Comment by Quarrelsome | Hacker News Reader

Comment by Quarrelsome | Hacker News Reader

Parent Full thread

Yes.

Civ games have never been known for their stellar AI. The way they keep the AI competitive at higher levels is to brutally cheat by getting free techs and units at the start and a beneficial modifier to their production and research.

Considering how quick the AI needs to run (you don't want "end turn" taking forever) and how complex the game is I can appreciate its simplicity, it is disappointing though.

I have examined the sources for the Civ IV AI and must say that it isn't very pretty (hardly any comments too!). One thing it doesn't seem to do that often though is cheat by examining what other players are doing (such as not building wonders it wont get) which is something at least.

But still.... can anyone point me towards a game where the devs actually spent a significant amount of time (relatively speaking) on the AI? It always seems like such an afterthought. Mods seem to always be a better bet. Although perhaps that's because the most efficient ways to play a game are always developed long after the game is actually released.

GalCiv2, with the expansions, has a very good AI and one which isn't entirely focused on total war. There are options to dedicate more CPU time to the AI to increase the difficulty. It's possibly the best AI I've seen for 4x games.

Was about to mention GalCiv2 as well. Stellar AI. I once read a blog post by Brad Wardell (of Stardock) where he mentioned he only added gameplay features once he thought through how the AI can use them properly.

Also, if you're tired of the more mainstream turn based strategy games, I'd recommend something like Hearts of Iron 3 or Europa Universalis 3.

Fear 1 would be an example for a game with a good AI. Also Far Cry 1. Can't remember on a strategy game with a really good AI though (maybe Heroes of Might and Magic 3?).

I don't think any of the games above had good AI - but rather, they had believable AI.

I once read this piece by one of the AI programmers on the Halo series (can't for the life of me find it now) where he goes into the difference.

A good AI will be able to beat you time and time again using ingenuity and creativity. But such AIs are rarely fun. Instead, we prefer AIs that merely give the illusion of competence (e.g., they take cover... but always behind predictable objects. They flank, but always in obvious ways. They retreat, but always under specific, easy to understand conditions.). What the AI does is utterly predictable, but still challenging.

An unpredictable AI will quickly become frustrating instead of fun, even though it would be more realistic.

On the other hand, the predictability of the AI of RTS-Games (and the Civilization-games) gets critized quite often.

A good AI is a believable AI, in games. The AI of a game shouldn't be too powerful, the player has to have a chance to win. But predictable? I agree that the AI shouldn't be incomprehensible. The player should be be able to predict some moves. But if it really is utterly predictable, it is also frustrating.

Take Greedcorp as an example - or chess. If the player would never be baffled by the move of the AI, it would be no fun at all to play against them.

PS: But believability really is important. I always hated the moment in late-game of civ when conquering cities of advanced civilization with nothing in them but a temple. Always made it look like the AI doesn't even build properly and only can compete because of cheating, which repeatedly destroyed my motivation to beat the game.

Soren Johnson talks about AI cheating in "Playing to lose: AI and Civilization" http://www.youtube.com/watch?v=IJcuQQ1eWWI&feature=playe...

This is the most enjoyable and diverse AI I've been able to play against.

AI Wars http://www.arcengames.com/w/index.php/aiwar-features

can anyone point me towards a game where the devs actually spent a significant amount of time (relatively speaking) on the AI?

Chessmaster 9000?

For most games, though, AI is an unlimited time sink; every thousand hours you sink into making the AI smarter will barely enhance the player's enjoyment.

Or, often enough, will make the player's experience worse; I'm pretty sure there's no way I could possibly single-handedly kill hundreds of bad guys unless those bad guys were complete idiots who run towards me one at a time and never bother to hide, strategize, cooperate or even pick up the powerful weapons which are just sitting around on the freaking ground.

A well-developed AI is not without merit. Many players enjoy higher difficulty settings despite the added time/effort needed to complete a task due to a greater feeling of accomplishment. A good AI can offer the same inflated sense of reward.

Playing Diablo III recently was a poignant reminder to me of how badly RPG developers need to consider fleshing out their AI. In the endgame kiting again predictably emerged as the dominant strategy; developers are apparently oblivious to as to how to prevent the epidemic. Is it not worth it to program some basic responses to being kited by a player?

Reply on news.ycombinator.com