A new arXiv paper traces the full history of computational belief revision, and it does not stop at the history lesson.
The paper is a narrative survey that starts with Doyle and London's 1980 taxonomy, the earliest attempt to classify how machines should update beliefs when new information contradicts old assumptions. It tracks how those early, practical ideas fed into the AGM framework, the formal theory that became the field's dominant model for belief revision. The authors work through Doyle and London's original taxonomy category by category, showing how each one evolved, survived, or got replaced once AGM's formal guarantees took hold. True to its title, the paper delivers both a survey and an implementation roadmap, though it stops short of actual working code, which the authors point to as a job for a follow-up paper.
Belief revision sits underneath any system that has to update what it 'knows' without breaking consistency, from 1980s expert systems to today's knowledge bases and reasoning agents. AGM theory has spent decades as an elegant but hard-to-implement standard, and the gap between its formal guarantees and actual code is a familiar complaint in AI research. A paper that connects the two properly, instead of just restating one side, is more useful than most literature reviews in this space.
It is still preparatory work, a foundation for the engineering paper still to come. But unlike surveys that only gesture at future implementation, this one includes the roadmap itself, not just a promise of one.