whatever the merits of that charge, there was no shortage of criticism for Dreyfus. Among others, Herbert Simon and Allen Newell—the Carnegie Mellon scientists who had come up with the first working AI program—lobbied Rand to suppress the report and nearly succeeded. It wasn’t published officially until 1967. Marvin Minsky and Seymour Papert, who were heading the AI program at MIT, launched another counterattack with a refutation by Papert titled The Artificial Intelligence of Hubert L. Dreyfus.

As support for their claims, Minsky, Papert, and the other champions of classical AI pointed to how well Eliza and similar “expert systems” were doing in the Turing test. That was something suggested in 1950 by famed mathematician Alan Turing, who first conceived what would become the modern computer. In his test, a human operator types questions that are answered by unseen correspondents in separate rooms. One of them is a person; one is a computer. When an operator can’t tell from the responses which is which, the computer has passed.

In the matrix of Cartesian and behaviorist notions then shaping AI, the Turing test was seen as ultimate justification. What came out of the box was all that mattered. The underlying architecture—how things worked inside the box—was beside the point.

Pressure mounted on Dreyfus, who was then teaching at MIT. He became isolated and eventually moved on to the University of California at Berkeley as classical AI’s stock kept rising. Throughout the 1960s and ’70s private corporations and the military poured enormous sums into the development of expert systems. It was believed that they would eventually dispense generations of accumulated wisdom and experience. But things didn’t work out that way. Instead, it soon became clear that they had to be upgraded whenever conditions changed, which was constantly. The problem was that they couldn’t learn on their own. As Weizenbaum and Dreyfus had tried to show, they didn’t know how to think.

Tags: AI, artificial intelligence, computer, Dreyfus, miltary, robotics, therapy, thought, Turing test, Weizenbaum

Email this to a friend | Add to del.icio.us | Add to digg |
Linkology by Kate (all links are unofficial) | Rights Info

Rating: 3.3, from 3 votes

Posted on May 12th, 2006 @ 3:00 pm