Geoscience Reference
In-Depth Information
my questions. I enjoyed the topic, though to this day I can't figure out what it had to do with warp
speed, time travel, or any related topics. But I know that already by that age I was fascinated with the
world of science.
Math and science were the subjects that had always come most easily to me; perhaps having a
father who was a college math professor had something to do with it. In high school, when other kids
were partying on Friday nights, I was hanging out with my computer buddies writing programs to
solve challenging problems. In fall 1983, after having seen the movie
War Games
, I became
determined to write a self-learning tic-tac-toe computer program, just as in the movie, a program that
could learn from its mistakes, a rudimentary type of artificial intelligence. The movie carried a thinly
veiled lesson about the futility of global thermonuclear war: There can be no winner in a tic-tac-toe
game expertly played; if neither player makes a mistake, the game will always result in a tie. Perhaps
if the computer—in the movie, it had seized control of America's missile program and was preparing
to launch a massive nuclear attack—could be brought to understand this paradox, it could recognize
the futility of nuclear war. For me at the time, however, it was just an interesting and challenging
computer problem to tackle.
Machine learning of this sort was in principle relatively straightforward. The real challenge was
in how to go about constructing an algorithm—a set of operations or calculations, here in the form of
a computer program—to solve the problem as efficiently and elegantly as possible. I had the
computer play itself, just like in the movie. That was the easy part. In the beginning, I simply had it
make random moves every turn. When it lost to itself, however, I would store both the final and
previous configurations of the tic-tac-toe board in a “blacklist”—moves that would no longer be
available to the computer.
13
The blacklist was used to ensure that the computer, while it continued to
make random moves, would not make the same losing move again; in this way it would gradually
“learn” how to play tic-tac-toe.
In practice, it might take a very long time for the computer to become skilled enough to avoid
losing because there are so many possible sequences of moves, and the program gets slower and
slower as it has to scan an increasingly long list of disallowed moves before each turn. But I
discovered a “trick”—the term scientists and mathematicians often use to denote a clever shortcut to
solving a vexing problem—to get the computer program to learn much faster. The trick was to exploit
the concept of symmetry. A tic-tac-toe game is the same no matter how you rotate the board, whether
you flip it vertically or horizontally, or whether you switch the role of Xs and Os. When you take that
symmetry into account, there are actually many fewer truly unique board configurations and many
fewer losing moves that need to be stored in a blacklist. Now I could get the computer to become
unbeatable in tic-tac-toe far more readily. The adrenaline rush, for the scientist, comes from finding
tricks that make a problem easier to crack. That—and eating pizza with my friends—was my idea of a
fun Friday night.
A Random Walk
A year later, in August 1984, as Ronald Reagan was completing his first term in office and Michael
Jackson's
Thriller
was the top-selling record album, I headed off to college at the University of
California, Berkeley (UC Berkeley). In part, I must confess, I was looking to get away from the harsh
