Newsgroups: comp.ai.philosophy,sci.philosophy.tech
Path: cantaloupe.srv.cs.cmu.edu!nntp.club.cc.cmu.edu!miner.usbm.gov!rsg1.er.usgs.gov!jobone!newsxfer.itd.umich.edu!gatech!news-feed-1.peachnet.edu!news.duke.edu!godot.cc.duq.edu!newsfeed.pitt.edu!ctc.com!news.mic.ucla.edu!library.ucla.edu!ucla-ma!oak!zeleny
From: zeleny@oak.math.ucla.edu (Michael Zeleny)
Subject: Re: Penrose's new book
Message-ID: <1994Oct21.210340.28435@math.ucla.edu>
Sender: news@math.ucla.edu
Organization: The Phallogocentric Cabal
References: <37v6ac$pa8@lyra.csx.cam.ac.uk> <1994Oct18.110514.16731@math.ucla.edu> <385oqf$b32@lyra.csx.cam.ac.uk>
Date: Fri, 21 Oct 94 21:03:40 GMT
Expires: December 31, 1994
Lines: 92
Xref: glinda.oz.cs.cmu.edu comp.ai.philosophy:21213 sci.philosophy.tech:16144

In article <385oqf$b32@lyra.csx.cam.ac.uk> 
mh10006@cl.cam.ac.uk (Mark Humphrys) writes:

>zeleny@oak.math.ucla.edu (Michael Zeleny) writes:

>>mh10006@cl.cam.ac.uk (Mark Humphrys) writes:

>>>Does Penrose deal with genetic algorithms, reinforcement learning,
>>>behavior-based AI, etc. in his new book?

>>Yes.

>>>He didn't seem aware that such things existed last time.

>>He does not have to be aware of their existence to deal with them.

>That's a great attitude to have to a field in which you aren't trained.

In this country, only soldiers and dogs get trained.  Scholars get
educated.  My friends at Oxford and Cambridge assure me that the same
is true of the UK.

>Read Holland or Goldberg on genetic algorithms, Koza on genetic
>programming, and maybe Langton on artificial life.
>Then re-read Penrose on 'NATURAL SELECTION OF ALGORITHMS?'.
>Embarrassing, isn't it?

Not at all.  However it must be truly embarrassing to lack the minimal
understanding of mathematics required to realize that Penrose's
argument, if true, admits of no exceptions, regardless of the
engineering technique.

>>>If I was interested in AI systems based on symbols and logic,
>>>I might (or might not) be interested in Godel's theorem and all that.
>>>But I can't see what it has to say about
>>>some mess of an evolved artificial neural net,
>>>undesigned, not guaranteed to be without errors and flaws,
>>>and about which nothing much can be said other than it worked better
>>>than the other messes.

>>If it can be modelled by mathematical means, it falls within the
>>purview of logic.  Which only goes to show that everything falls
>>within the purview of logic.

>>>How come popular critics of AI never seem to read in the above fields?

>>How come the artificial intelligentsia invariably lack awareness of logic?

>Logic is largely irrelevant to intelligence, consciousness
>and what makes us human in general. Get it straight - we evolved by natural 
>selection - we weren't designed, verified, or guaranteed to be consistent.
>No one ever contemplated or proved the validity of the algorithms
>that are us. There is no proof that our algorithms work, terminate, halt
>or are free from error. But so what? They helped us survive anyway!

Consider a total mathematical model of human performance -- guaranteed
to be consistent, for the simple reason that you cannot do and not do
P at once.

>And if you think such unreliable machines are funny things to do
>mathematics with, then you're right. Unfortunately, they're all we've got.

You really ought to think harder about such matters.

>>Suitable modifications of Goedel's argument can be promulgated in any
>>discipline whose language honors suitable closure principles.  There
>>is a neat example of one such application to inductive logic in Hilary
>>Putnam's latest collection of papers.  More to the point, it will of
>>necessity apply to the mathematical description of any mechanical
>>device whatsoever, regardless of its construction or provenance.

>OK then, does Godel's theorem show that an insect can do things
>that an evolved ANN can't? Please explain.

Evolution versus manufacture is a red herring.  Goedel -- by way of
Montague -- shows that if certain closure conditions obtain in L, then
it cannot be finitely axiomatized, with obvious implications for AI.
Arguments for closure from Platonism are likewise compelling.  Penrose
argues for the truth of Platonism, and in this way his argument is
valid, albeit not necessarily adequate.  Your option is to refuse
Platonism, which does not appear to admit of a conclusive proof or
refutation.  The best treatment of that option known to me, will be
found in Judson Webb's book on finitism.

>Mark Humphrys
>Computer Laboratory
>University of Cambridge

cordially,                                                    don't
mikhail zeleny@math.ucla.edu                                  tread
writing from the disneyland of formal philosophy                 on
"Le cul des femmes est monotone comme l'esprit des hommes."      me
