“What truly is logic? Who decides reason?.....”
― John Nash
Game Theory, quite unlike its name, is a serious affair to deal with when it comes to the configuration and planning of an AI model. In essence, while linear machine learning deals largely with single-dimensional elements in their very nature, the true power of AI is actually unleashed with game theory application, and it’s various facets. To understand game theory power in AI, however, it is essential to understand the basics of what actually constitutes game theory and its applications. So here’s the promised primer on what game theory actually comprises.
In its textbook definition, “Game Theory is the study of strategic interaction”. The concept gets its name from board games, where strategic interactions are most common. One player’s decision affects the decision and action of the other players and vice versa.
Real-world strategic interactions, however, can be quite complicated. These complexities can be circumvented by using models, a key aspect in the study of game theory. Some models include the Nash Equilibrium, Pareto Efficiency, the multiplicity of equilibria, Symmetric and Asymmetric and a whole host of other game models. Game theory is relevant to, and applied in almost every sphere of human existence, from war to business, to well, board games. However, in the world of Artificial Intelligence, where machines learn to play for the win, and multidimensional, multi-element interactions are imposed on the models, do things start to get really interesting.
Game theory is a crucial element in building AI models today. To begin with, game environments and models are increasingly becoming popular training mechanisms for machine learning, including imitation learning or reinforcement learning. The area where it is gaining ground is in multi-agent systems that can be made to participate in gamified interactions between other agents within a game model. Game theory is crucial in deep learning systems as well, to facilitate some of the important capabilities that multi-agent systems require to enable different AI programs to interact in order to reach a goal. Game theory bases itself on the assumption that the intent of each agent is to win the game, taking into account the actions of other agents within the same dimension.
How does this relate to AI? In reinforcement learning, multiple agents interact with each other within the gaming dimensions. Given that the objective is to win and other variables remaining unchanged, the agent that wins the game is finally considered as the most suitable. Multiple agents can either compete or collaborate in gaming dimensions to accomplish a task with accuracy and efficiency - the foundation for reinforcement learning in ML.
This also has the potential to solve real-world problems, and the use of game theory being applied in AI is spreading in different areas ranging from cybersecurity to healthcare diagnosis.
In typical settings, the game mechanism is designed with the end objective (output) in mind. The participants are also designed and all the agents and elements are facilitated to interact in the predesigned environment. The overall diagrammatical representation would be something in the lines of:
Game theory is just one of the aspects of Artificial Intelligence skills that you need to have in your arsenal as an AI engineer. Applications across several industries, not just gaming, includes the practical understanding of the real-world scenario and the design of participants and the mechanism to achieve the intended results. Looking to make a career in this exciting new space? Be one of the first to be on board with the world’s most respected and pure-play Artificial Engineer certification. Improve your employability in multiples now!
Sign up for the ARTIBA newsletter and get the latest blog updates straight to your inbox.Subscribe