An Early Introduction to AI
As a student at Carnegie Mellon University (CMU),b I learned about “
artificial intelligence” from some of the
field’s founders. My teachers were
clever but took a cavalier, “Try it and
fix it,” attitude toward programming.
I missed the disciplined approach to
problem solving that I had learned as
a student of physics, electrical engineering, and mathematics. Science
and engineering classes stressed
careful (measurement-based) definitions; the AI lectures used vague concepts with unmeasurable attributes.
My engineering teachers showed me
how to use physics and mathematics
to thoroughly analyze problems and
products; my AI teachers relied almost
entirely on intuition.
I distinguished three types of AI
˲ building programs that imitate human behavior in order to understand
˲ building programs that play games
˲ showing that practical computerized products can use the methods that
Computerized models can help researchers understand brain function.
However, as illustrated by Joseph Weizenbaum,
2 a model may duplicate the “
black-box” behavior of some mechanism without describing that mechanism.
b CMU was then known as Carnegie Institute
THE VAST INCREASE in speed, memory capacity, and com- munications ability allows today’s computers to do things that were unthinkable when I started programming six
decades ago. Then, computers were
primarily used for numerical calculations; today, they process text, images,
and sound recordings. Then, it was an
accomplishment to write a program
that played chess badly but correctly.
Today’s computers have the power to
compete with the best human players.
The incredible capacity of today’s
computing systems allows some purveyors to describe them as having “
artificial intelligence” (AI). They claim that AI
is used in washing machines, the “
personal assistants” in our mobile devices,
self-driving cars, and the giant computers that beat human champions at
Remarkably, those who use the
term “artificial intelligence” have not
defined that term. I first heard the
term more than 50 years ago and have
yet to hear a scientific definition. Even
now, some AI experts say that defining AI is a difficult (and important)
question—one that they are working
on. “Artificial intelligence” remains a
buzzword, a word that many think they
understand but nobody can define.
Recently, there has been growing
alarm about the potential dangers of
artificial intelligence. Famous giants
of the commercial and scientific world
have expressed concern that AI will
eventually make people superfluous.
Experts have predicted AI will even re-
place specialized professionals such as
lawyers. A Microsoft researcher recently
made headlines saying, “As artificial
intelligence becomes more powerful,
people need to make sure it’s not used
by authoritarian regimes to centralize
power and target certain populations.”a
a Interview with Kate Crawford in The Guardian,
March 13, 2017.
The Real Risks of
Incidents from the early days of AI research
are instructive in the current AI environment.
of AI methods
can lead to devices