Blackjack and Neural Networks
Problem being addressed
Blackjack is a card game in which the player attempts to beat the dealer by a getting a score that is better than the dealers score, but not more than 21. As a game of chance and skill, it's a good test ground for neural network architecture. How can your computer play Blackjack?
Blackjack has simple dynamics as a game, and can be modelled probabilistically. The basic play is hit or do not hit, so give the player another card, or don't give the player another card. Players who keep track of the cards that have been dealt can determine their chances of getting a better score, or going bust (over 21). In this research paper, an artificial neural network/reinforcement learning approach is taken to the game of blackjack. Further, a number of different play strategies and models are presented, due to the simple game dynamics.
Advantages of this solution
This is an early application of neural networks to card games, and is an entry point into understanding dynamically changing behaviour and modelling approaches.
Solution originally applied in these industries
Possible New Application of the Work
Aerospace & Defence Sector
Airplanes are complex structures with a number of components. The risk of failure of an airplane is correlated to the risk of failure of these components. One way to think of this problem is that it is the same as a player going bust in Blackjack. A possible application of this research is to model failure/risk in airplane components with neural networks.
Financial problems and card games have a long history. One useful extension of understanding a game like Blackjack would be to build simulators for financial systems, and, essentially, artificial economies. These could be used to test out economic theory and policy and understand behaviour in financial markets. This is more relevant now that blockchain systems can build custom financial markets.
Sports betting is a major industry and data science has been used extensively to understand the performance of sports teams. But what about lesser commercial games like darts and snooker? How can AI and probability theory combine to build models of these games; and of how people play under pressure?
Source DOI: #############
Source URL: #############