On programming gaming AI: "Good" vs. "Fun"

Thanks for the link. It is somewhat interesting, but its hard to see much in the way of a good argument for their actually being the distinciton between good vs fun AI. My main worry is that he is using the term 'AI' far too liberally for comparing algorithms that are meant to serve very different purposes. For example, the AI in a single player shooter campaign (such as enemy soldiers in half life2 single player) serve a very different purpose than the AI in bots in multiplayer shooters (such as in unreal tournament). It's like comparing the AI in Pacman with the AI GalCiv2. The function of the AI in both games is just far too dissimilar to make any interesting comparisons and this is not news to us. The author seems to think he is saying something new and interesting, but it seems a little trivial or outright vacuous as it stands. It only seems interesting if we thought that 'AI' only applied to games that have opponents that were on equal ground with human players. It also seems objectionable that the supposed 'fun' AI is subjective simply because it is not meant to mimic another human player.
Anyhow that is my winding ramble, which is probably only semi-coherent.

Successfully updated karma reason!

TheRealWarpstorm

Join Date 03/2003

+13

One thing to note, that GameSpy article is about 2 or 3 slides from the beginning of Soren's hour long presentation. It went a little deeper in the actual conference.

FWIW, I plan on asking Soren for advice if I get stumped on AI issues in my next game (I feel confident he could set me straight).

Successfully updated karma reason!

Good article. There can be many cases when AI could be almost as good as good human players. OR even better - in regards to his chances to win game (simplest examples would be FPS shooters - since advanced scriptable bots in Quake they could be much better than any human player).

But thing is not always fun to play such AI- he plays to win and players loses. Players dont like that, especially when they play single player .In fact the skill difference on the other hand is what drives similiar players out of MP.

On the other hand you have people who like challenge - and they usually move to MP pretty fast .Reason being that even if AI is hard to beat- it is usually because he is cheating hardcore, and even if it not really cheating much at all it still feels different from defeating another human player

Successfully updated karma reason!

But thing is not always fun to play such AI- he plays to win and players loses. Players dont like that, especially when they play single player .

Now see, I *am* a person that likes it when the AI can beat me, especially if it's not cheating.

One of the reasons I enjoy games like GalCiv 2, Shogun Total War, & Medieval Total War is because the computer actually has a chance of defeating me. Maybe my views are in the minority, but I love strategy games where I actually have a chance of losing to the AI -- I find a genuine challenge to be a lot more fun than games where I know I'm probably going to win.

Successfully updated karma reason!

TheRealWarpstorm

Join Date 03/2003

+13

Soren called his presentation "Playing to Lose" as that is essentially what he did for Civ4, made the AI play to lose in an entertaining way.

Successfully updated karma reason!

Now see, I *am* a person that likes it when the AI can beat me, especially if it's not cheating.

One of the reasons I enjoy games like GalCiv 2, Shogun Total War, & Medieval Total War is because the computer actually has a chance of defeating me. Maybe my views are in the minority, but I love strategy games where I actually have a chance of losing to the AI -- I find a genuine challenge to be a lot more fun than games where I know I'm probably going to win.

Yeah I am same way- I like challenging AI , but for a few rar exceptions (galciv2 on suicidal and Civ on emperor/deity) I never encountered good AI- and those are good only for sole reason of immense cheating (btw I never beat civ4 on deity -it is just crazy how many bonuses it gets- like 4 settlers on 1st turn vs yours 1). Total War series are great but I always thought AI there is very weak (both strategicaly and tactically) .

My point was that balancing is often made on "average" players. And average players are not really good at all. So they make AI, let some casual players play it and then balance it according to their feedback. In the end we have retarded AI everywhere since there is no push for good AI

Successfully updated karma reason!

That's not good AI in my opinion. That's just giving an AI such a huge bonus that it challenges you by sheer weight of numbers.

Successfully updated karma reason!

Now see, I *am* a person that likes it when the AI can beat me, especially if it's not cheating. One of the reasons I enjoy games like GalCiv 2, Shogun Total War, & Medieval Total War is because the computer actually has a chance of defeating me. Maybe my views are in the minority, but I love strategy games where I actually have a chance of losing to the AI -- I find a genuine challenge to be a lot more fun than games where I know I'm probably going to win.Yeah I am same way- I like challenging AI , but for a few rar exceptions (galciv2 on suicidal and Civ on emperor/deity) I never encountered good AI- and those are good only for sole reason of immense cheating (btw I never beat civ4 on deity -it is just crazy how many bonuses it gets- like 4 settlers on 1st turn vs yours 1).

Actually, it's 2 settlers in Civ 4 / Deity, but it gets free workers and all basic techs as well, and that's a lot of a difference. Also, upkeep cost is a killer for a player on that difficulty.

Successfully updated karma reason!

i liked the AI in Spaceward Ho!...
after time it learned to play like the players..

Of course this was very interessting with lot of different players involved

Successfully updated karma reason!

I'm by no means an AI developer, but from my experience, developing AI is extremely difficult. It's not easy to create an AI to play as well as an intelligent human. In chess, computers can brute strength all the possible moves and figure out the best choice. Add some pattern recognition and it can beat most human players. But in a game like GalCiv or Civilization, there is no way to brute strength all the possible moves.

Developers have to create 'cheating' AI because it's just not possible yet to make the computer good enough to beat the best human players.

This is largely the reason for the distinction between 'good' and 'fun', IMO. If the AI could be as good as the best human players, it would be fun. How can a player argue that it's not fun when it's playing fair? Instead, it operates within a smaller set of guidelines which limits its abilities.

There are games where the player is supposed to win. A campaign in a strategy game should be internally biased in some fashion. But when the AI is just another opponent in a deathmatch, it should play the same as a human.

In many cases, even with the AI doesn't have to cheat to win, it often doesn't play the same as a human. The reaction time is too fast or the mousing is too accurate or something similarly technical. That's another challenge for the developer to make the AI into a convincing human.

Successfully updated karma reason!

It seems that the only real way to make an AI that is fun to all levels of skilled players would be to make a different AI for each level. Since that is ridiculous with regard to time and money spent, the next best thing would be to balance the AI for the really good players (note that this can't be done until there ARE some really good players), and then nerf it a bit and then a bit more for each level down below that.

You could then have some 'underdog' setting where the AI gets bonuses for the players that like to have the odds against them.

Hey, does that sound like the AI in any games we know?

Successfully updated karma reason!

I believe that this same speech was discussed on AIGameDev.net. If I remember correctly, I believe that the consensus (or at least a common opinion) was that the two are not as mutually exclusive as the article makes it out to be. Good AI can be fun AI as well.

Successfully updated karma reason!