Artificial Intelligence and Nash Equilibrium: A Novel Way from DeepMind’s AI to Elegantly Solve Asymmetric Games
(Link)https://towardsdatascience.com/beyond-the-nash-equilibrium-deepminds-clever-strategy-to-solve-asymmetric-games-d571c998dac4
Alphabet Inc.’s, more colloquially known as Google, subsidiary DeepMind, of AlphaGo fame, has introduced a novel way to solve asymmetric games. This article from Towards Data Science explains exactly how this artificial intelligence system can systematically solve asymmetric games while maintaining Nash equilibrium. Symmetric games, in this context, is a game where there is a “dynamic in which the different players share the same strategy and goals,” while an asymmetric game is a dynamic “in which players have different and often conflicting goals and strategies.” This is where the Nash equilibrium comes in; the model mathematically derived the situation where each player chose the strategy in such a way that any other strategy would result in benefit assuming other players do not change. However, as the article states, the Nash equilibrium fails on two fronts: the unrealistic assumption of unlimited computing power, and the failure to account for the notion of risk that often comes in real-world implementations.
DeepMind picks up where the Nash equilibrium leaves off by breaking down the asymmetric problem into its symmetric single population counterparts. Via the usage of symmetric replicator dynamics and the fact that the Nash equilibria of the symmetric counterparts are formally related to the Nash equilibria of the original asymmetric game, DeepMind’s AI is able to successfully analyze the asymmetric game while maintaining Nash equilibrium.
This article provides quite an interesting insight into the future of AI and its role in analysis and implementation in relation to economics and networks. Such a technology could revolutionize the art of risk evaluation, analysis, etc. and advance inter-connectivity in various industries.
An illustration of how the AI deconstructs the asymmetric game into two symmetric counterparts.