civ3 ai article

[.dw.]

http://www.spectrum.ieee.org/WEBONLY...ec02/mind.html

interviewed Soren Johnson (Firaxis) amongst others

sorry if it was posted before

dw

[Peter Triggs]

Thanks.

Didn't see a link in Dan's news item and couldn't find the article through Google.

[alva]

nice article

most of it goes right over my head though, phew

[Coracle]

Soren:

The hardest thing in game AI is just making sure that the game never looks dumb

I'm sure it is. Very hard. And so far, Soren, you have been unable to prevent your AI looking dumb. When the human has a deal all set with another civ, and then throws in SIX free resources as a generous bonus, only a STUPID AI civ would then insult the human and cancel the deal. It happens. When an AI civ the human has just met hates the human for something that happened to a civ that has been extinct for almost a thousand years that AI is just plan weird.

That no doubt is the result of the whacky Diplomatic AI where humans get blamed for millennia for the most minor things and occasionally for nothing at all.

When one civ is almost totally conquered by a much larger civ and left with only its capital across a sea, a major conquered city "flipping" to that defeated civ, despite the proximity of a huge enemy army three tiles away, and a large garrison, is DUMB.

There are loads of whacked out in-your-face AI things that have annoyed a great many of us. Also, invading forces are often predictable and easily channeled and fooled.

Is the Civ 3 AI better than that in Civ 2? Yes. But the Civ 2 AI was more fun. True, the Civ 2 AI did such as build warships in lakes, but I've seen the Civ 3 AI do the same, along with never using artillery on the offensive.

BOTTOM LINE: considering the number of years between Civ 3 and Civ 2 I am surprised the Civ 3 AI isn't a lot better.

[Cyclotron]

Originally posted by Coracle
...only a STUPID AI civ would then insult the human and cancel the deal. It happens.

This has already been explained. Deal with it.

Hmmm... I would address your other points, but I can see clearly that none of them, CF or reputation, have anything to do with the AI. Just another troll, then...

[Coracle]

Originally posted by cyclotron7


This has already been explained. Deal with it.

Hmmm... I would address your other points, but I can see clearly that none of them, CF or reputation, have anything to do with the AI. Just another troll, then...

Your dismissive condescension is sickening. You don't agree with something, so the poster is a "troll", that despite my many hundreds of post here. So I say to you, "just another fanboy".

CF and reputation effects are not decided by the AI? OK, maybe so. But don't tell me the game doing stupid things is entirely separate from what the AI does. I also mentioned the way enemy forces invade - in predictable ways, easily channeled; better than Civ 2, but then the AI doesn't use artillery offensively in Civ 3, which is dumb.

Do us all a favor. List specifically what the AI does in terms of game play. I don't find Trade and Diplomacy, for instance, all that smart; in fact, without all its cheats and built in advantages I don't find the AI that smart at all using Civ 2 from six years ago as a point of reference. (Oh, don't tell me the AI doesn't cheat).

I expected an AI a LOT better than Civ 2. It is not. That is my objection, and disappointment.

[Cyclotron]

Thanks for responding. It's a first.

I also mentioned the way enemy forces invade - in predictable ways, easily channeled; better than Civ 2,

There seems to be mixed feelings on this. There have been posts from people who have witnessed focused invasions- for example, the writer of the piece above at the beginning of the article. I can't say the AI has done well every time, but too I see a marked imporvement over Civ2. How big that improvement is is subjective.

but then the AI doesn't use artillery offensively in Civ 3, which is dumb.

It uses ship and plane bombard very offensively. As for ground artillery, I have seen limited use by the computer and I agree that would be a good place to resume improvement.

I have already explained in threads on the issue how the "absurd" trade deals you speak of actually make sense. I'm happy to engage you further on the issue in a more relevant thread.

Cheats and advantages are in use by every game I know of, since even the best AI does not stand up to a human player. It seems to me that Civ2 had far more in the way of cheating than Civ3. Again, as to how intelligent the AI is compared to Civ2 or anything, that's just opinion.

