Talk:Artificial intelligence

For the benefit of the skeptics among us, wouldn't it be desirable to list some of the outstanding successes of AI so far? (other than beating Kasparov at chess, I'm not sure what those are; and even that I'm not sure counts as genuine AI) --Seb

There are much less such successes than average person would expect. There are some essayes about AI chatbots on http://www.alicebot.org/ (ALICE chatbot won "best chatbot" Loebner prize in 2000 and 2001, so author known what he is talking about), where author says that there was almost no progress in this field of AI since Eliza.

The most important things to notice:

we still don't have machine translation
we still don't have something that can do Turing Test better than Eliza
we still don't have automatic algorithm proving
no "important" math theorem was proved by automatic proving machine
computers almost universally are based on strictly deterministic algorithms, not heuristics

--Taw

"we still don't have machine translation": this is not really an argument. When people say they want to have machine translation, they usually mean "perfect machine translation". This however is unfair, because humans cannot translate perfectly either. I am pretty sure we can get near-perfect machine translation just using statistical methods, but I would not necessarily call that intelligence.--branko

Yanked from the article:

The philosophy of mathematics asks if adult human linguistic or symbolic intelligence is any more complex than mathematical proof itself, and whether this is in fact what humans mean when they recognize each other as being concious, wise, or aware. Turing's Test highlights these questions by suggesting that adult humans perhaps assume too much based on mere language:

Taking the second sentence first, that's a rather novel interpretation of the significance of the Turing Test, to say the least, and one that doesn't really fit with the reading of Turing's original paper IMHO. It might be a more reasonable response to some of the counterarguments raised about the Turing Test (notably the Chinese Room).

As to the first sentence, I can't parse it. --Robert Merkel

Fixed. The "mere language" issue is now illustrated by the apparent "human racism" of the theorists who reject Great Apes' intelligence. It's not that Turing's paper highlighted the issue of what matters in language but rather than Turing's Test itself did - by failing to convince people it was decisive.

The first sentence is also fixed, and refers more directly to the CHurch-Turing Thesis, which was the specific contribution, and which defined all forms of symbolic or linguistic intelligence as being equivalent to the Turing Machine, which in turn was equivalent to "mathematicians doing proofs in the usual way"

I don't know if you'd agree, but I think the original vision of AI has by now been thoroughly discredited by ethical problems. Creatures that satisfy the original definition of "intelligence", e.g. Great Apes, are not accorded the respect of personhood. Meanwhile, stupid programs of bad research continue to propagate themselves due to funding inertia and influence of top researchers like Minsky, who hvaen't produced anything worth a damn in years. Welcome to tenure, I guess, but the people building robot insects or hooking up humans into cyborg colonies (woops forget to mention Steve Mann under collective intelligence) or talking about augmenting apes witih speech synthesizers just don't believe any of the nonsense that Minsky believed...

Deception is the key difference between humans and Great Apes. Like four year old children, Great Apes do not have a "theory of mind" that enables them to lie *convincingly* by imagining how the other will think things are. Human children acquire this at four and a half or so. Great Apes never do. They lie very badly.

The thing called "intelligence" seems to me to be a combination of perception, planning, empathy, cognition and deception. One decides for oneself which to test for, and what to emphasize, and what to extinct. AI is a nonsense goal, just "infinite symbol manipulation" really, some of which symbols are maybe good enough maps to walk over rough terrain in a robot insect, but none of which are good enough to deceive a suspicious adult human.