From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!sun-barr!ames!elroy.jpl.nasa.gov!usc!rpi!ispd-newsserver!psinntp!scylla!daryl Wed Feb 26 12:54:15 EST 1992
Article 3986 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!sun-barr!ames!elroy.jpl.nasa.gov!usc!rpi!ispd-newsserver!psinntp!scylla!daryl
>From: daryl@oracorp.com
Newsgroups: comp.ai.philosophy
Subject: Re: Strong AI and panpsychism
Message-ID: <1992Feb24.211220.6854@oracorp.com>
Date: 24 Feb 92 21:12:20 GMT
Organization: ORA Corporation
Lines: 65

Stanley Friesen writes (about why a table lookup program is not intelligent):

> I see it as a matter of dealing with contingency. Precomputed
> responses are only good for dealing with *anticipated* situations,
> on-the-fly responses are capable of dealing with *unanticipated*
> situations, and I see this as one of the core capabilities of an
> intelligent entity.

The point of the table lookup was that in principle (although not in
practice, for obvious reasons) *all* situations a human may find
himself in can be anticipated. The argument goes as follows:

   1. Humans have finite ability to discriminate sense information.

This means that for any particular sense, say hearing, humans cannot
tell the difference between analog signals and sufficiently
fine-resolution digital signals.

   2. Therefore, humans only receive a finite number of bits of information
      about the world every second.

The highest number of bits per second occurs in vision, but there it
is still finite.

   3. Humans generally live less than 100 years.

   4. Therefore, in his lifetime there are only a finite number of
      possible situations in which a human can find himself.

This is a subtle point. Of course there is the possibility that two
different situations are the same for the purposes of sensory data
produced. But if that is the case, there is no way for the human to
decide which situation he is in. His (or her) responses cannot depend
on the situation if he doesn't *know* the situation.

   5. If a table contains appropriate responses to all these situations,
      then it will impossible to surprise it with something unexpected
      that must be dealt with "on the fly".

> And even here this can be looked on as a sort of efficiency, by
> delaying decisions until needed you reduce the amount of system
> capacity needed to store information (thus, likely, making the system
> both instantiable and general - otherwise physical limitations will
> force it to be specialized).

I agree (I think). Physical limitations on available memory or on
processor speed will force any practical intelligent system to some
mechanism other than table lookup; it would have to perform-replace
some "on the fly" computations. However, to me the efficiency or
practicality of the program says more about the intelligence of the
programmer than of the program.

> In short the table lookup scheme is not intelligent because it is
> incapable of dealing with unanticipated input (except by 'I don't
> understand' type responses). Just because all inputs it will ever
> actually recieve have been anticipated, this does not change its lack
> of flexibility.

By the way the table was defined, there are no unanticipated inputs.
The table includes responses to every possible input sequence.

Daryl McCullough
ORA Corp.
Ithaca, NY
 


