Throwing out a question concerning AI

[Checkpoint]

This is not my field of expertise, hence the query. Why is the AI always dumb? We know that they can make chess AI that beats the best human players in the world and so on, and I don't really know how much time that has been put into developing such AI, but what is the poor AI in computer games down to? Lack of time? Or is it simply an intentional decision to keep it at a certain level so as to not become too hard?

 

If it is the latter there obviously is a problem, because I imagine it is very difficult to create AI's that have different depth for the same game. Is it this problem they have to address? Do they go "people are too stupid, so for the better players we'll just have to give the AI cheats that won't help it anyway since it's too dumb to put its new-found cheat assets to any real use?"

 

The next problem is, would the majority of strategy-game players be that stupid? I'm not going to pretend I'm the best player out there, but I still manage to beat a lot of strategy games on the hardest difficulty setting. The AI is predictable and inefficient, and most of the time you can out-think it to compensate for its cheats. Is it simply just difficult to make good AI? Sometimes it just feels like they put a lot of time into a game, and then its ruined by the AI saying "we declare war! lolz!" and proceeding to mess things up. I just can't help suspecting that the game developers underestimate the brain power of the regular strategy gamer.

[Diamond]

Chess AI can't really be compared to one in games as chess has a very simple and static environment. Modern games, on the other hand, have highly dynamic environment and a lot more factors to consider.

[alanschu]

Chess is a pretty simplistic game when you sitback and look at it. All states are defined clearly, with clearly defined move possibilities. It's also a small scale game. Go is also an easy game, but it's board is many times larger than a chess board, so AI chokes on the branching factor.

 

The state space of a 3D strategy game is significantly larger.

 

Good AI is also very computationally expensive.

 

 

SOmeone seems to be venting in the general direction of Medieval 2: Total War :D

Edited January 13, 2007 by alanschu

[Checkpoint]

What makes you say that? " :D

I'm referring to games like Europa Universalis 2 as well, though.

 

I realise that chess AI is easier to program, but it was just a parallel given the near impossibleness of beating certain chess AI compared to the AI of a host of computer games.

Edited January 13, 2007 by Checkpoint

[alanschu]

Well, it's impossibleness is related to the "simplicity" of the game of Chess. Checkers is also dominating (to the point where human champions no longer play it), whereas Go is patently stupid, to the point where human champions won't play it because it's too easy.

 

 

I'd also argue that there are other things that will sell games more effectively (since these companies are in it for the money). Particularly graphics.

[213374U]

Are you referring to strategy games specifically? If so, I'd say it's quicker and easier to have the AI just cheat than actually develop an AI that can react to changing circumstances and come up with effective solutions to problems that aren't part of a set of predictions included in its original programming (as opposed to chess "AI"). Also, if a game did feature a true adaptive, learning-capable AI, I have no doubt it would be effectively unbeatable. And that's no fun either...

 

In the end, you are playing against a clock. A complex one, but a clock still.

[SteveThaiBinh]

It's a really interesting question. I read a game designer years ago who said that the problem with his game wasn't getting the AI to be challenging enough, but that it was often far too good for the player and had to be reigned back. I really wish I could remember which game it was he was talking about.

 

You so often hear reviews complaining about poor AI. Which games are generally thought by gamers to have unusually good AI? I've heard this said of Galactic Civlizations - which others?

[Checkpoint]

Are you referring to strategy games specifically? If so, I'd say it's quicker and easier to have the AI just cheat than actually develop an AI that can react to changing circumstances and come up with effective solutions to problems that aren't part of a set of predictions included in its original programming (as opposed to chess "AI"). Also, if a game did feature a true adaptive, learning-capable AI, I have no doubt it would be effectively unbeatable. And that's no fun either...

 

In the end, you are playing against a clock. A complex one, but a clock still.

<{POST_SNAPBACK}>

I mostly play strategy games, so yes. I'm obviously not advocating "unbeatable" AI, but I'm sick and tired of the dumb behaviour generally displayed to wonder what the reason is for all this. All this input actually helps me, though.

[Checkpoint]

You so often hear reviews complaining about poor AI.  Which games are generally thought by gamers to have unusually good AI?  I've heard this said of Galactic Civlizations - which others?

<{POST_SNAPBACK}>

I liked the progress made from Civilization III to IV, but the AI there is still very beatable. Particularly so on terra maps.

[Diamond]

I don't know, I always get my ass handed to me in Civ IV in land wars. I always try to wipe out the whole continent, because AI is weak in naval battles.

[Checkpoint]

Well, I'm not playing the hardest level in Civ IV. The difficulty there is very much based on cheats which I don't like. The problem with terra maps, though, is that the AI is bad at going after the new world and its resources. That, in part, is also related to its poor navies.

[Tale]

I hear good things about the AI in Company of Heroes.

[karka]

An interesting read

 

http://pc.ign.com/articles/750/750725p1.html

[Slowtrain]

Better AI costs time to develop and takes a lot of procressing power. SO it has a lot of cost associated with it, but at the same time its hard to say that better AI equals more game units sold. So AI devlopment is going to take back seat to pretty much anything else.

