|
MINSKY'S "THE SOCIETY OF MIND"
An Outsider's Brief Commentary
Marty Stoneman
Introduction
It's probably not usually fair to critically comment on a ten year old book
-- but this book seems, among the experts, to be surviving quite well. So here goes, at least for a few brief starter comments. By all rights, "The Society Of Mind" should be commented on in the style in which it was written, i.e., with about three hundred one-page comments very loosely organized -- as a "society" of "small ideas" -- and then state that when the reader "joins" enough of them -- poof! -- the "strangest mysteries" of this commentary will be revealed. Unfortunately, no writer with less than the high intellectual reputation of MIT's Marvin Minsky could possibly persuade a reader to jump into such a morass. So I'll be somewhat organized and topical even though each comment is about widely-separate parts of "The Society Of Mind". Furthermore, were I merely to compliment the book, that might not be interesting, so ---
"Complications" From Evolution, Etc.
Overall, Minsky (who is enough of a respected celebrity to be spoken of in
surname alone) takes the position that a (human?) mind must be very
complicated -- primarily because of all of evolution's accidental steps,
birds in the hand, side journeys, etc., in getting to more and more of
"mind". Further, he seems to feel that any intelligent machine entity will
share these complications and will have to work and be organized like
minds/brains -- or at least deal with the same kinds of complications as
intelligent biology. As an outsider, I wonder. Just because evolution took a
roundabout way to discover what kinds of systems work for intelligence, does
that mean that today, at a machine computational systems level, we can't
essentially do the main jobs in more straightforward ways than evolved
biology? In the same vein, Minsky's position toward psychology (as a field)
is that any overarching theoretical understanding will come from eventual
combining of "hundreds" of smaller theories -- as opposed to the possibility
of an early overarching theoretical "breakthrough" kind of model from which
many smaller models may be tested and refined. Again, I wonder. Great leaps
of understanding in science are often totally unforeseen -- why not here?
And provided machines can understand us in some overarching ways, they may
be able to get along with us even if their way of doing intelligence is not
available to us.
Pain And Pleasure
Professor Minsky's Zen-like approach to pain and pleasure ("simplified")
seems naive at best -- and certainly serves the purpose of avoiding talk
about structure or simulation or flowcharts or anything else useful to AI.
And his general talk about pain's "narrowing of interests" escapes essential
discussion of that most important of human pains -- call it "people pain" --
that deep pain behind the great majority of human suicides -- and the pain
ruling our lives most of the time in the form of shame, loneliness,
disloyalty, rejection, etc., etc., etc. And that, to an outsider, is quite
an omission.
Defining A Tiger
I am somewhat puzzled by Minsky's apparent abhorrence at defining something
too early or too .... whatever. Quoting, "You can know what a tiger is
without defining it. You may define a tiger, yet know scarcely anything
about it." It seems to me, as an outsider, that if we can begin to define
"tiger" using the sorts of representation and primitives and computation
enabling a deer or monkey or human to recognize and predict about a typical
tiger, that would be a good start. It might also take new ideas not in this
book.
Memory, Learning, And Episodes
Minsky seems to have adopted one of the major mental blocks of mainstream
psychology (excepting perhaps of evolutionary psychology) in his confusing
comments in the areas of "memory", "learning", and "episodic memory".
Certainly, if my computer is fouling up, I can add/modify some
data/programming to "teach" it to get it right. And then users of my
computer will know that it now acts differently (as though it had learned
something). As an outsider, it seems obvious to me that the new ways my
computer is now behaving have absolutely nothing to do with whether it has
stored anything about the time or circumstances of when its behavior changed
-- or about some historical "story" about its learning experiences.
Bothering with such storage and taking useful space up to do it seems
wasteful and not best for the computer. So why would a lizard or mammal
brain bother to do anything but learn to change behavior? If there is a good
answer to this, what is it? Or if anyone thinks they can show that a turtle
or rabbit "remembers" the sequential episodes of its "childhood", clue me in
with that evidence. Of course, in a social entity which constantly trades
stories with others of its kind, it is worth a very high amount of effort to
have an "episodic memory" of sorts -- usually of a poor sort. But nowhere
does Minsky argue that his views about learning and memory apply only to
languaging humans and not to rabbits or apes or other intelligent entities
-- and why on earth should he?
Cause, Explanation, Blame, Etc.
Most scientists might agree that the idea of "correlation" is a central one
in science, not the idea of "cause", which concerns most non-scientists
constantly. Since human concern with causation, with "what-came-before" (all
the way to big bangs and/or religious stories) is so universal, I expected,
as an outsider, that Minsky would point to possible brain/mind structures
which might be associated with this universal need. And I was disappointed
that, except for a hint or two, this important topic was not dealt with. It
seems to this outsider that there are identifiable links between our animal
needs for predictability, the way minds work, and our human preoccupation
with "cause". Will our intelligent entities be built to have the same
preoccupation? And, if not, how do we avoid it? And, if we avoid it, will
they understand us and our preoccupation enough to communicate empathically
with us?
Finally
Minsky, a brilliant and legitimate pioneer in Artificial Intelligence, is in
constant danger of also becoming a "guru" in the sense of the weight of his
authority becoming something semi-permanent. As an outsider, I hope, as in a
well-known phrase, that we need not sit at his feet but can climb on his
shoulders.
© 1996 Anthrobotics™
|
