Comment Re:AI (Score 5, Interesting) 604
I hope you don't mind, but I've taken the liberty of extrapolating on your question. It's a bit long, but I'm genuinely curious about this and really hope it makes it to Sid.
1. From a game designer's point of view, how much should you value game complexity against an effective AI? The new systems introduced in Civilization III, such as strategic resources and culture, add a new level to the game play, but also add another aspect of the game that the AI has to cope with.
No offense to Soren or the other AI coders on the project, but the AIs in Civilization seem to lack the ability to compete on these new cultural and resource "battlegrounds". Sure, they'll make attempts to pillage a resource tile here or there, and they'll build temples, but by and large these new game concepts just produce another way the player can exploit the AI. There are numerous exploits noted, such as camping scouts or other non-combat units on un-roaded enemy resources, and these are things the AI just never do. And, if the AI did do them, the player would get quite angry. Which leads to my next question...
2. Have you ever considered building in some sort of "frustration" level to the AI? One of the most common practices against the AI in Civ III was "herding" settler-stacks, by using multiple units to force the AI settler to move in one direction. Then, on the next turn, you would shift all the units in the same direction, forcing the AI settler to move back to it's original spot. This process could be repeated indefinitely, with the AI trying to head somewhere it will never reach. This problem arises because each turn is largely a whole new game to the AI, they have little if any memory of what the human did to them last turn. Wouldn't it make sense to have it remember such passive transgressions, and grow more and more angry at those behind the act? This same rule could be applied to remove the "exploits" known as "herding" and "oscillating wars". It's much like the existing reputation system, only driven by game actions, not just diplomacy.
3. It's clearly obvious, and acknowledged, that the AI cheats in Civ III at the higher levels. Do you see this as a valid method for compensating for AI code that cannot win a fair fight against a seasoned player? Playing a game of Civ III on "Sid" was simply absurd. Sure, it presented a challenge, but the game was so horribly skewed it wasn't Civ any longer; it was just mass produce and sprawl. While it could be argued that current AI technology on a standard PC will never be on equal footing with the human mind, do you not see it as a slap in the face to simply give the AI a 200% production bonus?
4. Have you ever considered having a "learning AI", much like that of Galactic Civilizations? I have been tinkering with the idea, and I really think it would be possible to create a sort of "game analyzer AI" that would accept some sort of "game history" file, which it could then analyze. From this analysis, it could extrapolate various patterns in human behavior, and add them to its heuristics when determining troop movement, production and diplomacy. I realize that it is not as simple as I have labeled it, "finding patterns", but in a finite environment such as any computer game is, a Neural Network could be devised to discover links between the actions a person makes. This could be done locally on the client machine, which would have the effect of making the AI adapt to the player's playing style. Conversely, players could upload their files to an online system that would analyze them and integrate the patterns found into a "collective AI", which could then be released through periodic AI-patches.
5. I'm graduating from Rensselaer Polytechnic Institute in December, finishing a degree in Cognitive Science (Computer Science and Psychology) in three years. Want to put a good word in for me at Firaxis? :)
No offense to Soren or the other AI coders on the project, but the AIs in Civilization seem to lack the ability to compete on these new cultural and resource "battlegrounds". Sure, they'll make attempts to pillage a resource tile here or there, and they'll build temples, but by and large these new game concepts just produce another way the player can exploit the AI. There are numerous exploits noted, such as camping scouts or other non-combat units on un-roaded enemy resources, and these are things the AI just never do. And, if the AI did do them, the player would get quite angry. Which leads to my next question...
2. Have you ever considered building in some sort of "frustration" level to the AI? One of the most common practices against the AI in Civ III was "herding" settler-stacks, by using multiple units to force the AI settler to move in one direction. Then, on the next turn, you would shift all the units in the same direction, forcing the AI settler to move back to it's original spot. This process could be repeated indefinitely, with the AI trying to head somewhere it will never reach. This problem arises because each turn is largely a whole new game to the AI, they have little if any memory of what the human did to them last turn. Wouldn't it make sense to have it remember such passive transgressions, and grow more and more angry at those behind the act? This same rule could be applied to remove the "exploits" known as "herding" and "oscillating wars". It's much like the existing reputation system, only driven by game actions, not just diplomacy.
3. It's clearly obvious, and acknowledged, that the AI cheats in Civ III at the higher levels. Do you see this as a valid method for compensating for AI code that cannot win a fair fight against a seasoned player? Playing a game of Civ III on "Sid" was simply absurd. Sure, it presented a challenge, but the game was so horribly skewed it wasn't Civ any longer; it was just mass produce and sprawl. While it could be argued that current AI technology on a standard PC will never be on equal footing with the human mind, do you not see it as a slap in the face to simply give the AI a 200% production bonus?
4. Have you ever considered having a "learning AI", much like that of Galactic Civilizations? I have been tinkering with the idea, and I really think it would be possible to create a sort of "game analyzer AI" that would accept some sort of "game history" file, which it could then analyze. From this analysis, it could extrapolate various patterns in human behavior, and add them to its heuristics when determining troop movement, production and diplomacy. I realize that it is not as simple as I have labeled it, "finding patterns", but in a finite environment such as any computer game is, a Neural Network could be devised to discover links between the actions a person makes. This could be done locally on the client machine, which would have the effect of making the AI adapt to the player's playing style. Conversely, players could upload their files to an online system that would analyze them and integrate the patterns found into a "collective AI", which could then be released through periodic AI-patches.
5. I'm graduating from Rensselaer Polytechnic Institute in December, finishing a degree in Cognitive Science (Computer Science and Psychology) in three years. Want to put a good word in for me at Firaxis?
