REPOST: Minsky's "Programs, Emotions and Common Sense"
![[repost because of rogue cancel]'s profile photo](https://lh3.googleusercontent.com/a/default-user=s40-c)
[repost because of rogue cancel]
On technetcast.com, there is an intriguing talk given by Marvin Minsky
as a preview of his forthcoming book. Minsky gives us a few reasons
why it is year 2001 and we still don't have HAL yet, along with his
hints on how we might achieve it.
Aside from his everlasting wit, Minsky delivers an applaudable
presentation; I have literally clapped him in front of the screen. His
criticism of the fallacies of popular AI is something so reminiscient
of our not-so-popular discussions at Bilkent CS dept., that I'd like
to comment on them. He states that there hasn't been much progress in
common sense or general intelligence since mid 1970's.(And yes, his
criticisms start with a bashing of FOL, so don't miss it) Neural
networks and genetic algorithms have been so intensely advertised that
most people nowadays assume that these are the state-of-the-art in AI.
If there would be a HAL, it would certainly be built with such
methods. Personally, I think this stems from the desire to have a
magic solution to a very hard problem without having any understanding
of it and in reality they do not solve majority of hard problems. As
Minsky suggests, there are kinds of problems that ANN are good for,
however he criticizes genetic algorithms harshly: how can we expect
that a superstitious worship of imitating organic evolution will work
out for us while it has bugs and takes millions of years to run?
Indeed, what genetic algo. people do is solving mediocre problems with
today's powerful machines in a worse way than traditional heuristic
search.
The part about statistical methods is a big hit. Although his opinions
are a bit too strong on this, he claims that using vectors of numbers
in a learning is an intellectual dead end. However, the kinds of
failures he cites are definitely realistics. You can have 90% accuracy
in, say machine translation, but the remaining 10% might be those that
require some deep thought. Think about speech recognition for another
instance; your language model might cover 95%, but the remaining 5%
depends on interaction with semantics and pragmatics, which shows that
a large percentage of accuracy does not necessarily mean that you have
solved a large part of the problem.
Without suggesting improvements, these criticisms would have looked
superfluous, and thus Minsky gives us a a glimpse of his next book
"The Emotion Machine".
He goes on to talk about consciousness which has "baffled so many
people, especially physicists", and expresses his belief that "there
is no so such thing". I agree with him. I don't think that there is a
thing such as 'subjective experience' or 'qualia' which is just a
romantic way of saying "I have no idea how any of these cognitive
phenomena happen". Consciousness, Minsky says, is a suitcase word "we
use as a name for a dozen very hard problems about how the brain or
the mind works". So, he does have a theory for solving "consciousness"
in his new book, which seems to me improving on the ideas of 'Society
of Mind'. He refers to two of his theories in particular; one which is
an architecture of mind, and "multiple representation". He does not
dwell much on the architecture, except by emphasizing that the human
mind is a highly evolved and complex distributed architecture and that
his architecture has five layers "including theaters which are places
where you simulate in one way or another what you think might happen
if you were to do something" with no central control. He seems to have
been inspired mostly by the architecture of human brain and he seems
to have been developing on ideas from 'The Society of Mind'.
Minsky touts multiple representation as a new way of looking at AI. He
begins by indicating the philosophical problems with a single
representation: if you understand something in one way then you won't
have understood it at all in another way. He indicates that we falsely
assume there is one definition instead of a network of interrelated
processes. When we would like an open-ended representation we should
be able to maintain all of them, and have them co-operate. Minsky
bears a diagram which partitions problems according to the scale of
their causes and effects and suggests that differing representations
work best for respective parts of the diagram. However, how we should
go about making NNs, statistical methods, logic, classical ai, etc. is
not very clear. He also indicates that he offers alternative
representations (network of k-lines from Society of Mind, etc).
If you haven't read or watched this talk, I suggest you to do so :) It
is as thought provocative and beneficially controversial. Comments are
welcome.
Regards,
__
Eray Ozkural
[ comp.ai is moderated. To submit, just post and be patient, or if ]
[ that fails mail your article to <com...@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]
[reposted because of rogue cancel]
Regards,
__
Eray Ozkural
========= WAS CANCELLED BY =======:
Path: news.sol.net!spool0-nwblwi.newsops.execpc.com!spool0-chcgil.newsops.execpc.com!newspump.sol.net!europa.netcrusader.net!203.248.240.126!newsfeed.dacom.co.kr!news.unitel.co.kr!usenet
From: y...@akibeorxpks.gov
Newsgroups: alt.test,comp.ai
Subject: cmsg cancel <9euqq5$ino$1...@mulga.cs.mu.OZ.AU>
Control: cancel <9euqq5$ino$1...@mulga.cs.mu.OZ.AU>
Date: 30 May 2001 01:11:48 GMT
Organization: Unitel and AT&T
Lines: 56
Sender: er...@bilkent.edu.tr (Eray Ozkural exa)
Approved: a...@cs.mu.oz.au
Message-ID: <3aqiq3$qqx$3...@mulga.cs.mu.OZ.AU>
NNTP-Posting-Host: dup-od731-215.ukr.net
This group is evil
Otthwam mtgai didm sioar strtnset thc gscsvsvo ee
nnoe naetr ct izc on us ii
ftvyeo ase eeaie efrdoi yt rhth rtio cssila ctnrt orr
ltvyrp ltutef lprdwt i reltcra trlr gne fivr
otpwor woo dmy pbae eynrpn y rtpcd onenx otao etd tharc
<remainder snipped>
[repost because of rogue cancel]
er...@bilkent.edu.tr (Eray Ozkural exa) wrote in message news:<9euqq5$ino$1...@mulga.cs.mu.OZ.AU>...
