An unrealistic AI

[Komoda]

Having the world generator working, its does not seem to matter how big the map is are where your civ ends up starting. For some odd reason the AI places any where from 1 or 2 additional civs on the same mass of land.
The map could be 300x300 with 5+ civs playing, which is the common number of players in a MP game. and it would still place 1 or 2 civs on the same continent (even an island).
Not meaning to gripe, but if the boys are serious about coming out with a civ4, I think they need to spend some real time with the AI to improve upon it.
Thye other thing I do not get, is how an AI civ can build a wonder with very little production, or have production at -10, and still complete it before the player does.
Another thing about the AI, which I find totally lame, is that it can spit out settlers and forces faster than you, and not have to worry about civic unrest.

The AI needs work.

 

[HungryMouse]

I am almost certain that the AI does not cheat with the rate of production. It does have some advantages related to the cost of buildings and units depending on difficulty level above Regent, maybe that's the case for your oppenents being able to produce settlers and units faster than you? Or they can for example make pop-rush, the thing that you would probably avoid doing early in the game...

 

[kris55]

I 100% agree, the AI in Civ is not as good as in other games. The AI is not very intelligent at all; it seems to react to the environment and not the situations in the game. It doesn’t calculate its own actions.

I have a fine example, I was in the modern age, and a Civ from the industry age, with no powerful units demanded goods from me, when I refused they declared war, and the funny thing was, i had nuclear weapons, but they still did it. I am sure that Zimbabwe wouldn't declare war on the USA in real life.

The AI does need some work to get back to standard of games out now, when it first came out 4 years ago, it was ok, but now, more is demanded from games in their realism.

When you have been playing CIV for a long time, you get used to the way the AI plays, and you can guess its moves 3 turns ahead. In the other games i play you can't do that, and i have tried! it doesn't work at all.

I know that making AI is extremely hard, i have a friend who does it, and he is very dismissive of the Civ game due to its sluggish AI.

I hope Civ 4 will have intelligent AI to match or surpass the amazing accuracy of other games out today.

 

[Doc Tsiolkovski]

I 100% agree, the AI in Civ is not as good as in other games.

Name any game with an AI even on par with Civ3. You won't find anything in sight.

That said, of course it isn't perfect.
But, the AI does never cheat - it does have its own ruleset (entire map known, sees all units, level boni), and thats it.
The map generator has its own flaws; yes, it tends to always put several Civs on the same landmasses. Still, that is reasonable - the game looks for the 'best starting positions' with a minimal distance. And often enough, several of those best starts are on some rivers in the moderate zones.

 

[vmxa]

Understand that RTS games do not have the complex demands that Civ does. So it is easier to code an AI that is passable. The only tricky thing RTS's have to deal with is path finding.

The other aspect is that if they make the AI so as to incorporate better strategy, it would cut off a lot of players. They would not be able to win.

The situation you mention, does not have to be all that foolish. If you look at it, that Civ is already doomed and has nothing to lose by going to war.

BTW read or watch the Mouse that Roared. I think it would be possible for Zim to declare war on US and not suffer. If they did not attack, they would be ignored, if they were to declare and send troops, oh well.

Now if you change the USA for maybe China or the old USSR????

 

[TyranusBonehead]

I agree that the AI does some pretty funky stuff sometimes, but I tend to think it's one of the better written AI's I've ever played against. Try playing it on Diety, that's tough. Each step up the difficulty ladder is won with trial and error and alot of learning along the way. You don't discover the AI's flaws and then beat it every time.

Anyways, just rambling...

 

[Brain]

I 100% agree, the AI in Civ is not as good as in other games. The AI is not very intelligent at all; it seems to react to the environment and not the situations in the game. It doesn’t calculate its own actions.

I don't know what you mean by that, but absolutely all game AIs (even chess) work essentially in the same way. They evaluate the current position and choose their moves based on a decision tree. This is not intelligence but computing power.

All current commercial strategy games (except classics such as chess) use fairly simple rule systems to decide what to do. Perhaps the next generation of games will start using some of the more advanced results of AI research, but for now most games are at or below the level that Civ is at.

I know that making AI is extremely hard, i have a friend who does it, and he is very dismissive of the Civ game due to its sluggish AI.

I think he doesn't have a clue of how complex a good AI system really is. Many people like to talk a lot, but they don't know s**t.

Try playing it on Diety, that's tough.

But it's not tougher because of better AI. The AI plays exactly the same way on Chieftain as it does on Sid. It only gets some advantages at the higher levels.

Understand that RTS games do not have the complex demands that Civ does. So it is easier to code an AI that is passable. The only tricky thing RTS's have to deal with is path finding.

Pathfinding is actually a well known and easily solved problem.

Actually, RTS is more complicated than turn based strategy because there is an infinite number of game states and an exhaustive search is not possible. Essentially, by the time the AI is done analyzing, the game state has already changed and it would have to start over. In order for the AI to be effective it must adjust its decisions based on continuous input and this is tricky.

The reason why our standards are higher for turn based games is that the human player has some time to think and evaluate everything. The game depends solely on strategic decisions and not on how fast you can click (which the computer can do very fast). Another reason why some strategy games seem to have better AI is because they are conceptually simpler. It's relatively easy to make an AI for a game where you just have to accumulate a large amount of money, build a huge army, and kick everybody's butt.

