|
If Two Minds Are Better Than One, Then How About Two Thousand?
A Review of Marvin Minsky's The Society
of Mind
Every once in a while, one encounters a book so strikingly fertile and
original that after reading it, one can never be quite the same. Its ideas
resonate, sounding depths in one’s everyday life, clarifying what was previously
obscure and raising questions that one never even knew existed. Such a
book is Marvin Minsky’s The Society of Mind. In this book, Minsky,
arguably the foremost researcher in the field of artificial intelligence,
shares with lay audiences ideas developed through many decades of careful
thinking about how minds, natural and artificial, might work. The late
Carl Sagan, a friend of Minsky’s, often lamented the fact that first-rate
scientists rarely condescend to explain their work to nonspecialists. We
can be greatful that Minsky saw the value in letting those of us who are
not specialists in AI and robotics have a glimpse of his work.
The cornerstone of Minsky’s theory is the conception of minds
as collections of enormous numbers of semi-autonomous, intricately connected
agents that are themselves mindless. As Minsky puts it,
This
book tries to explain how minds work. How can intelligence emerge from
nonintelligence? To answer that, we’ll show that you can build a mind from
many little parts, each mindless by itself.
I’ll call “Society of Mind” this scheme in which each mind is made of
many smaller processes. These we’ll call agents. Each mental agent by itself
can only do some simple thing that needs no mind or thought at all. Yet
when we join these agents in societies—in certain very special ways—this
leads to intelligence. (17)
Just as the mind, according to Minsky, is made up of many tiny agents,
so Minsky’s book is made up of many tiny, interrelated chapters. A typical
chapter begins by describing a phenomenon, such as young children’s tendency
to draw people as circular heads from which sticklike arms and legs protrude.
The chapter then explains, generally, an internal mechanism that might
account for such a phenomenon:
We’ll suppose
that the child does not have anything like a picture in mind, but only
some network of relationships that various “features” must satisfy. For
example, a child’s “person-drawing” feature-network might consist of the
following features and relations:
HEAD Large closed figure.
EYES Two circles, high in head.
MOUTH Object centered below eyes.
BODY Large closed figure.
ARMS Two lines, attached high on body.
LEGS Two lines, attached low on body.
To convert this description into an actual drawing, the child must employ
some sort of “drawing procedure.” Here’s one in which the process simply
works its way down the feature list, like a little computer program:
1. Consider the next feature on this list.
2. IF such a feature is already drawn, go to step 3. Otherwise draw
it.
3. IF list is finished, stop. Otherwise, go back to step 1.
When the child starts to draw, the first item on the list is “large closed
figure.” Sincere there isn’t any such thing yet, the child draws one: that’s
the head. Next the eyes and mouth get drawn. But then, when it comes to
drawing the body feature, step 2 of the procedure finds that a “large closed
figure” has already been drawn. Accordingly, nothing new is required, and
the procedure simply advances to step 3. (135)
This explanation of children’s drawings is typical of the kind to be found
in Minsky’s book. The
explanation is not meant to be a final, definitive statement about what
actually happens in the brain but rather shows how one might explain the
phenomenon from an engineering point of view. As one might expect of a
man who pioneered AI and made seminal contributions to robotics, Minsky
looks at a mental phenomenon and then asks himself what sort of mechanisms
in the mind could account for it.
Minsky is at pains to point out that much of what we take for granted
is really incredibly complex. Early
AI researchers soon learned that it was far easier to model on computers
the kind of sophisticated, conscious knowledge that experts have in a given
field than it is to model simple, largely unconscious abilities that we
take for granted, such as remembering how to get to our homes, balancing
ourselves on a bicycle, recognizing a parent’s face, or using language.
