Glacial Erratics

I've just returned from a lecture given by DouglasHofstadter? here at IndianaUniversity. The title was "Can Computers Understand Language? A ten-year booster shot against Searliomyelitis."    (2LF)

Back in 1980 JohnSearle? published a paper, "Minds, Brains and Programs", describing the now famous ChineseRoom? problem. Hofstadter believes the paper is full of errors, wrong, and akin to a virus in the way it infects otherwise right-minded folk with poor thinking.    (2LG)

Prior to the lecture I assumed Doug was going to give us a highly controversial refutation of Searle's thesis. I was prepared for an event--a painful inoculation. I didn't get that: at the end of the presentation simply asked that Searle and his cohorts lighten up and make room in their world for the simple idea that understanding of meaning exists not on an absolute black and white scale, but instead is a continuum on which progress--in the realm of computers understanding language--can be made: even if that progress thus far has been tiny.    (2LH)

I can get behind that. However, I think Hofstadter made some generalizations that were convenient for his argument when not explored but disruptive when considered more deeply.    (2LI)

When we ask if a computer is understanding language, we are asking more than whether it understands what we've said. A computer can give back reasonable responses to queries, so some kind of understanding is going on. The real question is: Do we believe there is meaning inside the machine. Are the symbols being manipulated by the computer related to "real things" with "real meaning".    (2LJ)

Hofstadter asks the same question of humans. When a human uses symbols when are they associated with "real meaning"? If a person uses baseball idioms without ever playing baseball, do they have a right to use the idiom? People use the idioms and we understand them when they do, so meaning is transmitted in some fashion.    (2LK)

I think Hofstadter misses the important question here. The question is not whether the idioms have meaning, but rather how do we know when a particular idiom is the right one to use? How do we judge or interpret the context in which we exist when we are communicating?    (2LL)

Hofstadter seemed to imply, although he may not have intended to do so, that children are programmed in much the same way as computers. Children, he said, know the difference between abstract and literal because we tell them. An audience member disagreed with this, saying there was far more subjectivity involved in the human child's judgment when compared with the computer.    (2LM)

The primary thrust of the criticism of Searle is that Searle's model of understanding is absolute, black and white and thus smacks of religious sacred-cowness. Hofstadter's alternative is a scale he calls semanticity from no understand to complete understanding. This is reasonable but it too has degree of absolutism. It assumes that there is only one dimension of understanding, and one ultimate peak of understanding which is the right one. We should certainly strive to improve understanding, and getting away from sacred cows is a good step, but let's not build another in the process?    (2LN)

Finally, throughout the presentation I was constantly reminded of what I consider to be a very important distinction: Behavior that is based on rules, such as programs, no matter how complex, can be decoded, eventually, by stepping backwards (perhaps in multiple dimensions) through the rules that were used to generate the behavior. It may be that the behavior can't be repeated, because of randomness in the system, but it still can be described by what amounts to programs. It's an article of apparent faith in some circles of the AI and/or cognitive science worlds that the complex behavior of the human brain is theoretically describable. I find it advantageous to disagree with this, not because I'm attached to meat as the only possible source of real intelligence but because believing it makes us emphasize the wrong problems, which I hope someday to describe at ClassesVersusCategories?.    (2LO)

Literary theorist should read more cognitive science. And vice versa. There's religion on both sides.    (2LP)