Reinforcement Learning

Reinforcement learning has been used very successfully in settings of single agent decision problems, where i'm the only one playing into aono drive my cara supermari. But it's also been used verysuccessfully in settings where there's an opponent, and the opponent may be unmodelled. One question that is sort of central here is, what happens when multiple reinforcement learning agents play against each other? Are they going to reach what we think is an equilibrium in the sense of rational agents, or are just going to reach something else that is not an equilibrium?