This is very nerdy so feel free to skip this and watch an episode of The Simpsons or even better The Big Bang Theory. I really do think I'm half-Sheldon and half-Leonard, ha. (I promise to write a normal article about MTG Forge next time.)
For my geeky readers, you might want to checkout this article about evolving a good evaluation function. The name of the article is "Bootstrap Learning of Alpha-Beta Evaluation Functions" which refers to the evaluation function that is used for alpha-beta, an optimized version of min-max. (The article uses the Greek symbols alpha and beta.)
I'll try to summarize the article. The min-max algorithm doesn't involve any learning since it uses a static evaluation function and the article proposes the idea that you save the game state tree that is generated and try to evolve a better evaluation function. The idea is to test the evaluation function with the previously generated game tree to see whether the evaluation function made good choices.
(Remember that the goal of the evaluation function is to determine which player is winning and by how much. Usually the evaluation function generates an integer number to represent how much the AI is winning or losing.)
Saving the whole game state tree that alpha-beta creates is difficult. I understand the alpha-beta algorithm but I don't understand it well enough to code it and the alpha-beta code that I have found is recursive, so I don't know how to save the game state tree that is created.
Basically the article is really interesting and has some really good ideas but I'm not sure how to practically use it. (Thinking about AI hurts my brain.)
I got my four year bachelor's degree in Information Science but we didn't touch AI. I really need a Master's degree or PhD in order to really understand AI. Anyone willing to foot the bill? :*)
I found a nice Java implementation of the alpha-beta algorithm in the download section for the book "Algorithms in a Nutshell". You can read more about the book and download the code from here. (To download the Java source code the link is on the left side in small letters, click on "Examples". Now you should see three directories, click on "Releases".)
Or you can download the 10 MB zip file from here.