Civ III AI - This explains everything (at least to me)

[Theseus]

I look forward to kicking your butt in MP, jt... that is, on the days when you don;t kick mine!!

[Jon Shafer]

We can all just kick each other around in a big circle!

In six months...

[Captain]

good job Soren, given the time and financial constraints. the ai is good, and downright impressive sometimes. too bad they didn't give you more help and more time. I'm curious as to what you'd have liked to do, if you weren't working in an 8 month time frame.

what would you change, or have added, to civ 3 if you were free to do so and had unlimited resources? what's soren's wish-list?

[Grumbold]

I'm impressed with the domestic expansion AI. The complaints here have all focussed around players being "forced" to rush colonise because the AI is too effective if you adopt any other strategy. Of course it is helped by its omniscence in knowing that there is available city space the other side of your empire but still, a very competent job.

The combat AI still has a lot of holes in it so I can't say the same there. Bait and trap with workers will catch the enemy every time. Artillery tactics are improving but still woefully poor.

All in all, its the best out-of-the-box AI from a random map TBS to date. Unfortunately the game lacks the extra functionality that has allowed dedicated fans to turn CtP1/2 AI into rampaging beasts capable of avoiding - or even using against you - all the players favourite tricks.

[Adm.Naismith]

Originally posted by Trip
We can all just kick each other around in a big circle!

And that will test players' Natural Intellingence?
Would you then tell that players who lose often where "rushed out"? How bad parents: who is the publisher to blame?

And about frequent winners?
"nine months of developement all done from a woman, after some starting input and general design by a man. Quite an accomplishment. Take this Mr Soren!"

Just kidding

[Adm.Naismith]

Originally posted by Captain
I'm curious as to what you'd have liked to do, if you weren't working in an 8 month time frame.

what would you change, or have added, to civ 3 if you were free to do so and had unlimited resources? what's soren's wish-list?

Good question. I'm afraid Soren can't answer it at his pleasure, because some concept can be in development for XP, too, so not disclosure agreement can be a pain in the neck here.

So, softening the attention from Civ III to TBS genre, where do you think AI development for TBS can gain more, Mr. Soren? Where results are more dependant from new algorithm, more computational resources, different approach, more link between game rules design and AI development (to mask limits and build more on stronger points)?.

[BillChin]

Re: Re: Civ III AI - This explains everything (at least to me)

Originally posted by Dominae
I see no evidence to support this. To create an AI that plays the game as well as it does requires intimate knowledge of all things Civ. I find it hard to imagine that Soren isn't in the top 1% of all Civ players, if you want to look at it that way. People who "write the book" on any topic are invariably experts.
Dominae

My evidence is hearsay. However, I do remember Mr. Johnson himself saying he is an average Civ III player on one of the chat sessions or interviews. I also remember someone saying that no one at Firaxis could beat Deity level more than once in a blue moon when the game was released. The top 10% of Civ III players can do better than that, so probably no one at Firaxis was in that category at time of release.

The other evidence is the AI engine itself. There are certain design decisions such as combined arms that an experienced wargamer (or Civ player) probably would have made a high priority instead of an after thought. Several patches after release, the AI still has no clue about combined arms. The AI sometimes can be observed running around with a single artillery unit escorted by a single defender, but this is perhaps the least effective way to use bombard units on the offense. On other threads I have suggested relatively easy ways to code the AI to include bombard units in an attacking stack without leaving them open to easy capture. Combined Arms is just one example, there are many more available.

When I factor in what I view as the averageness of Mr. Johnson's gaming skills, the AI is even a greater achievement. Kudos are in order.

[hexagonian]

Originally posted by Grumbold
All in all, its the best out-of-the-box AI from a random map TBS to date. Unfortunately the game lacks the extra functionality that has allowed dedicated fans to turn CtP1/2 AI into rampaging beasts capable of avoiding - or even using against you - all the players favourite tricks.

The CTP2 Modder community thanks you for your kind words

[Dominae]

BillChin, thumbs up on a very good reply.

Of course, I do have a couple of comments...

The game engine is an interesting place to look for the shortcomings of the programmer. It is also very misleading.

First, humans will always use available resources more efficiently than the computer (if they try hard enough). Why? Because a human player can learn from the AI. Although I admit that artillery is quite an oversight, there will always be something the AI "doesn't do" as well as us people. This does not reflect that programmer's gaming ability, as coding Civ3 is much harder than criticizing it.

Second, the AI was designed to be fun, and not simply challenging. People are forgetting this. If the AI gets so many free units at the beginning of the game, shouldn't it trounce us every time? I've often wondered why the Zulus, with their 10-15 Impi to my 6 Warriors, don't choose to wipe me out immediately. The answer is that it wouldn't be as fun that way (although it does happen for time to time to keep us on our toes). But if I started with extra units, I promise you I would beeline for the closest AI civ and eradicate it.

Third (this point is closely related to the second above), if the true gamers could design the AI, Civ would be boring. By "true gamers" I here refer to what you call the top 1% of the civ community. If this were the case, we'd have the AI ICSing full time, creating naval blockades till the cows come home, pounding cities into the ground with nukes, and plenty of other stuff that just isn't fun for the average gamer. Yes, it would be cool if Deity were like this. However, I firmly believe that such an AI is within Firaxis' capabilities. The reasons not to include it in Civ3 are surely (IMO) other than lack of gaming experience on their part.

