This Gamespy article talks about an AI that's playing to win and a fun AI that's playing to lose.
http://pc.gamespy.com/pc/civilization-iv/854509p1.html
With those constraints in mind, the differences between the two types of AI -- and the games you'll find them in -- is clear. Good AI tends to be based around multiplayer games, with a fixed rule set and equally-matched sides. You never want Good AI to cheat, certainly, and (unless it's a training program) you want it to use all available tactics. The quality of Good AI is objective and easily measurable: how well does it play the game? And in most cases the Turing Test is relevant: for example, a good chess-playing computer should be indistinguishable from a good human player in a blind test. In short, Good AI plays to win.

That's a sharp contrast to Fun AI, which you'll find in very different games. Fun AI is mostly for single-player games, with unevenly-matched sides. Even if it doesn't outright cheat, the AI probably plays with different rules than human players. It usually uses a limited set of available tactics. It's hard to measure the success of a Fun AI, as it's subjective. The Turing Test is irrelevant; nobody would think that the monsters in Desktop Tower Defense are controlled by people, or certainly, not anyone you'd like to meet. In short, Fun AI plays to lose.
I was playing Civ and I thought that Firaxis did a good job combining good and fun AI together.
What do you think? If you only had to pick one type of AI, would you want to have a good AI or fun AI?

Their description of "fun" AI doesn't sound very fun.

I'd prefer AI to try to be clever like Human players rather than simply give the AI advantages or different rules, however creating very good AI is hard to code. I reckon the AI in Civ4 are probably not the best going around at "thinking" compared to other games - however - I think Firaxis came up with a decent system of giving them advantages per difficulty level. While predictable they are still challenging in Solo games (at your preferred difficulty.) In multiplayer they can be added usefully as enemies or allies depending on what sort of game you are intending to play with your human friends.

The thing is that article is a summary of the first couple of slides of Soren's hour-long presentation (and it lost a bit in summarization) on developing Civ IV's AI. For a game, IMHO, the important thing is how the player experience is. Given that, I'd have to say "fun".

FWIW, I plan on asking Soren for advice if I get stumped on an AI issue on any of my future games.

That's a terrible description of the two positions on gaming AI. It's obvious from the writeup that they are arguing for "Good AI" (another poorly chosen term), which they take to mean an AI that "plays to win."

Let me put it this way: no AI in any game truly "plays to win." If it did, the game would be no fun at all. The Civ4 AI definitely doesn't play to win - which is a very good thing. If the AI played to win the game, it would ruthlessly declare war on the Human player at the earliest possible opportunity, since the Human has the best chance to win the game in almost all cases. You'd also better get used to the AI never voting for you in the UN, never trading you technologies/resources, and playing to screw you at every possible opportunity (e.g. ruthlessly whipping all cities into the ground when defending in war). Sounds like a lot of fun, huh? :rolleyes:

What good AI should do (and what the Civ4 AI does) is provide a challenge for the player to overcome, while acting in its own general self-interest. Note that this is different from doing everything possible to win the game. This is also similar to the argument of "realism" versus gameplay; good games should be willing to sacrifice realism in order to achieve fun and balanced gameplay. When you try to put realism first, you end up with disasters like Master of Orion 3. Furthermore, it's impossibly utopian to believe that you can have an AI play on equal terms with a Human without receiving some kind of boost under the hood...

We had many of these debates when designing Civ4. Generally speaking, the argument that the game should be fun won out over the argument that the AI should play to win. (This is why the AI doesn't declare on you in 3000BC, even if you leave your capital empty.) I think it turned out pretty well for us in the end. :)

^ Good points.

(This is why the AI doesn't declare on you in 3000BC, even if you leave your capital empty.)
Good points overall, but I'd like to note that this particular example would be acceptable to me on a high difficulty setting, so long as the human player has had some chance to defend his city.

It's also worth noting that Civilization is a bit different from many other games in that it's not the human versus the AI, it's the human versus several AIs, and the AIs versus each other. So, even if the AI was all-out to win, it may still trade with the human player if doing so would give it an advantage over a stronger AI-controlled civilization.

In general, I don't mind giving the AI some advantages at high difficulty settings, so I guess I lean more towards the "fun" side, just as long as it's not a total moron. Heck, even if the AI does do stupid things, that can make the game more entertaining. I expect all of us who played a lot of Civ2 have seen that game's AI build "great walls" around the south pole - see here (http://www.civfanatics.com/civ2/aistupidities/).

Idiotic? Yes.
Hilarious to see? Yes, and I'm willing to sacrifice a little challenge for a little humor. ^_^

A fun AI is a strong AI. That's why I like multiplayer modes.

I like na Ai that is fun, puts up a fight but isn't extremely hard or challenging. I don't mind losing sometimes, but I want it to be for a good reason and not "the AI cheats"

I used to love scorched earth and tank wars. In those games you could choose different ai personalities from moron to cyborg. With a range in between that were good depending on the settings. Playing with all cyborgs wasn't fun usually because I usually never even got a chance to shoot, but playing with all morons was annoying. My vote is more leaning towards good ai, but good in situations. They have their strengths, just like a human player does.

^ Good points.

Only if you assume that the AI nations are not competing against each other which they theoretically should be.

Good points overall, but I'd like to note that this particular example would be acceptable to me on a high difficulty setting, so long as the human player has had some chance to defend his city.

At least in Civilization IV at deity the AI would just immediately crush the player with their extra troops they are already given.

Good points overall, but I'd like to note that this particular example would be acceptable to me on a high difficulty setting, so long as the human player has had some chance to defend his city.

At least in Civilization IV at deity the AI would just immediately crush the player with their extra troops they are already given.

Just to muddy the water, Soren's title for the presentation is "Playing to Lose".

Good AI's that play to win without using cheats, that's good. "Fun" AI that is a general idiot and cheats to win because the developers were lazy, that's bad.

Fun > Good although you need to have the right balance. It's like cooking good food or making bombs, you got to get all the ingredients right.

AI should also strive for survival. It should not start suicidal wars against player just to annoy him/her. There's absolutely no sense for AI nation to start war against player with 100 to 1 ods against them no matter how anoyed they are about the player's actions. AI should also accept resonable peace terms if they are clearly losing a war. No more those 'I wont accept any peace terms even if you take 99% of my empire and siege my last city' situations.

Damn terrorists keep sneaking up on me on Terrorist Hunts. Now that is good AI, and it's fun because it's challenging.