First I think the game needs to be balanced and then you can work on a AI.
in the game of dice + poker I see a few problems:
1. Poker relies on the concept that you can "guess" what the enemy has based on 2 factors: The cards on your Hand + Table and how they bet. This only works because the cards are limited. So you don't can have 2 times 4 of Kings. This does not work in a dice game.
2. In Poker you reveal the card on the table one by one giving each player the opportunity to Bet
3. The cards in Poker have far more combinations then dices. So it is harder to tell if your hand is "bad" or not
I would suggest to either change the game to a other dice game with lying or add some meta mechanics outside the dice poker.
E.g. instead of making rerolling a part of the game, make it kind of a cheat. The player could distract Abe and reroll the dice, this needs to be balanced with a kind of downside if he got caught of cause. Or If you lie you have to win a kind of starring contest mini game.
Other dice game that comes to mind is Meier.
https://redcupshop.com/en/pages/trinkspiel-maexchen-meiern
It is a common party game in germany and uses dices + a believe/lie mechanic. The game also has very few rules which can make it easy to implement.
Good AI in my opinion has 2 criteria:
- Deterministic so the player can s strategize around it
- Random so the player can not predict it
Depending on the game these need to be balanced.
For a strategy game the AI must strategize. So think a few steps ahead. Then validate it by a criteria (in this case for example if he wins or losses chips). This can be done for X amount of times for a random decisions and then choose the best strategy or for all possibilities (if this is not to resource heavy) and then randomly select out of the best X.
There is alot of stuff to learn when it comes to game AI and I also only touched the surface :D so there is probably better methods out there.