Quote - IMO, the AI is never great in card games and the only way some of them can win is to cheat by deliberately dealing bad hands when the human player is winning. 

When I started the project I was expecting I would have to let the AI cheat in order for the game to be any challenge, and I had built in a function whereby a certain percentage of the time the AI would peek at the player's cards. In the end I never needed that, because the AI was quite strong enough without it. I was quite gratifed!

BUT

What I did do was choose a version of Poker very favorable to my purpose. It's not Texas Hold'em. In fact, as I don't know much about Poker, I don't even know what it IS called. The game uses a short deck of 32 cards, which maximizes the chance of high-scoring hands. There's no draw. Each player is dealt five cards, and then there follows at most three rounds of bidding.

The limited bidding rounds means that the player can't wipe the AI out on one hand, and the game is therefore tilted towards many short hands. This maximizes the computer's strength of knowing the exact percentages. The absence of a draw and the high score variability obtained from using a short deck also play to this strength.

Each "player" is characterized by a large number of variables such as proneness to bluff and so on, and I was able to find a combination of these that gave the human player a fairly hard time. I could usually beat it, having some inside knowledge of the algorithm, but the playtesters complained that they consistently got beat by the tougher opponents.