We study experimentally how players learn to make decisions if they face many different (normal-form) games. Games are generated randomly from a uniform distribution in each of 100 rounds. We find that agents do extrapolate between games but learn to play strategically equivalent games in the...