> Using multi-agent reinforcement learning to address the symbol grounding problem.

Somehow I think he's made a few machine learning models.