Path: utzoo!attcan!uunet!mcsun!ukc!edcastle!aipna!cam From: cam@aipna.ed.ac.uk (Chris Malcolm) Newsgroups: comp.ai Subject: Re: What AI is exactly. Message-ID: <3030@aipna.ed.ac.uk> Date: 15 Sep 90 17:20:31 GMT References: <25392@boulder.Colorado.EDU> <4123@servax0.essex.ac.uk> <3640@gara.une.oz.au> <3853@se-sd.SanDiego.NCR.COM> Reply-To: cam@aipna.ed.ac.uk (Chris Malcolm) Organization: Dept of AI, Edinburgh University, UK. Lines: 45 In article <3853@se-sd.SanDiego.NCR.COM> jim@se-sd.SanDiego.NCR.COM (Jim Ruehlin, Cognitologist domesticus) writes: >If you don't know how it works, how can you say it's intelligent? Well, I don't know how I work, so I'm not intelligent. Nor is Jim Ruehlin, or anyone else, unless there's some recent breakthrough in cognitive psychology or neurobilogy I don't know about :-) That's the strong form of the intelligence-is-method argument. It is more commonly found in its weaker form, which takes humans as being intelligent by definition. Then all one hs to do to demonstrate that the latest AI toy is not intelligent is to show that no matter how well it performs, it doesn't do it quite the same way that we do. Now we know that technology gives us many ways of doing the same thing. One can fly like a bird, or like an aeroplane. One can tell the time with a clockwork or digital watch. One can search for problem solutions forwards or backwards. So let us suppose, for the sake of argument, that my complete mental capabilities have been implemented in some other technology than biological, using other methods. This has given rise to certain minor differences in performance, for example, my simulacrum is faster at mental arithmetic than me but a bit wobblier on a bike, but these differences are within the normal variability of human performance. But, since how it's done is crucial, I am intelligent, but my simulacrum is not, and the research effort (and success) of building the simulacrum has not advanced our understanding of intelligence at all. Is this is useful position to adopt? I suspect that the popularity of the how-it-works argument comes from knowing that intelligence is not easily recognised. Even if intelligence can be defined completely in termw of behaviour, in practice it would be impossible to observe enough behaviour to be really sure, just as in practice on can never test a complex program completely. So in practice the attribution of intelligence depends on lots of presumptions about unobserved behavioural capabilities. But there is another way of finding out how something would behave: prediction from knowing how it works. That's very useful in any complex system which is understood; but it doesn't mean that how it works is more important than what it does: how it works is a way of getting a handle on what is important -- what it does. -- Chris Malcolm cam@uk.ac.ed.aipna 031 667 1011 x2550 Department of Artificial Intelligence, Edinburgh University 5 Forrest Hill, Edinburgh, EH1 2QL, UK