Newsgroups: comp.ai
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!snorkelwacker.mit.edu!news.media.mit.edu!media-lab.media.mit.edu!minsky
From: minsky@media-lab.media.mit.edu (Marvin Minsky)
Subject: Re: THE I OF THE BEHOLDER
Message-ID: <1991Jun14.194519.28465@news.media.mit.edu>
Sender: news@news.media.mit.edu (USENET News System)
Organization: MIT Media Laboratory
References: <25410@samsung.samsung.com> <1991Jun14.151446.20698@hawk.cs.ukans.edu>
Date: Fri, 14 Jun 1991 19:45:19 GMT

In article <1991Jun14.151446.20698@hawk.cs.ukans.edu> spratt@hawk.cs.ukans.edu (Lindsey Spratt) writes:
>
>Was Minsky's lambast about "growing" beyond the "childish, primitive,
>religious" concerns for truth meant to include non-classical logical
>and philosophical approaches such as situation theory and semantics?

That was a pretty childish remark of mine.  I do think that the term
"truth" has too many uses to be usable in technical discussions.  Yes,
we can all agree that Mickey is not a real mouse -- and if we want,
we can go on to say things like "'Mickey is not a real mouse' is
true."  But this is like pretending, in biology, that "alive" has a
meaning because we agree that tigers are alive and stones are not.
The  important and practical problems always come in the fuzzy areas.
I summarized my grumble with formal logic recently in an article in
the curret AI magazine, and another one in the current Artificial
INtelligence journal.  I'll just quote from the first one:

These limitations of logic begin at the very foundation, with the
basic connectives and quantifiers.  The trouble is that worldly
statements of the form, ``For all $X$, $P(X)$,'' are never beyond
suspicion.  To be sure, such a statement can indeed be universally
valid inside a mathematical realm---but this is because such realms,
themselves, are based on expressions of those very kinds.  The use of
such formalisms in AI have led most researchers to seek ``truth'' and
universal ``validity'' to the virtual exclusion of ``practical'' or
``interesting''---as though nothing would do except certainty.  Now,
that is acceptable in mathematics (wherein we ourselves define the
worlds in which we solve problems) but, when it comes to reality,
there is little advantage in demanding inferential perfection, when
there is no guarantee even that our assumptions will always be
correct. Logic theorists seem to have forgotten that any expression
like ($\forall X$)($PX$), in actual life---that is, in a world which
we find, but don't make---must be seen as only a convenient
abbreviation for something more like this:

   "For any thing $X$ being considered in the current context, the
assertion $P{X}$ is likely to be useful for achieving goals like $G$,
provided that we apply in conjunction with certain heuristically
appropriate inference methods."

In other words, we cannot ask our problem-solving systems to be
absolutely perfect, or even consistent; we can only hope that they
will grow increasingly better than blind search at generating,
justifying, supporting, rejecting, modifying, and developing
``evidence'' for new hypotheses.

------

TO be more constructive, what I'm saying is that there is something
wrong with the "basic" concept of instantiation -- and generalization.
Instead of doing "deductions" which require that sort of idea, I'm
saying that thinking is more centered around plausible matching,
concepts of similarity, heuristics of modifying descrptions, and so
forth.  My complaint is that the two dominant ideas of logic -- and I
just don't know whether the situation is different in "situational"
theories -- of *instantiation* and of *universal (or existential)
quantification* are rather low on my list of useful devices for cognition.
