New Directions in Cooperative AI

Cooperation is a crucial organizing principle across all scales, from cells to societies. The question, however, of how collective, cooperative behavior emerges from interacting, self-regarding individuals and feeds back to influence those individuals remains open. Complex systems theory focuses on how cooperation emerges from deliberately simple individuals. Fields such as machine learning and cognitive science emphasize individual capabilities in changing environments without considering the collective level much. To date, however, little work went into modeling emergent collective, cooperative behavior from individually self-learning agents. As a result, cooperative intelligence misses out on a holistic complex systems approach, yet-to-be-discovered basic principles on collective cooperative AI cannot find their way into advanced AI systems, and multiagent (human-) machine learning applications may lose out on fulfilling their full cooperative potential. I am proposing a research agenda on collective cooperative AI that will unify classic approaches of multiagent learning, such as reinforcement learning and multi-agent systems, with complex systems theory, i.e., non-linear dynamics and evolutionary game theory.