From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!jupiter!morgan.ucs.mun.ca!nstn.ns.ca!aunro!ukma!wupost!darwin.sura.net!gatech!psuvax1!hsdndev!husc-news.harvard.edu!zariski!zeleny Tue Nov 26 12:32:35 EST 1991
Article 1603 of comp.ai.philosophy:
Xref: newshub.ccs.yorku.ca rec.arts.books:10669 sci.philosophy.tech:1126 comp.ai.philosophy:1603
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!jupiter!morgan.ucs.mun.ca!nstn.ns.ca!aunro!ukma!wupost!darwin.sura.net!gatech!psuvax1!hsdndev!husc-news.harvard.edu!zariski!zeleny
>From: zeleny@zariski.harvard.edu (Mikhail Zeleny)
Newsgroups: rec.arts.books,sci.philosophy.tech,comp.ai.philosophy
Subject: Re: The Philosophical Foibles of John McCarthy
Message-ID: <1991Nov25.180643.5898@husc3.harvard.edu>
Date: 25 Nov 91 23:06:38 GMT
References: <1991Nov15.003438.11323@grebyn.com> <1991Nov15.160741.5495@husc3.harvard.edu> <JMC.91Nov24203029@SAIL.Stanford.EDU>
Organization: Dada
Lines: 216
Nntp-Posting-Host: zariski.harvard.edu

In article <JMC.91Nov24203029@SAIL.Stanford.EDU> 
jmc@SAIL.Stanford.EDU (John McCarthy) writes:

>In-Reply-To: zeleny@zariski.harvard.edu's message of 21 Nov 91 19:53:45 GMT

JMC:
>I have reread Zeleny's foibles posting several times.  I follow it
>(and agree) up to the point where he accuses me of prestidigitation.
>Then it gets a bit fuzzy.  I'm not sure what question I am supposed to
>have begged.  Circumscription has been treated syntactically by
>writing a certain second order formulas.  It is also treated
>semantically by talking about models of the axiom that are minimal in
>a certain ordering.  Given not too onerous conditions the syntactic
>and semantic treatments correspond nicely.

Allow me to reiterate the relevant parts of my argument, expanding 
where necessary.

Consider the following: your stated goal in the theory's development is
to formalize a certain ontological assumption, i.e. that "the objects
that can be shown to follow to have a certain property P by reasoning
from certain facts A are all the objects that satisfy P" (cited from
your 1980 article in "Artificial Intelligence"); however, in order to
capture this assumption, which you recently (see the article
<JMC.91Oct27133840@SAIL.Stanford.EDU>) characterized as "a kind of
formalization of Occam's razor, no more entities of a certain kind
exist than those whose existence the axioms imply", writing $A(\Phi)$
for the result of replacing all occurrences of $P$ in $A$ by the
predicate expression $\Phi$, you introduce as ``the circumscription of
$P$ in $A(P)$ [...] the sentence schema $$A(\Phi) \land \forall
\overline{x} \dot (\Phi(\overline{x}) \supset P(\overline{x})) \supset
\forall \overline{x} \dot (P(\overline{x}) \supset \Phi(\overline{x}))
,$$'' interpreting the condition ``as asserting that the only tuples
$(\overline{x})$ that satisfy $P$ are those that have to -- assuming
the sentence $A$.'' (op. cit.)

Now, the specious bit of prestidigitation represented in the above
claim, is your silent transition from talking about a property P and
certain facts A to the (extensional) first-order predicate P and a
sentence A.  Assuming the Quine--Church criterion of ontological
commitment (see the summary and references in R. Eberle, "Nominalistic
Systems", Dordrecht:  Reidel, 1970; see also the discussion in S.Haack,
"Philosophy of Logics", Cambridge: Cambridge University Press, 1978),
it is evident that, in passing from intensional entities like
properties and observables like facts, to extensional entities like
predicates and syntactical ones like sentences, respectively, the
ontology is already implicitly circumscribed by the choice of language,
with its variable ranges and implicit cardinality limitations; hence
the question of ontological commitment is begged; moreover, the sort of
circumscription that gets performed by formulating the axiomatization
is, by far, the most powerful kind. Consider first-order ZFC; is it
reasonable to assume that no more sets exist than those whose existence
the axioms imply? in other words, is Cohen's generic model to be
considered the right model for ZFC? On the other hand, the
circumcsription for first-order Peano Arithmetic is not even
expressible in a nicely recursive way.  So how would your machines deal
with the standard model of the integers?

