From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wupost!darwin.sura.net!gatech!pitt!speedy.cs.pitt.edu!geb Tue Nov 26 12:32:48 EST 1991
Article 1619 of comp.ai.philosophy:
Xref: newshub.ccs.yorku.ca rec.arts.books:10687 sci.philosophy.tech:1138 comp.ai.philosophy:1619
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wupost!darwin.sura.net!gatech!pitt!speedy.cs.pitt.edu!geb
>From: geb@speedy.cs.pitt.edu (Gordon Banks)
Newsgroups: rec.arts.books,sci.philosophy.tech,comp.ai.philosophy
Subject: Re: Daniel Dennett
Message-ID: <12504@pitt.UUCP>
Date: 25 Nov 91 16:31:30 GMT
References: <1991Nov21.005355.5696@husc3.harvard.edu> <centaur.690849720@cc.gatech.edu> <1991Nov23.022628.5799@husc3.harvard.edu>
Sender: news@cs.pitt.edu
Reply-To: geb@cs.pitt.edu (Gordon Banks)
Followup-To: rec.arts.books
Organization: Computer Science Dept., University of Pittsburgh
Lines: 34

In article <1991Nov23.022628.5799@husc3.harvard.edu> zeleny@zariski.harvard.edu (Mikhail Zeleny) writes:
>Since you chose to elide and ignore my semantical arguments, I feel
>justified in doing the same to your AI cheerleading, noting only that I am
>familiar enough with most of the programs you mention, to the extent of
>knowing their very real failures.  As an example, I offer you visual
>pattern recognition.  Show me a program that can do the thing nearly every
>normal human child can: reliably recognize a human face.  No "experimental"
>designs will be accepted: I want the same reliability as manifested by
>humans.  All contenders will be tested by a bona fide expert in visual
>pattern recognition and measured against a randomly selected 5-year-old.
>Put up, or shut up.
>

Let's not pretend this is a trivial task.  The human brain devotes
dedicated hardware to this particular task, which when knocked out
can leave a person unable to recognize faces (aprosopagnosia).
We don't know exactly
the size of the neural network involved, but it likely contains a few
million neurons.  I don't think this is something do-able at the present
state of the art at the same level a 5 year old would.  However, it is
very unlikely that any symbolic processing is involved in this task with
humans, and thus a conventional AI approach would not be a good choice
to attack this.  Certainly we have neural network programs that do
simpler variations of this.  Arbib has one that mimics a frog and
can recognize enemies, prey, and other frogs.  One of my students made
one that can recognize and diagnose visual field defects.  I expect
that within a few years we will have the horsepower to attack such
tasks.

--
----------------------------------------------------------------------------
Gordon Banks  N3JXP        | "When in danger, or in doubt
geb@cadre.dsl.pitt.edu     |  Run in circles, scream and shout" --Heinlein
----------------------------------------------------------------------------


