Tit-for-Tat

Learning to Model Other Minds: OpenAI Releases LOLA Algorithm

OpenAI has released Learning with Opponent-Learning Awareness (LOLA), an algorithm designed for multi-agent settings where each agent accounts for the fact that other agents are also learning. LOLA discovers self-interested yet collaborative strategies such as tit-for-tat in the iterated prisoner's dilemma. The work represents an early step toward agents capable of modeling other minds and reasoning about opponent behavior.