(that was in comp.ai)
> Without suggesting improvements, these criticisms would have looked
superfluous, and thus Minsky gives us a a glimpse of his next book
"The Emotion Machine".
I really appreciate Eray's excellent review of my lecture. I don't
think I could have condensed it so well. ( I think this was the talk
I gave last month at the Games Development Conference in San Jose,
Calif.)
For more details, I've just put a draft of the first half of my book,
"The Emotion Machine," on my web page at
http://www.media.mit.edu/people/minsky/
It is still pretty rough, and I would appreciate all comments on
details: either about parts that are hard to understand, or parts that
seem just plain wrong.
(Incidentally, a lot of sections were inspired by discussions on
comp.ai and comp.ai.philosophy. Tell me if I quoted you without
mentioning the source.)
[repost because of rogue cancel]
On technetcast.com, there is an intriguing talk given by Marvin Minsky
Without suggesting improvements, these criticisms would have looked
superfluous, and thus Minsky gives us a a glimpse of his next book
"The Emotion Machine".
He goes on to talk about consciousness which has "baffled so many
Regards,
__
Eray Ozkural
[ comp.ai is moderated. To submit, just post and be patient, or if ]
Eray Ozkural exa
> er...@bilkent.edu.tr (Eray Ozkural exa) wrote in message news:<9euqq5$ino$1...@mulga.cs.mu.OZ.AU>...
>
> think I could have condensed it so well. ( I think this was the talk
> I gave last month at the Games Development Conference in San Jose,
> Calif.)
>
It was a pleasure. It is indeed the talk at Game Developers Conference
that technetcast is broadcasting. I must say that I had enjoyed
reading "The Society of Mind" which kept me full of thoughts for a
long time. I'd written a humble review of the book for an AI course
I'd taken from Varol Akman, which states that the book attempts to
"answer how a mind can work by presenting his [your] theory that a
mind is composed of smaller processes which are called agents" and
that you "intend to construct a wonderful toy with tiny bits and
strings of a very humane nature" and finished by remarking that "a
review can identify only the surface of such a brilliant book" and
that the book succeeds in presenting a "mechanistic framework for a
theory of mind", that it "contains reasonable explanations for
phenomena that are otherwise most difficult to consider"; that "Minsky
is well aware that some of the ideas would work, and some others would
be very difficult to realize" and "evidence from simple common sense,
ordinary facts of human mind, unambiguous psychology and elementary
biology, together with his great insight make his opinions truly
strong"; "Accessible by anyone who is interested, it speaks an
exciting tone. The proposed theory is sure to cause proliferation of
thinking and research on cognition."
It's certainly one of the most influential books I've ever read. I had
been thinking how your ideas could be extended and I am delighted to
see that you are working on a sequel.
> For more details, I've just put a draft of the first half of my book,
> "The Emotion Machine," on my web page at
>
> http://www.media.mit.edu/people/minsky/
>
> It is still pretty rough, and I would appreciate all comments on
> details: either about parts that are hard to understand, or parts that
> seem just plain wrong.
>
I will definitely have to read this draft. :)
Regards,
__
Eray Ozkural <er...@cs.bilkent.edu.tr>
[repost because of rogue cancel]
er...@bilkent.edu.tr (Eray Ozkural exa) wrote in message news:<9euqq5$ino$1...@mulga.cs.mu.OZ.AU>...
(that was in comp.ai)
> Without suggesting improvements, these criticisms would have looked
superfluous, and thus Minsky gives us a a glimpse of his next book
"The Emotion Machine".
I really appreciate Eray's excellent review of my lecture. I don't
think I could have condensed it so well. ( I think this was the talk
I gave last month at the Games Development Conference in San Jose,
Calif.)
For more details, I've just put a draft of the first half of my book,
"The Emotion Machine," on my web page at
http://www.media.mit.edu/people/minsky/
It is still pretty rough, and I would appreciate all comments on
details: either about parts that are hard to understand, or parts that
seem just plain wrong.
(Incidentally, a lot of sections were inspired by discussions on
comp.ai and comp.ai.philosophy. Tell me if I quoted you without
mentioning the source.)
[ comp.ai is moderated. To submit, just post and be patient, or if ]