The turing test is a sack of crap. I don't recall where I read this, but somebody put it this way (I"m of course para-phrasing) :

we should be focusing on what AI accomplishes, not how it looks doing it and who it fools in the process. We achieve flight with machines that in no way resemble those things that fly in nature (birds, flies, bees, etc).

er...something like that. Of course the original author could articulate much better than myself, but you get the point. I feel what you're saying though, Bert. I just wanted to throw my two cents in about Turing.

http://www.cogs.susx.ac.uk/users/blayw/tt.html
explains that point of view.

Of course Turing did a lot more for computing than coming up with the Turing test.

Alex

How would you propose testing adaptive reasoning? A game of chess ? The Voigt Kampff Empathy test ?

The fact that the test has worked until now to effectively discern natural language from natural language processing perhaps only means that Turing was clever but might also mean that his test was well specified.

Bert

[disclaimer : my words are based less on fact than on opionion.]

Relativly. The whole perspective on AI is too general (currently). We're looking to AI to be just like bob (general name, no real finger-pointing). But that, like you said, is pretty far off. The Turing test could prove useful (but not an absolute standard) for AI progs designed to be human-like (physchatrist (sp?), AI bot who talks about sex, any expert system ...), but we shouldn't [yet] expect an AI prog to be able to talk about anything we can.

If I hire a robot housemaid (like in the Jetsons) I want [him = (him/her)] to know all about his job. I want him to be able to smell dirt in the air and suck it up before it hits the ground. I don't care if he is able to tell me I'm hooking up my new mobo wrong (well, bad example, that might come in handy, but you get the point). Surely my botmaid will fail the Turing test. But so what? The botmaid has some very human-like qualities, but the whole reason I hired him for the job is because most of his qualities are not human (won't get tired, won't get grossed out, won't sell my secrets to ad companies).

So I find the Turing test to be irrelavant (I don't hate the Turing test, and I only called it "a sack of crap" just to state that I don't find it relative) and this is mainly due to AI, while posseing many human-like qualites, is often chosen because it is NOT human, and will NOT go crazy from doing the same thing over and over (of course, we all do that, but that's another topic).

Back to your orignal question, Bert :
"How would you propose testing adaptive reasoning?"

I think it's realitive. I know you had a hint of sarcasm, but chess is a good example - if you're making a chessbot. I'm unfamiliar with "Voigt Kampff Empathy test". Sorry, I don't feel like looking it up. But yes, if it's a less narrow topic than "emulating a human", that's a good measure of adaptive reasoning.

..... continuation of above post (NS6 sucks when there are more than 500 chars in a textbox)

"The fact that the test has worked until now to effectively discern natural language from natural language processing perhaps only means that Turing was clever but might also mean that his test was well specified."

Well, you'll not hear be dispute the cliam that Turing was clever. But you'll forgive me if I fail to see how you were able to come to the conclusion that "his test was well specified.". I can't tell if I"m just being slow, or if I really don't get it. In either case, can you rephrase that (or dumb it down)?

The test was well chosen - in answer to the question "How do you test intelligence?" Turing could have devised any number of tests - writing a love letter, reading the news, or a game of poker.

If a human cannot tell whether he is communicating with another human or a machine, the machine can be said to be artificially intelligent.

Do you think intelligence can be accurately measured? Would we be able to scientifically discern an intelligent being? If you think so, then I can understand your lack of faith in the Turing test. But most people don't think it's possible and neither does the scientific community.

So the Turing test is like a wall of unspecified height, which no-one has managed to jump yet so (and because) no-one knows what's needed to jump it . That's why it's a good test.

Hope this explains a little better.

Bert

Seems a little arrogent to compare intelligence to how we are. But it could also be argued that we really have nothing else to compare it to. Because the definition of intelligence is relative, I would venture to say it is of the same importance (with AI, not in general) as being knowledgable. Making desisions in good taste and relative to the situation can be a suitable definition for intelligence for the field of AI. Having said these things, I don't feel that we should be comparing AI to us [humans], and should consider it an entity in it's own right. Have we never met another being that shows signs of "intelligence"?

Well, yes, but then we would need to get a dolphin or a chimp or an alien to say whether the computer shows intelligence and they're only interested in fish, bananas or making crop circles respectively.

Bert

or write programs that are intrested only in these things

Lackluster, I apologize if I offended you. I was speaking from my own VERY light experience with PERL and legacy code. Shoulda known better. I would be curious to see what sort of AI systems could be produced in it, though, and learning about its OO capabilities (which I thought were nonexistent at the time).

A friend today sent me this interview with the creator of PERL, in case you've not seen it on slashdot yet. http://interviews.slashdot.org/artic...mode=nocomment

You didn't offend me, I just hold PERL in very high reguard. Thanks for the link, btw. I didn't know Larry Wall was Christan. He was the first person in 10 years to make me question my lack of religion. But then I wised up to his scheme .