Of course, you could argue that under the formalist assumption that a
theory is reducible to its syntax, you haven't done anything wrong in
tacitly dispensing with the semantical considerations; however, on this
view you aren't entitled to talk about facts and objects at all.  Facts
and objects exist before all theoretical descriptions thereof, and
semantics is the only bridge between the latter and the former; ignore
it, and your grand theory is reduced to a mere *flatus vocis*.  For
regardless of your philosophical prejudices, it pays to feign realism
when trying to construct an artificial intelligence; as I recall, you
already noted something similar in an earlier publication.

There exist even more fundamental problems implicit in your approach,
in that it relies on a simplistic application of Occam's razor.  As
noted Church in his 1951 article, "The Need for Abstract Entities In
Semantic Analysis" (Proceedings of the American Academy of Arts and
Sciences 60: 100--112, reprinted in Martinich, ed, "The Philosophy of
Language", New York: Oxford University Press, 1985), not every
subtraction from the entities that a theory assumes is a reduction in
the variety of entities, and, more importantly, the need of ontological
parsimony should be balanced by the need of the workability,
simplicity, and generality of the theory; were you to succeed in
formalizing a principle stipulating the former, there's no reason to
suppose that the latter would lend themselves to easier formalization
than the much more restricted general problem of algebraic
simplification, currently considered to be intractable.

JMC:
>Please avoid Latin in arguing with me; I don't know it.  No German
>either.  You may use French or Russian if you must.  Also I would be
>grateful for a full reference to the Church article.  Circumscription
>is not an application of Occam's razor; rather it is a formalism in
>which some applications of Occam's razor can be expressed.

My own Latin is rather minimal, so I use it sparingly, and translate
nearly all I cite.  On the other hand, `mouthfart' doesn't sound nearly
as good as `flatus vocis'.

JMC:
>As for salvaging my credibility, I am complacent about it.  Quine,
>Dennett and several other philosophers have been quite friendly.

Of course, Quine's own credibility tends to vary depending on the field
of inquiry; I don't know of a single mathematician that would take him
seriously; as for modal logicians and semanticists, they tend to view
him more as an object of exorcism than a constructive thinker.  As for
Dennett, the less said, the better.

JMC:
>It occurs to me that Michael Zeleny has a future as an anti-AI
>philosopher.  He has apparently read rather little in AI, especially
>in the area of formalizing common sense and nonmonotonic reasoning,
>but this little is enormously more than Hubert Dreyfuss, Sir James
>Lighthill, John Searle, Hilary Putnam, Gian-Carlo Rota, Roger Penrose
>and James Fetzer bothered to read, judging from the bibliographies
>associated with their pronouncements.  Apparently they have such
>strong intuitions that AI must be nonsense that they feel no need to
>read papers before writing.  There is a need, however, because they
>have not succeeded in transferring their bare intuitions to many of
>their fellow philosophers and scientists or even many of the literati
>who write reviews.

On the contrary, I believe that AI has a radiant future ahead of it,
especially after it gets itself a proper name, like `technonoetics'.
Briefly, I see the situation as similar to that of alchemy and astrology:
given a few centuries of pompous pronouncements, and numerous calving of
successful derivative disciplines which, embarrassed by their shameful
parentage, will invariably choose alternative, descriptive names (remember
my earlier suggestions: `proairetics', `hypoleptics', and `phronetics',
Greek words for deliberate choice, judgment, and practical wisdom,
respectively), occupying themselves with limited, and hence useful tasks of
modeling various aspects of the mind, the grand old dame will comfortably
settle into her anecdotage, occupying herself with parlor tricks and
publication of penny sheets.

