We study the game theoretical problems that arise in multi-agent learning. In particular, we are interested in the simplest possible setting of symmetric zero-sum games,i.e, games that look like rock-paper-scissors, in the sense that there is no dominant agent/strategy. A recent paper that tries to address this problem, talks about a variant of previously proposed multi-agent reinforcement learning algorithm called the Policy Space Response Oracles (PSRO), where they showed improved performance compared to learning independent reinforcement algorithm. We will look at these two algorithms and interpret the game theoretical intuition behind them. In addition, we re-created some experiments in to understand the working of the algorithms.