Fourth (and most importantly), the Civ3 AI is sufficiently good for 95% of players out there. Coding an AI that accomplishes demonstrates knowledge of strategic game mechanics as well as "fun" game mechanics.

Anyway, I'm not going to push this issue further. I'm simply tired of people criticizing Firaxis left and right (as I unfortunately do, from time to time). From your previous posts I know that you're not a whiner, so all is well.


Dominae

[hexagonian]

...and Dominae,

Your post is an excellent response to the whole issue of how hard it is to create an AI that would truly challenge a human player - something that I believe cannot be done at this time to the extent that everyone seems to want.

The problem is that the AI, for all its number crunching ability and built-in bonuses, simply cannot plan out a long term strategy. It uses built-in cheats to simulate a thought process. Cheats have to be the norm at this point in time - and the AI cheats are countered by a human having the ability to exploit the way an AI is coded to react in a given situation, because ultimately, the AI is a program.

Take this situation and try to figure out how the AI would figure out what is the best plan of action...

An AI needs to be programmed to take a particular route to get at a target. The question needs to be asked, 'What constitutes the best path to get at that target - the fastest, or a path that may take much longer, but may pay off with a greater gain because it may take the player off-guard?' The choice will vary from situation to situation and is actually a very subjective matter to determine - ask 20 different players and you would get 20 different responses. How do you program the AI to think in a subjective manner?

All everyone has to do is look at the howls of outrage when it was discovered that the civ3 AI was able to see where your defenses were placed in cities and was moving its forces to the most lightly defended city (an obvious cheat). It is interesting to note that once that characteristic was noticed, players had come up with a way to exploit that feature against the AI too.

Your second point is a good one, because it is so true. If the AI played to win (especially with those bonuses), those enemy forces would be on your doorstep ASAP and would be crushing you into the ground. The goal of the human player is to reach the same point when it can do the same against the AI - so the balancing issue for game companies is to give the semblance of a challenge without totally unleashing the AI.

[Jon Shafer]

I think those fools who challenge deity should have their civs brought to their knees by 3000 BC.

[Coracle]

hexagonian,

The reality is that never in Civ 2 did so many people complain about the cheating AI. But Civ 3 is so blatant and irritating in the manners in which it cheats that it kills a lot of the fun. What the hell's the point of being "challenged" by an AI that breaks its own rules and does outrageous things in order to be "competitive"?

Make no mistake about this either, the Civ 3 AI does STUPID things, not just cheat. It wastes a lot of its resources sending out settlers to every crappy piece of land anywhere, it throws away settlers marching them into enemy territory, it makes idiotic attacks because it is coded to always go after undefended cannons or workers or resources, etc. These and other stupid AI moves do not need to be in the game; they are there as Soren rushed this thing to market at the behest of his Infogrames masters.

[monkeyman]

What I want to know is whether it is possible to use genetic alogorithms to select and adapt the ai to get a better fit to what happens in the course of a game or series of games? That way, if you start playing in a rut, some variability in the ai might crop up and force you to change your play style. This is in effect what Soren is doing iteratively with each patch, but Soren's manual approach is (somewhat) more predictable in its effectiveness (but not necessarily ala communist collapse of ai in 1.17f). I guess this is also what the CTP mod community is doing. Again, could this approach work out of the box?

-mm

[Dominae]

monkeyman, no genetic algorithm (that I know of) could at present help "fix" the Civ3 AI.

A genetic algorithm requires some framework to determine where "mutations" occur, and what exactly they consist of. For example, if the state of the game could be represented as a bit string, it would be simple matter to change some of the bits so that the AI could alter in subtle ways and eventually adapt to the human players. There are two problems.

First, mutations, taken individually, would have to cause insignificant changes to the AI's existing behaviour, or the human player would, more often than not, playing against an AI that was altering its behaviour in seemingly stupid ways. But insignificant changes, taken together, require too long to evolve into a "good" version; the human player would have to play literally thousands of games to see any effect.

Second, Civ3 is too complex to represent as a bit string, so classic genetic algorithms are out. The "mutations" would have to occur over some complex structures (i.e. game mechanics). For example, slight changes could be made to the pathing algorithm to determine the "best" way of getting from point A to point B. But again, it would require so much time to train the AI to learn what is "best" in various situations that players would get bored with their "stupid 'genetic' AI" (as I can hear them say already!). Furthermore, applying genetic algorithms to these complex structures is (I think) possible, but certainly no piece of cake. It would definitely be more work than Firaxis is being contracted to do.

The Civ3 AI is inherently "simple", but still manages to generate some pretty complex behaviour. The fact that we humans learn to adapt to it is a credit to our incredible brains, not the AI's "stupidity".


Dominae

[Dominae]

hexagonian, very thoughtful post. The Civ3 AI is a perfect example of artificial intelligence that is in no way intelligent.

The designers made a conscious decision to create a game-playing program that would be competitive with humans. To do so they anticipated human strategies and weaknesses, and coded against them. They also coded in some "cheats" to ensure the AI's competitive edge.

These are essentially rule-based constructs. As has been said many times before, creating a game AI that is as difficult and as fun as that of Civ3 without resource to any fancy algorithms warrants a tip of the hat to the Firaxis team.


Dominae