JMC:
>If Zeleny could bring himself to read more papers, he could easily
>become the most important philosophical critic of AI.  He wouldn't
>even have to say anything sensible.

If I could get myself on their publishers' charity lists, I most
certainly would; as the matters stand, my budget is limited to books
with yellow covers.  Furthermore, I believe myself to be far too young to
take leave of my senses; if and when I get to be your age, all bets would
be off.

JMC:
>Finally, Zeleny, like his fellows considers the forty years of AI
>research to be completely unuccessful.  My opinion is that progress
>has been made in this difficult area.  Like the problems of biology,
>it may take hundreds of years.  My further opinion is that progress
>is being made by many people along the logical lines proposed in my
>1959 paper, "Programs with Common Sense" reprinted in my 1990
>book "Formalizing Common Sense" published by Ablex.  To evaluate
>that, it would be necessary to read quite a few papers.  However,
>fundamental conceptual problems remain.

In fact, I rather like things like Prolog, circumscription,
non-monotonic reasoning, image recognition, computer algebra, pattern
matching, and so on; however I like them in the same way I like the
elephant: they are nice to look at, but I wouldn't want to take them
home with me.  I believe that a researcher in computer science is much
better off studying the semantics of programming languages (cf. Scott's
Denotational Semantics, and better, intensional approaches like
Girard's "Proofs and Types") than vainly trying to simulate natural
languages.  As for your own research, to the extent that I am familiar
with it, I believe it to be worthwhile as an attempt to formalize
practical reasoning, and in fact stated so earlier.  What I don't like
is the crass commercialism and fantastic hype one often finds
associated with the claims made by purveyors of expert systems,
inference engines, and reasoning tools; fortunately, they seem to be on
their way out.

So think of me as located on the other side of Church's Thesis, emphasizing
fortuitous, rather than effective computability.  You, on the other hand,
seem to choose siding with your predecessors McCulloch and Pitts, with
their semantical finiteness assumption already implicit in the quaint title
of their original AI manifesto, "A Logical Calculus of the Ideas Immanent
in Nervous Activity", their explicit, specious, unjustified identification
of "computability by an organism" with Turing computability, their
counterfactual claims contradicting the preceding: "every net, if furnished
with a tape [...] can compute only such numbers as can a Turing machine,
[...and] each of the latter numbers can be computed by such a net" (where
does the tape come from?), and, finally, their monumental misreading of
Church's Thesis by conflating computability and effective computability:
"This is of interest as affordind a psychological justification of the
Turing definition of computability and its equivalents, Church's
$\lambda$-definability and Kleene's primitive recursiveness: if any number
can be computed by an organism, it is computable by those definitions, and
conversely." (See Boden's Philosophy of AI anthology, page 37.)  In spite
of all the flaws of Penrose's argument, I'm more inclined to trust his (and
Church's) intuitions than this sort of discombobulated reductionist
sloganeering.

JMC:
>In a BBC debate with Professor Lighthill, I tried to make an analogy
>saying, "Physicists haven't solved the problems of turbulence in 100
>years and aren't giving up".  I was flabbergasted by Lighthill's
>reply, "They should give up".  Unfortunately, the BBC didn't
>include this exchange, which served to calibrate Sir James's
>attitude, in the tape they broadcast.

One may safely conclude that Sir James is a pompous ass; one wouldn't
be right in concluding that every AI critic is one.

'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`
`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'
: Qu'est-ce qui est bien?  Qu'est-ce qui est laid?         Harvard   :
: Qu'est-ce qui est grand, fort, faible...                 doesn't   :
: Connais pas! Connais pas!                                 think    :
:                                                             so     :
: Mikhail Zeleny                                                     :
: 872 Massachusetts Ave., Apt. 707                                   :
: Cambridge, Massachusetts 02139                                     :
: (617) 661-8151                                                     :
: email zeleny@zariski.harvard.edu or zeleny@HUMA1.BITNET            :
:                                                                    :
'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`
`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'`'


