Path: utzoo!yunexus!geac!syntron!jtsv16!uunet!seismo!sundc!pitstop!sun!amdcad!ames!ncar!gatech!rutgers!columbia!cs!andy
From: andy@cs.columbia.edu (Andy Lowry)
Newsgroups: comp.ai
Subject: Re: Limits of AI
Keywords: Intelligence
Message-ID: <20@cs.columbia.edu>
Date: 30 Oct 88 21:35:41 GMT
Article-I.D.: cs.20
References: <1651@ndsuvax.UUCP>
Reply-To: andy@cs.columbia.edu (Andy Lowry)
Followup-To: comp.ai
Organization: Columbia University Department of Computer Science
Lines: 52

In article <1651@ndsuvax.UUCP> ncthangi@ndsuvax.UUCP (sam r. thangiah ) writes:
>One of the students in my class raised a point that:
>"Man is not capable of producing a machine that is more intelligent than
>oneself".  Is this a valid statement?

I certainly don't think so.  Humankind has created machines that are
stronger, faster, more precise, more rugged, and "better" than humans
in many other respects.  Why should "more intelligent" be magically
excluded from the list of possibilities?  The only reasons I can think
of are: (1) "Intelligence" is not well defined and is difficult to
measure, and as long as this situation continues, it will always be
easy to discount any machine behavior that somebody calls intelligent;
(2) Our intelligence is something we generally hold very sacred, and
it makes some people extremely uncomfortable to contemplate the
possibility that we might not have an exclusive claim to it.

Intelligence is clearly not something that one either does or does not
possess.  Some people are more intelligent than others; some species
of animal are more intelligent than others.  To assume that humankind
possesses the highest attainable level of intelligence, just because
we have not encountered more intelligence elsewhere in the animal
kingdom (or thus far in the machine "kingdom"), seems an extremely
arrogant attitude.  And to propose that humankind is incapable of
creating machines that exceed human levels of intelligence runs
counter to our experience in countless other domains.

Here's a more radical proposition: Suppose we manage to design and
build mechanisms for learning, perception, abstraction, etc. that are
superior to our own.  Will we then build a bunch of machines that will
be our mental superiors?  How about an alternative: we apply what we
have learned to "re-engineer" the human mind.  How?  I don't know.
Genetic engineering?  21st century magic?  22nd century magic?
Whatever...  But what would we call the results?  Human beings or
machines?  Is this scenario scary?  Is it more palatable than
coexisting with mentally superior machines?  Is all this so hard to
take that we should stop trying to push machine intelligence?  Can we
ignore these problems and stop worrying because the student quoted
above is correct and the situation is inherently inconceivable?

My own views: we can and will fabricate intelligence levels exceeding
our own.  What we learn (and what our creations learn) about the
mechanisms of intelligence will enable us to improve human
performance, though without altering the human physiology, barriers
will be met.  (Analogy: what we have learned about mechanics, human
physiology, nutrition, etc. has allowed us to push the performance
level of athletes, but it is inconceivable that a human runner will
ever break the sound barrier or a strongman lift ten tons.)  The
evolutionary processes that result in physiological changes allowing
greater intelligence will come much more slowly than our ability to
build intelligent machines.

-Andy