Poker AI

Okay, I'm working on a genetic algorithm to produce a good poker AI. It's not using Bayesian statistics or anything to analyze and attack your style of play, but it should be a near optimal value (i.e. some value near that of part of an optimal pair of strategies for a two player game).

Here is a trial run. This is a display of the best strategy at the end of each generation. The values listed are used to create 20 new AI with points near those, with some variance based on a kind of a skewed binomial squared (with sign preserved) mutation function. The fitness function is just 1,000 hands of poker played between every AI for P1's position and P2's position.

For poker nerds this is a one bet limit (ante 1, bet size 1) with a 1 bet cap (no raising), limit poker setting.

Every time you run the program it asks for new number of AI per generation, new bet and ante sizes, new number of hands per game (every AI plays one "game" (fixed number of hands) with every AI for its opponent's position (every P1 plays every P2)), as well as how many generations should run.

Here is a screne shot:

How he was born

I was thinking the core of Strong AI would really be the same. It should be able to use periferials to see and hear human feedback. It should be able to assess its own performance in terms of success and failure. It needs to be able to alter its behavior in creative ways.

If you can get some AI to do this, it will not be useful for just some problems, but it perhaps can be part of all AI. All AI seems like it could benefit from this behavior. Hopefully, it would be overkill for some tasks, and it would come to this conclusion on its own and remove this "creative/Strong AI" part from its code, becoming more traditional of an AI.

HRM

The hard part is trying to come up with a way for AI to decide what needs to be done, and how to do it. Once you have an AI performing a task, a Strong AI can assess its success based on feedback, and with negative feedback, regress to older behaviors, and with positive feedback, continue to progress coming up with new innovations.

Strong AI

Alexander Marsh is a computer programmer who works for Strong Intelligence, and artificial intellegence research company. His job is simply to translate the reviews of the different interesting AI written by the researchers into a database that can be read and viewed by the AI programs. This is his story.