Probabilistic Reasoning in Intelligent Systems
Pearl set out the full theory of Bayesian networks: how to build the graph, how to propagate evidence, and why that is computable.
Why it matters
Probability returned to AI not as decoration but as the main apparatus for reasoning under uncertainty.
The book gives belief propagation for trees and polytrees, works through d-separation, and shows when a graph permits purely local computation. It displaced MYCIN's certainty factors and similar heuristics that had no semantics. Graphical models grew from this book, and later Pearl's work on causality.