Game Theory in Artificial Intelligence
https://towardsdatascience.com/game-theory-in-artificial-intelligence-57a7937e1b88
The article “Game Theory in Artificial Intelligence” by Pier Paolo Ippolito explains how game theory is a vital concept in multiple scenarios/situations in our daily lives. It plays a role in business, war and defence, transportation, political science, and without a doubt artificial intelligence. Ippolito describes how game theory is applied to artificial intelligence models and algorithms using two different examples. First he starts off by explaining how game theory works with Adversary training in Generative Adversarial Networks (GANs). A GAN has two components, a generative model and a discriminator model, where the generative model takes in some features and tries to understand how those features were produced by explaining their distributions, as opposed to the discriminator model which takes in input features to predict to which class the sample might belong. Ippolito looks at a particular GAN where it produces new images and then looks at them to distinguish which ones are fakes and which ones are real. Here, the two players involved in this game theory application are the two different models (the generative model and the discriminator model). The generative model begins by creating a new image, while the discriminator model fails at distinguishing between the real and fake images at first. The discriminator model then learns through every iteration/new image that the generative model creates until finally each model gets to a state where they cannot improve anymore.
The second example that this article dives into is Multi-Agents Reinforcement Learning (MARL), which can be applied to AI-powered self-driving cars. Game theory is applied here to determine what is the most optimal method to program self-driving cars when they work against both the environment and other self-driving cars around them. This is an example of a more difficult game theory application because it is not just between two players (in this case the environment and one self-driving car) but instead multiple players because there are multiple self-driving cars. Nash Equilibrium is applied to both of these scenarios because in both cases, two or more players get to a point until each player (model or self-driving car) reaches a certain point in which all the players playing are in the most optimal position of the game. Likewise, none of the players will diverge from their strategy for the game because there is no incentive to. This article contributes to how new technology can be either be built or improved through basic concepts of networking. Game theory is a simple method to determine a necessary Nash Equilibrium for optimization in ground breaking technology such as machine learning and self-driving cars. This shows that game theory can later and should later be explored in other artificial intelligence such as machine learning in the healthcare industry.