From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!think.com!wupost!micro-heart-of-gold.mit.edu!news.media.mit.edu!minsky Thu Dec 26 23:57:55 EST 1991
Article 2347 of comp.ai.philosophy:
Newsgroups: comp.ai.philosophy
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!think.com!wupost!micro-heart-of-gold.mit.edu!news.media.mit.edu!minsky
>From: minsky@media.mit.edu (Marvin Minsky)
Subject: Silicon Neurons (was: In the news (re: Searle's response to silicon brain?))
Message-ID: <1991Dec21.155741.16304@news.media.mit.edu>
Sender: news@news.media.mit.edu (USENET News System)
Cc: minsky
Organization: MIT Media Laboratory
References: <40858@dime.cs.umass.edu> <45183@mimsy.umd.edu> <709@ckgp.UUCP>
Date: Sat, 21 Dec 1991 15:57:41 GMT
Lines: 63

In article <709@ckgp.UUCP> thomas@ckgp.UUCP (Michael Thomas) writes:
>David Harwood writes:
>>         The "neurochip" is an analog device, using very
>> little power compared to conventional designs, and apparently
>> simulates some internal electrical properties of neurons,
>> and adaptively spikes output.
>>        The article concludes with an assessment by Marvin
>> Minsky, "I don't think this tells us anything we didn't
>> already know," and, "To me it's such a long jump to the
>> human brain that this isn't interesting. For one thing,
>> there are several hundred kinds of neurons. Which one's
>> this supposed to be?"

>Yes it is true, that there are several hundred different
>"classifications" of neurons. Another point was overlooked though,
>neurons are like snowflakes, no two are alike! None of mine match up
>with any other of mine or anyone elses... The connections between neurons

 [Etc., with several remarks I agree with]

>NOTE: I tried not to direct this towards Marvin Minsky since he could
>      have been miss-quoted.
>
>================================================================================
>Thank you,            ||  "Sol est invisiblis in hominibus, in terra vero
>Michael Thomas        ||   visibilis, tamen ex uno et eodem sole sunt ambo"
>(..uunet!ckgp!thomas) ||                    -- Theatrum Chemicum (Ursel, 1602)


Thanks.  I really appreciate your considering that possibility.  After
all these years I'm still not used to worrying about being taken too
seriously.  And in fact it was me who screwed up here and not the
reporter -- and I wish I could cancel/retract those remarks. 

Yes, the information was uncooked at my end.  The reporter called me
up to ask about the new Caltech Silicon neurons.  I had not heard
about them and was trying to be helpful, but I simply forgot that I
was talking to a reporter!  I haven't read the paper and was reacting
to the reporter's summary.

Actually, I haven't seen the quoted remarks attributed to me, but they
sound more or less accurate.  But I don't think they gave the
impression I intended, because I was *npt* responding to a question
like "is this an important breakthrough in building artificial
brains?"  I was mostly answering a different question -- and a good
one -- that the reporter asked.  The question, more or less was, "is
it true that analogue simulation on a chip is orders of magnitude
faster than digital simulation"?  And my answer was, more or less,
"yes, if most of the functionality can in fact be modelled mostly by linear 
processes together with nonlinear ones that can  be suitably realized 
in the semiconductor context.  And Carver Mead and his associates 
are outstandingly good at that.

The quoted remarks appear to be reponses to the first question -- and
the answer to that should have been, something like, "it is not very
interesting unless the models can be put into (variable) networks that
can simulate heuristically useful kinds of adaptations" -- just as
Harwood observes.
  -- Marvin Minsky






