jim (12/08/82)
In reply to the argument that a mathematical theory of brain behavior
will not contribute to producing an intelligent machine, consider that,
instead of talking about brains and intelligence in 1982, we are talking about
plutonium and energy in 1932. If we had a suitable mathematical theory
describing the processes which go on in the nucleus of the atom, couldn't
we maybe take a 6kg. hunk of plutonium, take a conical 2kg hunk out of the
side, displace it away from the main hunk on a kind of gunlike device, and
set the whole thing up so that the gunlike device rams the 2kg hunk down
into the main hunk, releasing energy? The answer, in retrospect, is *yes*,
although the numbers may be off slightly. The point is that the brain is
no different from any other physical-biological system for which a suitable
basic science theory could be expected to yield applied science gains.
With regard to the comments on self-organizing systems, the brain is
obviously not just a random mess of neurons; in fact, there is probably
as much structure there as in any other highly organized system, like an
ecosystem, or even a Vax 11/780 for that matter. The question is how
that structure results in the observed brain states. I also don't find
the thought that the brain is "only" a collection of neurons any more
depressing than the fact that a 6kg. hunk of plutonium is "only" a collection
of atoms, except, possibly, for the thought of what such a collection of
neurons has chosen to do with the knowledge of how to use plutonium.
I agree with Serle that, until, Putnam, Fodor, Dennett and others advocating
a "knowledge level" can come up with evidence of such, the self-organizing
approach stands to gain more information on how intelligence works in the
only system where we can presently observe it.
I am not familiar with Minsky's book (is a more exact title available?)
or Harnad's article, and thank those users for drawing my attention to
them. I am skeptical, however, of people who try to model conciousness
or the brain with a finite state machine. An infinite state model is
required for a bucket of water heated from below by a Bunson burner,
and the brain is certainly more complex than a bucket of water. Being
a classical (ie. nonquantum) system, its states should properly be
mapped onto real numbers, and not some countable set, in my opinion.
This also makes me skeptical that a program running on a von Neumann
machine will ever be able to do more than simulate certain aspects of
intelligent behavior, and also answers the question of how a simulation
differs from reality (ie. in the bits lost by forcing a real number into
a finite word size). I will certainly check out the book and the article.
Since I seemed to have started this discussion, I guess I'd have
to weigh in for the creation of a new category, net.ai, though only
if a significant fraction of the user community wants it. How about
sending out a request for a vote to all sites?
Jim Kempf
...!arizona!jimmark (12/12/82)
Sorry, but net.ai has received only 7 favorable votes. It appears that the discussion belongs right where it is, in net.misc. Mark
lemmon (12/16/82)
a couple of questions:
1) how would you characterize the states of the brain? if you
have a theory that attempts to account for intelligence
in terms of brain states, it seems you must characterize
those states somehow.
2) does it make any difference which brain cells fire when?
if so, what difference? and isn't this different from the
plutonium? and if not, how can behavior be directed?
if the operation of the brain depends on the coordinated operation
of cells, then we must find some way of organizing the combinatorial
number of states (sorry, a finite number if there are enough
threshold effects operating) into equivalence classes.
maybe the "knowledge level" is just a way of doing this?
for me, i'll stick to knowledge.
alan lemmon
linus!lemmondebray (12/21/82)
Here's an eighth vote for a net.ai ! Saumya K. Debray SUNY @ Stony Brook
danb (01/04/83)
Here's a ninth vote for net.ai !
bts (01/05/83)
Yet another yes vote for net.ai
paul (01/07/83)
Another vote in favor of net.ai, which hopefully can cover other assorted goodies.
neiman (01/07/83)
Yes to net.ai!
seth (01/11/83)
#R:pegasus:-15700:hp-cvd:7600002:000:12 hp-cvd!seth Jan 10 10:11:00 1983 Aye to ai.