TECHNOLOGY
Bayesian Networks
Judea Pearl's 1980s contribution—graphs that let a machine propagate probabilities through a web of dependencies the way water finds its level, and the first rigorous handling of uncertainty in AI.
A Bayesian network is a way of representing what depends on what. Devised by
Judea Pearl in the 1980s, it is a directed graph whose nodes are variables and whose arrows encode probabilistic dependence, together with a procedure for updating beliefs as evidence arrives—letting a machine propagate probabilities through the web of dependencies the way water finds its level. Before this, artificial intelligence had no principled way to reason under uncertainty; the
symbolic systems of the era manipulated certainties and brittle rules, and the alternatives were ad hoc. Bayesian networks gave the field a mathematics for graded belief, and the contribution alone would have secured Pearl's place in its history. Yet he came to regard it as a way station. The networks were brilliant at the first rung of
the Ladder of Causation—tracking what goes with what—and that very brilliance exposed the deeper lack: they could tell you that wet grass and rain go together, but not that the rain wets the grass and the grass