Anyway, everytime the topic of AI comes up I hear the same kind of comments and I'm getting tired of them. AI is hard to make. Those who think they can do a better job should quit whining and try making an AI of their own, and quickly realize how stupid they were for even thinking about it.

 

[Serutan]

IMO that argument is not valid. The AI having a different set of rules than the human player is, by my
definition, a cheat.

That said, I agree that writing an AI
for something like this is not at all easy, and it's all
too easy to poo-poo the Civ3 AI.

 

[budweiser]

Name any game with an AI even on par with Civ3. You won't find anything in sight.

MOO. ten characters

 

[vmxa]

Brain I have not seen anything to supports your position that RTS is complex, it is not. I did not say that pathfind was imposible, only it is the tricky part and in the past they failed to handle it well.

Where is the dipolmacy, trade in RTS games? Remember CivIII is 4 year old engine by now and really much of it goes back many years before that.

Much of what the ai does poorly could be corrected, but you have two issues:

1) targeted game systems (do not want to required latest ones
2) do not want to make it unbeatable, except by dedicated players (reduces sales)

 

[Brain]

Brain I have not seen anything to supports your position that RTS is complex, it is not.

Perhaps the following will convince you:
http://ai-depot.com/GameAI/GameTree.html

I did not say that pathfind was imposible, only it is the tricky part and in the past they failed to handle it well.

But it has nothing to do with RTS games. Turn-based games have exactly the same pathfinding problems. Please look up A* (A star) or Dijkstra's Algorithm on Google. It's a well known computer science problem that has been solved a million times before.

Where is the dipolmacy, trade in RTS games? Remember CivIII is 4 year old engine by now and really much of it goes back many years before that.

I'm not saying the AI in RTS games is better. I'm saying that it's more complex to implement. Therefore, it should actually be worse than the average turn-based AI because it's harder to do a good job with it. Also, as I said, our expectations are higher for turn-based AIs than for RTS AIs. There's no diplomacy in RTS games because players don't expect it. The AI in RTS games will never go as deep in its analysis because players don't expect it.

 

[Scuffer]

Extremely powerful and specifically designed computers struggle to beat the best humans at chess, a game with a really small board, few units and few parameters.

Civ runs on computers designed to do all sorts of things, has a massive game board, many opponents, huge numbers of units, techs, resources, terrain, combat system that isn't absolute.

It simply isn't possible to devise a human-like AI for civ. I mean, how would Deep Blue cope if it had to factor in the fortification bonus for a knight, or worry about caputuring a bishop if the rooks had bombard, or whether it could make a positional breakthrough before the opponent discovered Nationalism from a goodie hut?

 

[Brain]

Chess is not the same thing. Chess is a tactical game where the computer calculates the combination of many moves and counter-moves. Civ on the other hand is a strategic game. The combination of individual unit moves doesn't really matter as much as the overall strategy. Computers are very good at tactics, but suck pretty badly at strategy, and that is essentially why there are no good strategy game AIs.

 

[vmxa]

I read the one page at the link, but it never compared RTS to TBS perse. It merely was addressing a different way to handle the Ai in RTS games.

BTW I love RTS games, since way back on WarCraft and Command and Conquer. So I am not saying one is better than the other either. Both are great and satisfy difference needs. I love RPG as well.

 

[Breunor]

MOO. ten characters

Budweiser,

I haven't played MOO but I played MOO2 -- in my opinion, the MOO2 AI is MUCH worse than Civ's. Specifically, it always keeps its big fleet together, so the esy 'hit em where they aint' strategy is pretty easy to pull off. I don't know if this applied to MOO original.

I think the AI here is quite good. I do think a few games are better, but I give the AI here a good mark considering the complexity of the game.

For instance, I think the HOMM 3 AI is about equal, maybe a tad better. The games by Keating and Trout have very good AI -- both long-time wargamers who put making the AI really good as a high priority. However, despite their good AI their games weren't well received, so too much AI focus may not be great either.

EUII is one of the most brilliant games ever made, but despite constant tinkering, the AI isn't real good (again, be gentle here. This game is much more complciated than Civ.) One reason I turn to Civ over EUII is I th inkt he AI at Civ is better (I think they are both wonderfully great games).

The AI for Stronghold and its expansion is very good -- ever see how the computer plays King Richard? This was a great peice of work.

But, that's about it for me! I've probably played about 30 - 40 major games, and I don't think I have too many others I would even compare. Even war games, which are built pretty much for their AI, can rarely hold the grade.

So, I would give the CIV III Conquests AI an A- like grade, closer to an A than to a B+ .

Some of you may remember an earlier post where I talked about chess AI -- you can't compare it mostly because of the time, energy, and money spent on it is probably greater than all other PC games put together. Companies liek IBM spent millions, and the AI has been improved over 40 years. No commercial company can ever be held to that standard -- not the least of which, as others have pointed out, chess is a game that is easier for AI to handle by its nature.

Best wishes,

Breunor

 

[Breunor]

{Sorry, I accidentally sent twice. Second deleted}.