From newshub.ccs.yorku.ca!torn!utcsri!rutgers!jvnc.net!yale.edu!yale!gumby!wupost!zaphod.mps.ohio-state.edu!cs.utexas.edu!uunet!mcsun!uknet!qmw-dcs!abreu Tue Jul 28 09:41:23 EDT 1992
Article 6451 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!torn!utcsri!rutgers!jvnc.net!yale.edu!yale!gumby!wupost!zaphod.mps.ohio-state.edu!cs.utexas.edu!uunet!mcsun!uknet!qmw-dcs!abreu
>From: abreu@dcs.qmw.ac.uk (Abreu)
Newsgroups: comp.ai.philosophy
Subject: Re: Defining other intelligence out of existence
Message-ID: <1992Jul15.013626.24984@dcs.qmw.ac.uk>
Date: 15 Jul 92 01:36:26 GMT
References: <1992Jul8.092458.3088@otago.ac.nz>
Sender: usenet@dcs.qmw.ac.uk (Usenet News System)
Organization: Computer Science Dept, QMW, University of London
Lines: 88
Nntp-Posting-Host: it118.dcs.qmw.ac.uk

In <BILL.92Jul13114604@cortex.nsma.arizona.edu> bill@nsma.arizona.edu (Bill 
Skaggs) writes:

 > abreu@dcs.qmw.ac.uk (Abreu) writes:
 >   The first step in measuring intelligence is defining it. The second, is
 >   devising ways to quantify it. We all think we know what intelligence is,  
 >   can't quite put our finger on it, but know it when we see it.

 > I'm not convinced that it is impossible to give a decent descriptive
 > definition of intelligence.  I think we can get pretty close to the
 > mark by defining intelligence as the *ability to solve problems*.

Hmmm, a bit limited though. I'm not familiarised with Alan Newell's work
but I will be soon. Meanwhile, describing the ability to solve problems
as pretty close to the mark is pushing it a bit, I think.

I'd say that the ability to solve problems is a RESULT of intelligence,
and certainly that there's more to intelligence than logic. If we look
at human production in the fields of literature, art, music, (you get
the picture) I think we'd all agree that they require intelligence, but
they're not exactly what we'd call problem-solving. Obviously, we can
look at every single area of human endeavour from the point of view
of problem-solving. However, I'd say that such an approach is hardly
close to the mark.

On the other hand, EVEN if we accept such a definition, the following
doesn't convince me: 

 > A "problem", in turn, is defined by a system (consisting of states and
 > allowed transitions between states), an initial state, and a set of
 > goal states; a "solution" is a set of transitions taking the system
 > from the initial state to one of the goal states.
 [...]
 > Intelligence, in this formulation, is the ability to search a space of
 > transition-sequences.

No way on Earth. As you rightly say, the concept of "intelligence as
search" is the essence of the traditional approach to AI. In my opinion
that's why AI has produced respectable results in some limited areas
(game-playing being the most obvious). The problem inherent to such an
approach is the size of the transition-sequences space, or, more
specifically, the ability of an intelligent being to expand or restrict
this space. Now, few spaces could be more restricted than game-playing,
and even with state-of-the-art hardware (although Von-Neuman archite-
ctures), state-of-the-art tree searching algorithms heavily optimised
for performance, table-based evaluation functions, all kinds of
heuristical knowledge that can be integrated into such systems, we still
can't beat the best human players.

One fascinating thing about game-playing in particular, is that machines
can and do beat the best human players in Othello, whereas they can't do
the same in Chess (yet). Although infinitely more time has been spent in
studying Chess than Othello, we must consider that Othello is inherently
simpler and therefore has a more restricted space. The larger the
transitions-sequences space is, the more problems a space-searching
approach to intelligence has.

If we keep in mind the size of the space in Chess and compare that with
other spaces, it will be dwarfed to an infinitesimal size. Thus the
shift from Situational to General intelligence. The chances that a 
problem-solving space-searching approach to intelligence will be able to
cope with such spaces should be reduced at least in direct proportion to
the dwarfing (INSTINCT(?!!) tells me that the reduction is probably
heavier than that).

 [...]
 > Understanding the nature of human intelligence means understanding the
 > kinds of problems we are capable of solving and the strategies we use
 > to solve them.  The same applies to alien intelligence, or for that
 > matter to artificial intelligence.

Debatable, in the least! I don't mean to say that problem solving is
negligible to a definition of intelligence; maybe it does have a relevant
part to play. I do mean to say, though, that there must be much more to
intelligence than problem-solving. Our knowledge of the parallel processes
occuring in the brain is still very, very limited. Maybe even, some
time remotely distant in the future, we'll discover that the brain is
nothing but a hyped-up implementation of an alpha-beta search. I don't
think so, but who the hell knows... :-)

Hamilton


PS: As you might have noticed if you read my last message, the last
remaining bug has been ironed out from this otherwise immaculate
piece of software, that you, lower forms of life, may addre
AI_LeftHem_DecodeReference: out of space - infinite regression at &a2e3ffff
core dumped


