Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!samsung!uunet!mcsun!ukc!edcastle!aipna!cam
From: cam@aipna.ed.ac.uk (Chris Malcolm)
Newsgroups: comp.ai.philosophy
Subject: Re: AI - the real problem
Message-ID: <4085@aipna.ed.ac.uk>
Date: 1 Mar 91 19:17:18 GMT
References: <1473@ucl-cs.uucp> <23398@well.sf.ca.us>
Reply-To: cam@aipna.ed.ac.uk (Chris Malcolm)
Organization: Dept of AI, Edinburgh University, UK.
Lines: 62

In article <23398@well.sf.ca.us> nagle@well.sf.ca.us (John Nagle) writes:
>G.Joly@cs.ucl.ac.uk (Gordon Joly) writes:

>>There was a suggestion in the AI Journal that 2.5 million years of
>>humanity compared with the rest of life on earth meant that simulating
>>the IQ of an earwig or a lizard meant that you were almost home and
>>dry, and that human IQ was within spitting distance.

>     Reference, please.

Not the requested reference, but others relevant.

Moravec, H.,{\em Locomotion, Vision, and Intelligence}, in Robotics
Research 1, eds Brady and Paul, MIT Press, 1984.

This says it quite explicitly, and suggests that the proper basis for
an AI research programme is therefore to start with a simple
responsive mobile creature and then recapitulate the "ox-trail of
evolution".

See MIT AI MEMO 899, R. Brooks, "Achieving Artificial Intelligence
through building robots", 1986, which suggests just such a research
programme.

AUTHOR: Evans , Christopher  1931-1979
TITLE: The mighty micro  : the impact of the computer revolution
EDITION: New ed.,2nd impr.
IMPRINT: London Victor Gollancz c1982

This earlier popular book gives the same sort of numbers and arguments
which form the basis of Moravec's recommendation.

But I am sure that the general idea has been part of the
systems/cybernetics culture for as long as enough has been known to do
the sums, i.e., I think it's at least as old as AI, but I'm sorry, a
rapid rummage of my bookshelves has failed to turn up a reference.

The very earliest robotics research was along the kind of lines
suggested by these arguments, e.g., the MH1 MIT hand of the late 1950s
early 1960s. The later successes of the analytic approach, e.g., the
position-controlled robot arm with inverse-kinematics giving Cartesian
end-effector position specification, and the use of inverse Newtonian
optics to deduce object position from stereo images -- these successes
(IMHO) tempted research away from what then was a barely-articulated
paradigm into the "classical approach", i.e., which performs
"sensor-fusion" via world models, and in which the behaviour of the
system is made to conform to the formal model by using the formal model
as an ingredient in the implementation. The funding crisis of the 1970s,
when AI researchers were criticised for being "unscientific", i.e.,
"where's the formal model then?" helped the shift along.

"How do you know which leg to move next?" asked the ant of the
centipede. The centipede thought for a long time, started waving its
legs experimentally, and fell over.




-- 
Chris Malcolm    cam@uk.ac.ed.aipna   +44 (0)31 667 1011 x2550
Department of Artificial Intelligence,    Edinburgh University
5 Forrest Hill, Edinburgh, EH1 2QL, UK                DoD #205