In a new paper, researchers at DeepMind propose using the game Diplomacy as an environment for training reinforcement learning agents.