Before I go to the topic of explanation, I would like to say a few words on the role of Al in such applications as statistics, public health, and social science. One of the reasons that I find these areas to be fertile ground for trying out new ideas in causal reasoning is that unlike Al, tangible rewards can be reaped from solving relatively small problems. Problems involving barely four to five variables, which we in Al regard as toy problems, carry tremendous payoffs in public health and social science.

Joseph Weizenbaum, of M.I.T., who has recently come to critize computer education for children as a context for "toy problems," nonetheless got more than a toy gasp out of the public in 1968 when he unveiled his project ELIZA — an IBM 7094 programmed by Weizenbaum to "practice" psychotherapy.