Let me make something clear: I did not, as you would like to think, call you a troll because you disagreed with me. I called you a troll because you were complaining about topics completely unrelated to the topic of the thread, and this is hardly the first time you've done that.

You think the AI is stupid? Fair enough. But please, try not to put things like your hatred of CF in practically every post you write.

[Cyclotron]

Originally posted by Coracle
that despite my many hundreds of post here.

Ah, multitudinous posts does not intelligence make... Just check the OT sometime.

[Aeson]

Coracle,

The AI is smart enough to let you capture it's cities (leading you on with it's 'dumb' military tactics), so it can wipe out your forces and make you quit after a culture flip.

Seriously though, the AI is so much better than it was in Civ2. Even with much the same 'cheats', Civ2 AI was beatable, OCC, any difficulty level, every game.

[Ijuin]

Regarding how reputation is handled in Civ 3, I do think it is utterly stupid that your reputation falls when a deal is cut off due to circumstances beyond your control (such as your trade route going through a third-party Civ, who then declares war on either you or the Civ you are trading with). I mean, when Iraq invaded Kuwait and cut back Kuwaiti oil exports, did the world get angry at Kuwait for not exporting, or at Iraq for preventing Kuwait from exporting?

[Jaybe]

Ijuin, while I basically agree regarding a player's rep being decapitated, understand also that without VERY/involved/CPU-hungry sophisticated programming ...
it would be too easy for a player to exploit.

Life is tough and unfair, in some small ways so is Civ3. Some people would like natural catastrophes, mini ice ages, and plagues added in.

[DeepO]

In games "it's like this big loop: build a little bit of AI, see how it does, make it better, see how it does, continually and continually testing," he says, "whereas in the academic community, there is no real reason to do any sort of iterative process. The important thing there is that your AI is as close as possible to the model that you're trying to study." Academic AIs have a related advantage over their game cousins—they do not have to be shipped on time as an integrated component of a polished product.

Actually, I disagree on many points with Soren here... academic AI not being an iterative process? AI is always iterative, it only works when you do something, build on it, view the results and go back a few steps to adjust your initial assumptions. There is no difference between academic and industrial (or game) AI in this respect, it's always part of the process.

Further, Academic AI would only be tested to a model, where game AI would relate to the real world?? Strange... I would consider it the other way around. I agree, some academics work from a model, but the biggest part of AI is trained or modeled to reality, not to a simulated model. Of course, new techniques are better of if they first are compared in controlled environments, to existing benchmark problems, but once this step is done, no academician would object to throwing his algorithm out in the open. Or at least no-one working in engineering would, which is still the faculty where most of academic AI is developed.
OTOH, game AI is always related to a model, and in more then one case the model (=game) gets adjusted to the AI, so that the combination of both work better.

As to time pressure: I can imagine why Soren says this, but I have to disagree. I've worked in industry before, and am working in uni now, and while the time pressure is different, it's twice there. If new algorithms don't produce the results in time, you got to work harder to make it work, before someone else publishes about them.
He has one big point, though, and that is that games are pushed out the door on merciless time tables, just to make it in time for Christmass. That's where the real problem lays of game AI: the current publisher-game developer relationship, where a publisher knows that any game has to have the latest eye candy, which inherently means it won't have the best possible AI (as this is done at the end of the development, when the graphics are done, and 'waiting').

Other Industrial AIs don't have this problem, as time pressure is always present, but not taken to the extreme... which is healthier in all respects, but it does require a customer who will value quality above all, which the current game community does not.

I've got some other remarks on the rest of the article as well, I don't consider it that good at all. It is clearly written by someone who knows small pieces, but is no expert in AI, with the usual simplifications. Finite states with 'less finite' states? Use fuzzy logic. Yeah. Right. Adaptive AI needed? only Neural Nets can help... Right again. But at least it tries, and it has some nice pointers, so thanks for the read.

DeepO

[Thrawn05]

From one programmer to another... you have once again amazed me DeepO.

Perhaps I should hit the books again some time.

[Datajack Franit]