Belief change survey from Doyle to AGM maps AI implementation roadmap

Belief change — how an AI system updates its knowledge when new information conflicts with existing assumptions — just got a comprehensive historical and practical roadmap. Yuri Almeida and Arthur Casals published "From Doyle to AGM: A Survey and an Implementation Roadmap for Belief Change" in the European Journal on Artificial Intelligence (SAGE, 2026), with a 65-page accepted manuscript now on arXiv. The survey starts from Doyle and London's foundational 1980 taxonomy of computational belief revision, then traces how the field transformed through the influential AGM framework (Alchourrón, Gärdenfors, and Makinson) and into modern AI approaches. Rather than treating these eras as disconnected, the authors demonstrate direct relationships between early computational pragmatism and AGM's formal postulates, showing how each taxonomical category evolved after AGM. This historical analysis identifies recurring theoretical foundations and implementation challenges.

For engineers building knowledge-based AI, the paper's key contribution is its implementation roadmap. By synthesizing historical insights with formal guarantees, Almeida and Casals provide a systematic baseline for "engineering-focused belief change research" — a gap often noted in applied AI. This matters for systems like knowledge graphs, autonomous agents, or databases that must revise facts without collapsing into inconsistency. The survey bridges 46 years of theory into actionable design guidance, making it a valuable reference for anyone wrestling with belief revision, non-monotonic reasoning, or knowledge lifecycle management in production AI.