Neat AI does Blackjack using NEAT and a Genetic Algorithm

Learn Blackjack Video Source & Info:

Its time to check out Blackjack and see if card counting really works. So I coded it up and got a neural net to evaluate a betting strategy based on the True Count to see how it compared to the theory.

Turns out that it works very well !!

Music:
https://www.bensound.com

Holy Rollers:


Source: YouTube

Share this video:
Neat AI does Blackjack using NEAT and a Genetic Algorithm

6 thoughts on “Neat AI does Blackjack using NEAT and a Genetic Algorithm

  1. Damn your neat ai vids are really neat. Nice data visualization.
    Why do you always add a bias of one as input? I don't find that anywhere else, but on your channel 😅

  2. How did you added the true_count to the existing neural networks ? Did you just added 1 input node to the 5 already existing + 1 ouput node for the bet size ?

  3. Unlike poker, Blackjack is one of the rudimental games, where you don’t really need a playing cards, just a 13 numbers*4 with 1&11 being equal and couple rules of interaction. I think, applying the same rules to the different size of number set or modifying the goal of the system may lead to far more interesting games generated by AI.

  4. The AI is being trained to learn BS which may not be the optimum strategy for a particular set of rules. You don't include insurance or say how many times a player can split, for example. Would be interesting to see how it bets with a limited amount ofcash to start.

  5. I wonder if you could train an AI to count cards but also evade detection while still making at least some profit.

  6. Nice video, but the fitness function is cheating a bit. It's using the known answer to define its fitness, so it's really just a supervised problem where the answer is already known. It would be more interesting if the fitness function had to instead learn Basic Strategy via the data it's collected, like average profit or something more complex.

Comments are closed.