This paper extends the classic two-armed bandit problem to a many-agent setting in which I players each face the same experimentation problem. The main change from the single- agent problem is that an agent can now learn from the current experimentation of other agents. Information is therefore...