Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!mit-eddie!media-lab!minsky
From: minsky@media-lab.MEDIA.MIT.EDU (Marvin Minsky)
Newsgroups: comp.ai.philosophy
Subject: Re: emergence
Message-ID: <3619@media-lab.MEDIA.MIT.EDU>
Date: 7 Oct 90 22:28:15 GMT
References: <1990Oct4.173933.7319@ux1.cso.uiuc.edu> <6@tdatirv.UUCP> <8712@milton.u.washington.edu> <8724@milton.u.washington.edu>
Reply-To: minsky@media-lab.media.mit.edu (Marvin Minsky)
Organization: MIT Media Lab, Cambridge MA
Lines: 29

In article <8724@milton.u.washington.edu> wcalvin@milton.u.washington.edu (William Calvin) writes:

>GENES NEED ONLY be approximately correct, as a little behavioral
>versatility can do the rest. 

Your essay looks profound, and I can't wait to see the book.  

A point somewhat like this was made by Jeff Hinton, not long ago: he
pointed out that a small genetic "hint", together with some learning
mechanisms, can speed up evolution enormously. 

I think his argument went something like this: Suppose some 40 binary
accident were needed to accompish some amazing performance, and you
happened to possess 20 of them, by genetic chance.  This could easily
happen in a population substantially greater than a million.  Now,
suppose eacg such creature lives long enough to try 2^20 behavioral
variants -- and if the miraculous 1/2^40 event occurs, it gets a
selective reproductive advantage.  Then !!!!! we have selection for a
one-in-a-trillion conbination.  

Then, quite rapidly, perhaps, that sub-population will accumulate
genes that further predispose them to now need only 2^19, then 2^18,
etc., learning trials.  When that gets down to, say, 2^10, we have
something that might occur regularly, in the first few hours or days
of infancy -- and then there might be no selective advantage
in further genetic fixation.

Anyway, that's what I think Hinton was suggesting.  Sounds like you
also noticed something of that sort.