I have lost the reference, but sometime within the last 72 hours, I
read a quote by an official of the Defense Advanced Research Projects
Agency (DARPA), the agency that
helped launch the Internet and recently got into trouble for creating
a "futures market" in terrorism. This official bemoaned the
stupidity of his laptop, which doesn’t know what he wants it to do;
he called for much more public investment in artificial intelligence
(AI).
I have an interesting colleague in computer science, Ben
Shneiderman, who strongly criticizes AI research. His argument is
not that the machines will take over the world and make us do their
will. Rather, he argues that AI tends to make machines less useful,
because they become unpredictable. When, for example, Microsoft Word
tries to anticipate my desires by suddenly numbering or bulleting my
paragraphs, that can be convenient—but it can also be a big nuisance.
Shneiderman argues that computers are best understood as tools; and
a good tool is easy to understand and highly predictable. It lets us
do what we want. All the revolutionary computer technologies
have been very tool-like, with no AI features. (Think of email, word
processing, and spreadsheets.) Meanwhile, untold billions of dollars
have been poured into AI, with very modest practical payoffs.