Joho the Blog
An Entry from the Archives

Thanks for the links. I actually talk in "Small Pieces" about Searle's Chinese Room and a Hofstadter essay in "The Mind's I" (which I reviewed lo these many years ago for the Philadelphia Inquirer). But I hadn't seen Searle's comments on Kurzweil's book. I'll blog the reference. Thanks again.

It's gotta be 10 years since I had any kind of close look at AI, but what I remember is that we tried to build inference engines based on expert frameworks. Rules were only one way to apporach this. At that time, with 486 processors and a max of 128 MB of memory, the whole thing seemed problematic. Maybe it's time for someone who actually worked on LISP or one of those projects to dust it off and try it out on a current platform. In translation programming, it seems like inferences could be drawn based on associative rule sets and simple word pairings.

This means something, but I'm having a hard time translating it from english into english.

Hello David: Excuse my silliness (much lack of sleep), but when I read this post I was thinking of the old anecdote about a program that was being tested to translate English into Russian. "The spirit is willing, but the flesh is weak" was translated into "The vodka is free, but the meat is rotten." -BK

I just want to make sure that the Canadian aspect to this story doesn't get overlooked. Couldn't the choice of the Hansard actually have a great deal to do with the success of this program?

Pinkers 'Words and rules' is worth reading on this. He showed using PET scns thta different parts fo the brain are used for regular and irregular verbs - rules for regular; look-up tables for irregular.

Association and rules are complementary

Would that work for Chinese?

Sylvia Wong also gives some background to machine translation techniques, including some discussion of the Candide System with the modeling background for support - if anyone is interested.

The sound bite is "Every time I fire a linguist the performance of the recognizer goes up" - reportedly said by Fred Jelinek of the IBM speech group - as quoted in Jurafsky and Martin, 'Speech and Language Processing'.

The question of whether or not this leaves a door open for AI is to me unimportant. The only accomplishment of Searle's argument is to preserve an illusion of the superiority of the human mind.
The much more interesting question of whether machines are capable of performative superiority relative to human beings is completely unadressed.
Searle's argument therefore not only is irrelevant to the development of technology like AI (which it is, it strongly argues that NO capability of the machine is relevant in the argument).
It actively obscures important discussion on the possibility and power of complex machines.

Claus, I disagree with your reading of Searle. The question that you state is unimportant is precisely the one that Searle is addressing, so it's unfair to complain that he doesn't focus on another question, especially since you (IMO) mischaracterize his opinion on that other question.

He is not arguing that the human mind is superior. He agrees that a $1.95 calculator calculates better than humans do and doesn't deny that Deep Blue can beat the best human players. He is disputing, however, that Deep Blue is conscious or that it understands that it's playing chess. I.e., Searle is arguing against strong AI. And I think it is a convincing argument.

You are right of course. Superiority is my reading of why the argument is put forward with such energy in the first place. But my real criticism of Searle remains:
If the discussion of the possibility of AI becomes a discussion of consciousness and if we accept Searle's definition of that term then the outcome of that discussion has no bearing on the development of the kind of technology being developed under the heading of AI (strong or otherwise). The only conclusion one can draw from Searle's argument is whether or not one can apply the word consciousness to AI.

Searle very specifically argues that the observable capabilities of a complex machine is irrelevent in judging about the possible consciousness of the machine. Searle's argument is NOT that the Turing test cannot be passed, but that it does not matter to the question of strong AI.
While the discussion on consciousness is intellectually important, it is in my opinion also derailing a necessary discussion about a possible technological future. The consequences (practical or ethical) of constructing machines that are observably as sophisticated as the human mind are in no way diminished by insisting that these are 'mere' machines.