Comparing AIs for chess and Billard to 4X AIs make no sense, they are too fundamentally different.
It makes perfect sense, if you realize that I'm only talking about them from a perspective of what happens when you actually make a perfect AI (which happened in those two cases), and how the market actually doesn't like it a whole lot. A game where the AI is the best player on the board, all the time, would not do very well, even if it could be built (and with the budget available, it can't be built).
Nobody actually said the AI in those games and the AI in this game looks like at a code level, because they don't.
I think AI developers should focus more on trying to mimic human players. Why is it that we can win against AIs who have strong favorable bonuses?
Because we can think, and learn, and the AI can't. We also get annoyed when the AI uses the advantage it should have - being the game, it can know where everything is. But we tend to call that cheating.
In part because we exploit anything we can in the games and are complete bastards about using loopholes against them. As loopholes are found, I think the AIs should be programmed to use them too. Just this simple thing would make them really more interesting and lower the " bonus cheat " requirements for hi level play.
Entirely doable, with the budget for continuous AI development as those loopholes are discovered.
Another thing that should be considered when making AIs is making a list of the stupid things they do over and over again and program them to NOT do them. For example, in endless space, once you had defeated an enemy main fleet, the AI would send you his new ships as soon as he built them, Which is a completely stupid thing to do. You would defeat a 2-3 ship fleet every turn which did not even make a dent in your fleet. A human player would fall back, regroup, and counter-attack a few turns later with a decent fleet.
They do that already, during development. After development, it's entirely doable with a budget for continues AI development as the foolish behavior is discovered.
hmm, I seem to be using that B word a lot.
Also, Just implementing some of the cheap tactics that most human players use is a pretty simple way to enhance an AI I would say. In Civ IV the AI was pretty good and they did act in some interesting fashion, like planning attacks the same way a human player would and it made it quite interesting.
If you played Civ IV you will know that when Monty (Montezuma) declares you war, you needed to be ready for it 10 turns before. As the very next turn he appears out of the blue with 8 galleons full of jaguar Warriors and he drops them right next to your capital.
Civ IV's AI was a lot better than Civ V's AI, for sure. In part it's because the game is simpler for the AI to play. The AI in IV can launch this kind of attack by building up an army in a stack, figuring out who it wants to attack, sneaking up close, and declaring war. Attacking is a matter of moving the Stack of Death to a target and going nuts. The V AI has to figure out how to position all its units on the board optimally to fight, which it really struggled with.
One of those cases where the game design has a significant effect on the performance of the AI.
In GalCiv 2 I noticed the AI is not such a Jerk. From the moment they DOW, you usually have time to prepare proportionally to how far they are. I understand sometimes they DOW because someone else bribed them so they have no way of preparing then.
The AI in GalCiv 2 was pretty good but it still sucked at some specific things, like evaluating what is needed to capture a planet. It seems they really suck at bringing enough transports for the job. A Human will usually not bother trying to capture a planet unless he is pretty much certain he will capture it.
I'd expect the transport issue to go away this time, because yeah. The AI really shouldn't mess that up. It's like it was calculating what it needed at the time it sent the transport out, not factoring in that the planet might grow or get fortified while on route.
Anyway... my point in all this is simply that the goal in every game is to make an AI that's good enough to compete. The goal is not to create an unbeatable super AI, both because they really can't afford to do that given the complexity of these games and because if they somehow managed to, it wouldn't be terribly good for sales. You'd wind up having most players playing against an easier version.
All the processing power available to GC3 should help the AI a fair bit. It's going to have a lot more power to try to play a few moves ahead and see what happens when deciding what to do, along with some other tricks that it didn't have in the past. So I have high hopes (but not in the alpha, of course).
