By picking actions that have higher regret with higher probability, algorithms can learn how to play games and converge to Nash equilibrium. An example is given with AI learning how to play poker.