TITLE: Psychohistory, Economics, and AI
AUTHOR: Eugene Wallingford
DATE: August 03, 2011 7:55 PM
DESC:
-----
BODY:
Or, The Best Foresight Comes from a Good Model
In
my previous entry,
I mentioned re-reading Asimov's Foundation Trilogy and made
a passing joke about
psychohistory
being a great computational challenge. I've never heard a
computer scientist mention psychohistory as a primary reason
for getting involved with computers and programming. Most of
us were lucky to see so many wonderful and more approachable
problems to solve with a program that we didn't need to be
motivated by fiction, however motivating it might be.
I have, though, heard and read several economists mention
that they were inspired to study economics by the ideas of
psychohistory. The usual reason for the connection is that
econ is the closest thing to psychohistory in modern academia.
Trying to model the behavior of large groups of people, and
reaping the advantages of grouping for predictability, is a
big part of what macroeconomics does. (Asimov himself was
most likely motivated in creating psychohistory by physics,
which excels at predicting the behavior of masses of atoms
over predicting the behavior of individual atoms.)
As you can tell from recent history, economists are no where
near the ability to do what Hari Seldon did in
Foundation,
but then Seldon did his work more than 10,000 years in the
future. Maybe 10,00 years from now economists will succeed
as much and as well. Like my economist friends, I too am
intrigued by economics, which also shares some important
features in common with computer science, in particular a
concern with the trade-offs among limited resources and the
limits of rational behavior.
The preface to the third book in Asimov's trilogy,
Second Foundation,
includes a passage that caught my eye on this reading:
He foresaw (or he solved his [system's] equations and
interpreted its symbols, which amounts to the same thing)...
I could not help but be struck by how this one sentence
captured so well the way science empowers us and changes
the intellectual world in which we live. Before the rapid
growth of science and broadening of science education, the
notion of foresight was limited to personal experience and
human beings' limited ability to process that experience
and generalize accurately. When someone had an insight,
the primary way to convince others was to tell a good
story. Foresight could be feigned and sold through
stories that sounded good. With science, we have a more
reliable way to assess the stories we are told, and a
higher standard to which we can hold the stories we are
told.
(We don't always do well enough in using science to make
us better listeners, or better judges of purported
foresights. Almost all of us can do better, both in
professional settings and personal life.)
As a young student, I was drawn to artificial intelligence
as the big problem to solve. Like economics, it runs
directly into problems of limited resources and limited
rationality. Like Asimov's quote above, it runs directly
into the relationship between foresight and accurate models
of the world. During my first few years teaching AI, I was
often surprised by how fiercely my students defended the
idea of "intuition", a seemingly magical attribute of men
and women forever unattainable by computer programs. It
did me little good to try to persuade them that their
belief in intuition and "gut instinct" were outside the
province of scientific study. Not only didn't they care;
that was an integral part of their belief. The best
thing I could do was introduce them to some of the
techniques used to write AI programs and to show them such
programs behaving in a seemingly intelligent manner in a
situation that piqued my students' interest -- and maybe
opened their minds a bit.
Over the course of teaching those early AI courses, I was
eventually able to see one of the fundamental attractions
I had to the field. When I wrote an AI program, I was
building a model of intelligent behavior, much as Seldon's
psychohistory involved building a model of collective
human behavior. My inspiration did not come from Asimov,
but it was similar in spirit to the inspiration my
economist friends' drew from Asimov. I have never been
discouraged or deterred by any arguments against the
prospect of artificial intelligence, whether my students'
faith-based reasons or by purportedly rational arguments
such as
John Searle's
Chinese room argument.
I call Searle's argument "purportedly rational" because,
as it is usually presented, ultimately it rests on the
notion that human wetware -- as a physical medium -- is
capable of representing symbols in a way that silicon
or other digital means cannot.
I always believed that, given enough time and enough
computational power, we could build a model that
approximated human intelligence as closely as we
desired. I still believe this and enjoy watching (and
occasionally participating in) efforts that
create more and more intelligent programs.
Unlike many, I am undeterred by the slow progress of
AI. We are only sixty years into an enterprise that may
take a few thousand years. Asimov taught me that much.
-----