AI4FM: A new project seeking challenges!
Gudmund Grov, Cliff B. Jones · Verified Software: Theories, Tools, Experiments · 2010
The “proof obligations” generated from many formal methods tend to be simple and can often be discharged by modern automatic theorem provers or SMT systems. However, those proof tasks that need hand –or interactive– intervention present a barrier to the use of formal methods. Theorem proving was one of the earliest challenges addressed by researchers in the area of Artificial Intelligence and enormous progress has been made in the provision of general purpose heuristics. The approach in the recently started AI4FM project is different: we hope to devise a system that will learn from an expert user how they tackle one interactive proof and then apply the discovered high-level strategy to other related proof tasks. We are fortunate in having access to many such problems through the DEPLOY project but are aware of the dangers of devising an overly specific approach. This short paper appeals for challenge problems from other sources.