Civilization AI: Playing to Lose

Civilization Designer Soren Johnson has a long one hour talk about the AI in the Civ series and games in general. It turns out to be a very interesting discussion and I learnt quite a few things about the AI in Civ and another way to perceive AI in games.

Interesting points I gathered from the talk:
  • Soren mentions that in the Civ games they were targeting for three types of players: ones seeking a challenge (i.e. the powergamer I guess), Sandbox players (i.e. the kind that comes from playing SimCity or The Sims) and the ones looking for a Narrative (they go for the roleplaying, looking for a history they can recollect). I have to say I probably fall in the latter category.
  • Game difficulty was added to cater for the aforementioned three types of players (principally the Powergamers and Sandbox players). Settler difficulty is so easy that it is effectively like playing in a sandbox and the harder difficulties were for powergamers
  • Soren even mentions Puzzle Quest in his talk when discussing how some perceive the AI to cheat. He refers to the fact that during the AI turn it appears that they get lucky chains (i.e. 4-of-a-kinds, 5-of-a-kinds etc.) but it doesn't seem to be the case during the player's turn, when in fact the player has the same chance as the AI. Soren believes that the Infinite Interactive developers could have toned down the difficulty somewhat - and adopt the same mantra of "playing to lose" with the AI (since Puzzle Quest is after all a casual game).
  • Even though "Noble" difficulty is meant to be the level where the AI is on the same playing field as you (i.e. no advantages/disadvantages) it in fact still "cheats". The AI gets a bonus against barbarians, doesn't pay the same amount for unit support, unit upgrades, inflation and war weariness doesn't affect the AI as much. So it looks like all those years of accusing the AI of cheating was true! Apparently the reason the AI got these "cheats" was because it needed "help" in these areas as it couldn't do it as effectively as a human.
  • In Civ I and II, there was actually code to give the AI "free wonders" every so often, which seems pretty unfair!
  • In Civ I and II, there was actually code that instructed the AI to gang up on the human player if the human player was winning and it was past 1900!
  • In Civ III there was a funny exploit to use against the AI when it came to amphibious landings. The code basically instructed the computer player to land units in the city with the least number of units. Humans would counter this by leaving units out of a city until the last second and then place a unit back inside. They would then proceed to remove a unit from a city on the other side of their empire. Ultimately by repeating this behaviour you could get the AI's fleet of transports into an infinite loop!
  • AI code only accounts for about 2-3% of the entire codebase of a Civ game
  • Usually 1.5 years is spent on testing AI during a closed beta as it is deemed to difficult to test any other way.

Thanks to Danno for the heads-up!