9.1 Rule Induction and Decision Trees

In what follows, much of the mystery of rule induction ("creating rules directly from data without human assistance") will be removed. When you see the nature of the approach, you'll probably think, "Of course! It's obvious!" With a little imagination, you might even create some rule-induction strategies of your own.

There was a time when computers were only programmed. A human being that understood how to solve a particular problem would create a step-by-step procedure (program) for solving it. When this procedure was slavishly followed by a piece of electronic circuitry, the same solution was produced as long as nothing unexpected was encountered. This sort of "procedural cognition" has offloaded a lot of work once done by humans, but tends to be very narrow. It isn't really what anyone would want to call "intelligence."

This approach worked well until there arose the need to automate solutions to problems that no human being understood, or that couldn't be reduced to hard-coded sequences of rote operations.

Suppose, though, that certain instances of the very problem of learning itself could be automated. Were this the case, perhaps...