"Expertise in Expert Systems"
When I arrived at CMU in 1982, expert systems were all the rage: find an expert, encode the rules by which he makes decisions into a computer and presto-magic we can replace humans with droids. That vision didn't quite work out. Then came the statisticians with their focus on formal probabilistic models. Now the neural-network guys are telling us that we can dispense with both the experts and the models, and that all we need is some data and the chain rule.