 

I remember Warren Spector said that one of the reasons the AI was so poor in Deus Ex was because they simply left it to the end because they considered other things more important and therefore only had a very short amount of time to get the AI working. ANd the result was obvious.

[Plano Skywalker]

well, I think the Holy Grail with regard to AI in computer games is to simply mimick a human player.

 

in other words, not only should the AI being trying to win, but there should be a distinct "personality".

 

 

for instance, let us say that you like to build more outposts than you really need to in your strategy games...but that is your "signature style"...that is how you LIKE to play.

 

well, I think the AI should go for more than just the shortest mathmatical path to victory...there should be different personalities that appear (randomly determined) each time you play.

 

one personality might like large fleets, one might like tech r&d, etc. they all amount to valid ways to win but each has a distinct flavor.

 

I also like the idea of an "adaptive AI" that intentionally plays one notch BELOW your level. That way, it is still challenging but you can make a mistake or two and still have a realistic chance of winning.

[Checkpoint]

An interesting read

 

http://pc.ign.com/articles/750/750725p1.html

<{POST_SNAPBACK}>

Interesting indeed!

[metadigital]

An interesting read

 

http://pc.ign.com/articles/750/750725p1.html

<{POST_SNAPBACK}>

Good computer AI can slow down the computer a great deal, as well. If the AI has to evaluate a lot of data, that can take a lot of time. One solution to that is through multi-threading. That is the process in which the game multitasks within itself. While the human player is moving, the computer players are calculating their turn. That's why in Galactic Civilizations, there's no "Please Wait..." dialog when you hit the turn button. The AI has typically already figured out where it's going to go and is only making small adjustments based on what you just did.

That's exactly how Kasparov was beaten by the IBM Deep Blue; the computer was busy scanning all the possible moves from the current position, so that when Kasparov eventually made his move the computer moved immediately. Part of the psychological strategy of the IBM team. (It is a commonly held belief that Kasparov was the superior player; the IBM team used all sorts of meta-gaming techniques to put him off his best game.)

[Checkpoint]

An interesting read

 

http://pc.ign.com/articles/750/750725p1.html

<{POST_SNAPBACK}>

Good computer AI can slow down the computer a great deal, as well. If the AI has to evaluate a lot of data, that can take a lot of time. One solution to that is through multi-threading. That is the process in which the game multitasks within itself. While the human player is moving, the computer players are calculating their turn. That's why in Galactic Civilizations, there's no "Please Wait..." dialog when you hit the turn button. The AI has typically already figured out where it's going to go and is only making small adjustments based on what you just did.

That's exactly how Kasparov was beaten by the IBM Deep Blue; the computer was busy scanning all the possible moves from the current position, so that when Kasparov eventually made his move the computer moved immediately. Part of the psychological strategy of the IBM team. (It is a commonly held belief that Kasparov was the superior player; the IBM team used all sorts of meta-gaming techniques to put him off his best game.)

<{POST_SNAPBACK}>

If only a strategic computer game could psycologically intimidate the human player...

[metadigital]

Well, time pressure is one way: people get harried when they have less time to make decisions (and too much information to process); that's what RTS games use to good effect.

[kirottu]

I still remember in Wolfenstein 3D when Hitler had stopped behind a corner to camp. I don

[Pidesco]

Monolith's games beg to differ.

[kirottu]

Monolith's games beg to differ.

<{POST_SNAPBACK}>

 

I haven

[Zoma]

It's a really interesting question. I read a game designer years ago who said that the problem with his game wasn't getting the AI to be challenging enough, but that it was often far too good for the player and had to be reigned back. I really wish I could remember which game it was he was talking about.

 

Might be refering to the game "Brothers in Arms". The AI in the first game is considerable dumb and predictable whilst the second game can be a little too overwhelming for some players due to the 'smarter' AI.

 

Not many people seemed to notice that the AI in "The Punisher" game is quite 'smart' in a way. The enemies are observed to run away and take pop shots at the player at the corner of the wall, even intelligent enough to take cover, retreat further and take cover and shoot when the player approaches them.

 

Though I think the The Punisher the player plays is somewhat godlike which results in unable to notice the AI's at work.

Edited January 13, 2007 by Zoma

[Calax]

I hear good things about the AI in Company of Heroes.

<{POST_SNAPBACK}>

it has a set routine that once you get it down you can usually beat them silly. Generally it'll take whatever unit it has the resourses for and will spam it. (the two usual choices by an axis comp are Pioneers and Stugs, Allied computer uses engineers and Shermans, unless their airborne then they get all antitank weapons.)

[Spider]

Monolith's games beg to differ.

<{POST_SNAPBACK}>

 

Yes, the AI in F.E.A.R. was quite excellent. To the point that I have a hard timge getting into other FPS now. That game spoiled the FPS genre for me on so many levels.