In some ways, a good game and a good AI are opposing goals. A good game is deep, with deceptively simple rules leading to a wide variety of complex behaviors. Approaching the goal of complexity on a shallow level will create a game with a book full of complex rules, special cases, and distinct mechanisms. Something like that will leave any current game AI baffled. On the other hand, if the game is shallow, it will be relatively easy to program an AI that can beat it. If you can reduce the strategy down to a simple recipe, like you can on various levels of civ3, then you should be able to create an AI to follow that formula. A deep game would mean there is no perfect strategy, but rather something the human player can figure out intuitively on a per-game basis. Writing an AI to cope with that depth is a tall order.
My hope is that the AI will be able to play as well as a decent human player. The one saving grace of the above dichotomy is that the AI can apply brute force, making up with calculation the gap between it and human intuition. That may help explain why I am fixated on the game having a simple set of intuitive mechanisms. The game mechanisms have to be well-chosen, so that the right thing seems intuitive to the human player, and so they can lead to complex behaviors. The mechanisms have to be simple so that human players can understand them easily, and so that the AI can use its brute-force computation advantage to reach the same conclusions we do intuitively.
In other words, you can't separate the AI from the game. On the one hand, too much complexity in the game rules leads to an AI that can't cope, meaning the AI just has to cheat. Too little complexity, on the other hand, makes the game shallow and dull, and the AI too powerful. There's a tricky middle ground where a simple set of rules can lead to complex, challenging behavior. Think chess or go.
!)
moves very often.
Sullla.