As Minsky puts it, “Easy things are hard.(29)” Minsky points out that in
the 1960s, he and Seymour Papert spent several years developing a program
that would have the ability to build towers out of blocks. Such a Builder
agent is itself made up of subagents such as Begin, Add, and End. Add is
made up of the subagents Find, Get, and Put. Get is made up of the subagents
Grasp and Move. All these mindless agents are connected in bureaucracies,
in which agents report to other agents, and most of this work occurs below
the level of consciousness, which is why it is so difficult to think clearly
about them. As Minsky puts it,
Consider just the seemingly simple problem of not reusing blocks already
built into the tower. To a person, this seems simple common sense: “Don’t
use an object to satisfy a new goal if that object is already involved
in accomplishing a prior goal.” No one knows exactly how human minds do
this. Clearly we learn from experience to recognize the situations in which
difficulties are likely to occur, and when we’re older we learn to plan
ahead to avoid such conflicts. But since we cannot be sure what will
work, we must learn policies for dealing with uncertainty. Which strategies
are best to try, and which will avoid the worst mistakes? Thousands and,
perhaps, millions of little processes must be involved in how we anticipate,
imagine, plan, predict, and prevent—and yet all this proceeds so automatically
that we regard it as “ordinary common sense.” But if thinking is so complicated,
what makes it seem so simple? At first it may seem incredible that our
minds could use such intricate machinery and yet be unaware of it.
In general, we’re least aware of what our
minds do best. (29)
In the course of his book, Minsky addresses many fascinating questions
in light of his Society of Mind theory. He
does an admirable job of exlaining how collections of mindless mechanisms
could account for phenomena as diverse as memory, learning, jokes, fashions,
self-images, intentions, classification, apprehension of analogies and
metaphors, the difference between seeing and remembering, motivation, fantasizing,
stages of development, commonsense versus logical reasoning, autism, the
formation of identity in spite of our underlying multiplicity, and the
“necessary illusion” of free will. Minsky is adamant that all events are,
at base, physical phenomena and are either caused or occur by chance. In
Minsky’s view, people are extremely complex machines, though Minsky would
argue that when we use the word machine, we tend to think, erroneously,
of very simple machines that “behave only in lifeless, mechanical ways”
rather than in terms of machines as complex as the human brain, with its
“billions of cells, each one complicated by itself and connected to many
thousands of others” (30). To make
his point about brains as machines, Minsky asks us to imagine replacing
each cell in a brain with a computer chip designed to perform the same
functions and connected to the other chips exactly as the brain cells are
connected. According to Minsky, “There isn’t any reason to doubt that the
substitute machine would think and feel the same kinds of thoughts and
feelings that you do—since it embodies all the same processes and memories.
Indeed, it would surely be disposed to declare, with all your own intensity,
that it is you” (289).
At one point in this fascinating, accessible, yet challenging book,
Minsky quotes Wolfgang Pauli as saying, “That theory is worthless. It isn’t
even wrong!” Minsky would be the first to admit that many parts of his
theory are highly speculative. Even where Minsky’s ideas prove, one day,
to have been wrong, they will have yielded powerful results, providing
us with ways to begin thinking clearly about who we are and how we work.
Minsky has spent his life in a demanding field in which a theory of a process
is as good as the program that one can write or the robot that one can
build to instantiate it. Time and time again, in reading Minsky, one has
the satisfying feeling that one has when assembling a complicated appliance
and having it operate smoothly. Ah, yes, that works. And even when it doesn’t
work, by actually attempting to build the machine that instantiates the
theory, we can learn what parts don’t operate and begin to conceptualize
alternatives. One of the reasons why artificial
intelligence is important is that it provides a testing ground for psychological
theorizing, forcing theorists to make their thinking concrete and operational.
Minsky’s book is an example of operational thinking at its best and will
doubtless serve as a roadmap for much of the work in artificial intelligence
and cognitive psychology in years to come.
References
Minsky, Marvin. The Society of Mind. New York: Simon and Schuster,
1986.
Minsky, Marvin. The Society of Mind CD-ROM, Macintosh Version.
New York: Learn Technologies Interactive, 1996.
Minsky, Marvin. The Society of Mind CD-ROM, Windows Version.
New York: Learn Technologies Interactive, 1997.
|
