OT - Strategy games with GOOD AI

[analytic_kernel]

I have to agree with those who are recommending computer chess, even if it was tongue-in-cheek. As Agema has mentioned, games with simpler sets of rules can generally get deeper-thinking AIs programmed for them.

I've actually had some experience at working on improving AIs for open source TBS games. In one case it was for an existing TBS game engine that could run different kinds of games, depending on the game definition files you fed to it. I was younger then: more ambitious and less reflective. As I got further and further into that project, the full enormity of the task became apparent. It is truly a non-trivial undertaking to make an AI that plays well in a variety of circumstances when a lot of variables are in play.

I also started to develop a Python AI for Wesnoth about a year-and-a-half ago, but became frustrated by the limitations of the sandbox environment they had set up (for security reasons). Someone else actually had put together a Python AI that was a decent tactician, but it was computationally expensive and there was no "strategic level" thinking. I took some of his ideas and worked on one that performed some strategic analysis and kept coherent goals across multiple turns. It was doable, but I don't think it was a huge improvement, because there are so many situations where one kind of strategic analysis doesn't work well. (Closeness of starting positions, victory conditions, etc... can all force significantly different approaches to the game.)

So, I think you can sink a lot of thought and a lot of code into making a good AI, but after a certain point there are diminishing returns. "Use cases" that someone cares about do slip through the cracks. My guess is that professional game development shops realize this, determine a threshold that they deem good enough, and then don't expend much effort beyond that point.

As mentioned in an earlier post, humans are generally more challenging intelligences to play against. If I could go back in time and change some of my participation in the TBS engine project, I would not have touched the AI and would have instead focused my efforts on getting the networking code back in good shape. Working multiplayer functionality trumps having a decent AI, in my opinion. (Of course, if you have both, then more power to you.)