Creating AI Bots to Beat the Banker

 

Blackjack is a classic card game that many players are familiar with. Encyclopedia Britannica writes that it could have evolved from either French, Italian, or British card games. However, its exact origin is unknown. To this day, it's a staple in casinos, drawing crowds from all walks of life. Recently, people have begun experimenting with Blackjack and AI technology, to test how its programs would do in predicting the game's outcomes.

Using AI to Win at Blackjack

For the uninitiated, Blackjack has pretty straightforward rules. A guide to Blackjack on Gala Casino explains that each card in the deck has a numerical value, and at the start of each round, you'll be presented with two cards. The goal is to get 21 or a value that's closer to it than the dealer's (a.k.a. the banker) without going over. You can choose to "hit" to get a third card from the deck, but it risks going over 21, resulting in an automatic loss. Whether you win or lose relies solely on probability, which is precisely why this game works so well with AI technology.

One notable way to utilize AI is to have the program count cards for you. A feature on Vice details that card counting is when players keep accurate track of the cards that have been dealt, and then use information that to accurately anticipate what's left in the pack. Using what's known as the "high-low count system", they're able to calculate the probability of getting a good card should they decide to hit. This technique is quite hard to master, given how much the player needs to remember in a short amount of time. However, with the help of a machine learning detection program that can identify cards, players can easily deduce what cards are still in the deck.

Recent innovations have also shown that an AI can replace the human player. An article on Towards Data Science shares how one could teach a neural network how to play Blackjack. The writer goes into detail about how he trained the neural network to pinpoint the correct move for a given situation. In this case, it had to know when to hit or to stay. When the neural network was run through around 300,000 Blackjack games, it won 42% of them and tied in 9%. The developer notes that while it didn't beat the banker, it still had a decent performance. To compare, the average win rate of human players is around the same number at 42.22% with an average tie rate of 8.48%. This shows how AI technology can be used to beat the bank, but it isn’t necessarily better than humans with the same information.

Using AI to Catch Cheating Players

On the flip side, AI technology can also be used to the casino administrator's advantage. In online casinos, AI programs are used to spot cheaters. By keeping an eye on player behaviors and flagging any suspicious patterns, casino administrators can catch cheaters in the act.

However, it's worth noting that these AI programs are not 100% accurate. After all, some players might just be having a really good day. And that raises the question: Who is morally accountable for the actions of an AI? At present, there aren't many policies that dictate how to handle mistakes in machine learning, making it difficult to hold anyone accountable. Given this, these technological innovations, as useful as they may be, ought to be implemented with caution.