Computational Game Theory AI Solution with a Clear Air Force Stakeholder Need

Most military settings are imperfect-information games. Yet today they are not treated as such. Instead, planning, real-time re-planning, and training are done using simulation and tabletop exercises (or worse: gut feel) where the strategy spaces of Blue and Red are poorly explored. There is significant opportunity to improve using modern computational game theory AI techniques. Over the last 17 years, Dr. Sandholm’s group has become the world leader in this field. They won the world championships in the AI-versus-AI competition multiple years in a row, then in 2017 reached superhuman game solving in two-player games [Science, 2018], and then in 2019 in multi-player games [Science, 2019]. There is now a great opportunity to apply this technology to the Air Force. Strategy Robot, Inc., a Carnegie Mellon University spinout founded by Dr. Sandholm, has already been working on a certain important Air Force planning problem class for the Chairman of the Joint Chiefs of Staff since November 2018. Here we propose to extend that work to Air Force planning problems that are more general in several important ways.