Flashback to 1975

From: Michael Wilson (mwdestinystar@yahoo.co.uk)
Date: Sun Jul 17 2005 - 04:55:29 MDT

I've just read John Anderson's 'Language, Thought and Memory'; light reading
for a weekend break. It has the highest concentration of GOFAI fallacies I
have seen yet in a single book, which is pretty impressive after Newell's
'Unified Theories of Cognition'. There's a lot of black humour in retrospect,
but in particular I would like to share the following paragraph with you:

'Despite the fact that it is largely lacking in demonstration programs,
Norman and Rumelhart assert that their computer system has great potential.
That is, the representation, parsing schemes, interpretative schemes, etc.
are so structured that, given sufficient investment of programming time, they
could form that basis for a very powerful language system. Such claims for a
system's potential rather than for a system's actual performance are
frequently made in the artificial intelligence (AI) and carry considerable
weight in the AI community. It is argued that it would be a waste of
artificial intelligence researcher's potential to require them to spend the
years of hack programming required to translate potential into reality. This
would be an eminently reasonable argument if there were objective criteria
for evaluating the potential of a system other than its actual performance as
a concrete system. But there are no such objective criteria.'

Almost as relevant today as thirty years ago; while contemporary academics
are perhaps a little more prepared to actually build things, the fraction
that progresses beyond nontrivial toy examples is very small. While the
situation is perhaps rather better in the overall space of AGI projects, it
is unfortunate that while some have realised the importance of working
implementations, virtually no-one has been working on the problem of
objective criteria for evaluating AGI theories. Anderson wastes a couple of
chapters formally analysing the Turing-equivalence of various models, and
seems to consider that chunk of logic (along with formal specification of his
program) adequate to permit the rest of his design characterisation to read
as a collection of hunches, guesses and wild extrapolations. Alas in this
regard his work is still compares quite favourably with most contemporary

 * Michael Wilson

Yahoo! Messenger - NEW crystal clear PC to PC calling worldwide with voicemail http://uk.messenger.yahoo.com

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:51 MDT