I understand now you’re focused on an academic definition in the game theory sense, personally I don’t think this has much utility in considering actual games, but I’ll acknowledge that by that definition you’re probably correct. I suspect by that most “AIs” in games wouldn’t pass the bar of counting as an agent, even generous definitions that would accept a flow chart would probably concider most AIs to be part of the game state rather than another player (eg the nazi soldiers in wolfenstein aren’t playing to win, they’re set dressing for you to kill). The opponents in Civ are more likely to count as agents perhaps.
Yeah, certainly, sorry if that wasn’t clear. Up above I tried to stipulate that I was speaking from a game theory perspective.
And yeah, you can model the AI in a game in whichever way is most useful. I said as long as they have utility functions that differ from the player(s), but then you also can recursively define games in terms of winning games.
Ex. the famous case of the US deliberately losing battles to not give away that they had cracked the German cipher. Each battle could be modeled as a game, and the war could be modeled in terms of battles.
Similarly, a single room in wolfenstein could present an contained “game”, the outcome of which is applicable to which ending you get in the larger “game” (I haven’t played it), and thus the AI would be agents at one level, but state/strategy at another.
I understand now you’re focused on an academic definition in the game theory sense, personally I don’t think this has much utility in considering actual games, but I’ll acknowledge that by that definition you’re probably correct. I suspect by that most “AIs” in games wouldn’t pass the bar of counting as an agent, even generous definitions that would accept a flow chart would probably concider most AIs to be part of the game state rather than another player (eg the nazi soldiers in wolfenstein aren’t playing to win, they’re set dressing for you to kill). The opponents in Civ are more likely to count as agents perhaps.
Yeah, certainly, sorry if that wasn’t clear. Up above I tried to stipulate that I was speaking from a game theory perspective.
And yeah, you can model the AI in a game in whichever way is most useful. I said as long as they have utility functions that differ from the player(s), but then you also can recursively define games in terms of winning games.
Ex. the famous case of the US deliberately losing battles to not give away that they had cracked the German cipher. Each battle could be modeled as a game, and the war could be modeled in terms of battles.
Similarly, a single room in wolfenstein could present an contained “game”, the outcome of which is applicable to which ending you get in the larger “game” (I haven’t played it), and thus the AI would be agents at one level, but state/strategy at another.