Poker as a Mixed Strategy Nash Equilibrium for AI
https://www.technologyreview.com/s/603385/why-poker-is-a-big-deal-for-artificial-intelligence/
While Artificial Intelligence has become increasingly dominant at games such as Chess and Go, games such as poker have proved difficult. This is partly a result of poker being a game played with limited information – in Chess, players have access to all information relevant to the game, while in most poker games players only have access to certain information, and do not have knowledge about the values of others’ cards. Because of this, as the article discusses, every state in poker has to be viewed as a mixed Nash Equilibrium. Even though one particular move in poker may be immediately better than another, a professional player can surely adapt and counter that strategy, and so a good AI must be able to randomize its behavior, using mixed Nash Equilibrium, in an optimal way that is difficult, or even impossible, for a player to directly counter.
If Machine Learning techniques are employed to train poker AIs to learn these optimal mixed Nash Equilibrium, it might be possible that these AI will learn optimal mixed strategies that no top poker player was aware of before, or perhaps it will be discovered that a strategy previously believed to be sub-optimal (such as folding on the vast majority of turns) is, in fact, viable.
AIs like this, that are capable of learning optimal mixed Nash Equilibrium, can be applied to many other relevant domains as well. For instance, football coaches could give an AI many different values and variables related to their offensive success, and that AI could return on what percentage of plays that team should be running or passing. In certain eSports, where different game characters have different mechanics, a similar AI could calculate how often each character should be passive or aggressive for the best chance of success when facing a certain opponent. With all of these breakthroughs in Artificial Intelligence learning mixed Nash Equilibrium, in the near future AIs might even end up taking the jobs of offensive or defensive coordinators, or they might at least assist them in training and organizing their players and team properly.