Interactive learning in 2×2 normal form games by neural network agents

► Neural networks concurrently learning to play 2×2 normal form games are simulated. ► Agents are randomly rematched with opponents, and games are randomly drawn. ► Neural networks learn to play unique pure strategy Nash equilibria. ► Neural networks abide by dominance and iterated dominance principles. ► Risk-dominant equilibria are played more often than payoff-dominant equilibria.