Few months ago I discovered a wargame called "Conquest of the Aegean" with a very exciting AI. No script at all, no cheat at all, and could decently handle every maps (even players creations), by simply "understanding" what an objective or a threat is. He knows how to define simple tasks (defend, assault, delay, retreat) and execute them properly. AI limitation comes to lack of anticipation and complexity for the CPU to handle strategic situation better than a human not pure dumbness that we usually see in games. So at least, when you beat him (which occurs time to time), you're feeling smarter than him, not sorry to have find another flaw in the AI design.
When you read about the guy who make the game (he was almost alone on this task), he did not come with strong mathematical theory or algorithmic fundamentals, but with pragmatic view of what the AI should do, some skills to implemented them properly, a limited but well defined field of application (world war II wargame), and a well-crafted game design at the first place.
My experience is that :
One should never forget that AI is not meant to beat you, but to improve the game experience. Depending on the game, making an unbeatable AI is quite trivial (give it unlimited resources, or invicible units), but making an AI enjoyable is far more complicated. In this regards, cheat, repeatability and scripting are bad design that should be avoided.
AI is not only your dumb CPU opponent. It is also one of your friends when it comes to pathfinding, NPC, GUI enhancement, micro-management, and other tedious tasks where it excel at. Furthermore, today almost every games could be played online, with (against ?) other human players. So there is less need to spend millions in making a decent opponent AI than before.
AI should strongly be linked to the game design itself, and not be done after it.
Slashdot Mirror
Few months ago I discovered a wargame called "Conquest of the Aegean" with a very exciting AI. No script at all, no cheat at all, and could decently handle every maps (even players creations), by simply "understanding" what an objective or a threat is. He knows how to define simple tasks (defend, assault, delay, retreat) and execute them properly. AI limitation comes to lack of anticipation and complexity for the CPU to handle strategic situation better than a human not pure dumbness that we usually see in games. So at least, when you beat him (which occurs time to time), you're feeling smarter than him, not sorry to have find another flaw in the AI design.
When you read about the guy who make the game (he was almost alone on this task), he did not come with strong mathematical theory or algorithmic fundamentals, but with pragmatic view of what the AI should do, some skills to implemented them properly, a limited but well defined field of application (world war II wargame), and a well-crafted game design at the first place.
My experience is that :