MIT researchers demonstrated that a previously overlooked category of algorithms performs significantly better than expected in a defined class of games with implications for AI strategic reasoning.