Influence Diagrams—Historical and Personal Perspectives

The usefulness of graphical models in reasoning and decision making stems from facilitating four main computational features: (1) modular representation of probabilities, (2) systematic construction methods, (3) explicit encoding of independencies, and (4) efficient inference procedures. This note explains why the original introduction of influence diagrams, lacking formal underpinning of these features, has had only mild influence on automated reasoning research, and how Bayesian belief networks, which were formulated and defined directly by these features, became the focus of graphical modeling research.