My gaming forum of choice has just discovered Tom Francis’ after action report of a Galactic Civilizations II campaign. As I noted in my brief mention of this AAR a couple of months ago, there’s some curiosity about how much rationalizing Francis is doing to account for the AI’s actions.

This rationalization could be seen as the fevered delusions of a games journalist trying to justify the screwy thing that happened to him in a game he was supposed to lose.

I prefer to see it as evidence that the game is good.

Gamers love to complain about AI, but, like many gamer complaints, I’m not sure if they are saying what they mean. “Bad AI” is used to refer to both predictable computer opponents and unpredictable computer opponents. The term is a catch-all criticism that boils down to “That’s not what I would do in that situation.” There is an unrealistic expectation that the computer controlled enemies or allies play like a human does. And when game designers give the AI a bit of an edge just to keep them from being rolled over, there is a vocal contingent out there that decries this leveling as “cheating”. And nobody likes a cheater.

What most gamers want, I think, is an opponent that seems human. Face it, few of us know enough about how AI works to really know what is happening in the code behind the actions on the screen. So there is a natural inclination to connect what the computer opponent does to some sort of reason or logic. Which makes sense, since computer programs are pure logic. But you know that a game is doing its job when gamers talk about “deceiving” the AI not in terms of exploiting a known blind spot, but in terms of pulling off a great con.

Things get interesting when people start assigning human characteristics like personality and emotion to the AI. We speak of “angering” opponents. We talk about “making nice” with enemies. And of course these are just rule sets translated to our screens. This is harder to pull off without assigning “personality” traits like Civ IV does.

This is part of why Combat Mission: Shock Force is such a colossal disappointment. The AI is unhuman. If the setup tells it to advance, it will until it meets fire. Then it stops, never to move until you decide to meet it. If it sees enemy armor, it will attack it with the nearest available unit, so good-bye infantry squad. It won’t seek cover from enemy artillery or make use of terrain in any sensible way. Every battle becomes a turkey shoot. Where the problem with Theatre of War was getting your own soldiers to stop doing stupid things, CMSF is plagued with enemy forces not doing anything most of the time and when they do things it’s always the wrong thing.

So the AI performs in a way that isn’t anywhere near optimal. And the illusion that I am playing against a thinking being collapses. Unless I want to imagine that the soul of General Haig has seized control of all the armies in the Middle East, there is no way to rationalize what is happening on my screen.

Compare this to the brilliantly convincing, if not brilliant, AI in the earlier Combat Mission games, a computer opponent that would advance through rough ground, place its guns in intelligent places and even play peek-a-boo with its tanks. Mortar fire would send men running for the nearest building or crater. Yeah, you could trick it, but it never felt like you were taking advantage of it. The Germans were just buying your feint. Or they were shooting too soon because they were so afraid of your reputation as a leader of men.

A good AI, then, isn’t the one that always provides a challenge or that can compete with you without cheating. A good AI is one that brings you into the world the designer has created. Civ IV does this by assigning traits and personalities to its leaders. GalCiv 2 does this by some sort of alchemy of priorities and race typologies. X-Com did this by carefully balancing tactical limits and your natural fear of the dark. AI is, ideally, just one ingredient in a stew of immersion.