The AI is fairly good, but not superb. I’ve only lost once to any of the three AI players included. An inexperienced player should get fairly stiff opposition for a while, though. What’s most interesting, the AI players are basically text files listing a bunch of parameters: how the player should value the buildings in different stages of the game and so on. Modifying the AI players or creating new is very simple.
I spent about five minutes meddling with the parameters and tested for few rounds to create an AI player that kind of matches my playing style. It proved quite effective: when it finally worked the way I wanted, it kicked my butt. I’ve yet to beat it.
Of course, this leads to an obvious idea: any takers for a San Juan AI tournament?
As an interesting side note, the creator of the program, Jim Getzen, seems to value producing a lot more than I do. It’s written in some of the assumptions his program has. For example, AI players move from stage one to stage two and three based on production. You can change the limit, but it’s always about production: when I have established this much production, I’ll move on.
Which is funny (and caused my AI player originally to get stuck in stage one), because I have no problems finishing the game (and being victorious) with just the one indigo plant I got in the beginning. Had I done the program, the production capacity wouldn’t have such a big importance.
Well, that’s something you can work around and I think there’s enough variables to allow for a fairly diverse group of AI players, but of course more would always be better. Not that I have any suggestions, though. It isn’t always that easy to transfer unstructured knowledge in your head to a formal shape for others to use, particularly if we’re talking about a computer program, which needs a lot more guidance and instruction than a fellow human being.
Anyway, if you have a Mac and you like San Juan, there’s no reason not to try Jim’s San Juan.