How to Safely Exploit Predictions in General-Sum Normal Form Games

Given a prediction of opponent behavior in a generalsum two-player normal form game, it is difficult to select a strategy that balances the opportunity to use the prediction to inform one’s action with the risk of becoming predictable. We propose Restricted Stackelberg Response with Safety (RSRS), a novel way of generating such a strategy. RSRS uses an r-safe Stackelberg equilibrium in a modified game, which is created to reflect the assumption that the prediction might be inaccurate. With appropriate parameter selection, RSRS produces strategies that can play well against the prediction, respond well against a best-responding opponent, or guard against worst-case outcomes. We describe an algorithm that selects appropriate parameter values, which we have tested on multiple general-sum games, comparing its performance to that of other algorithms.