Civ games complexity

[Solver]

Anyone able to provide some facts/arguments on how mathematically complex civ is? From what I see, it actually seems to be more complex than games like chess, because of the way more variables in situation.

Also, the best chess computers are on same level with best chess players, and I'd expect computers to be ahead in several years. However, even the best civ AIs aren't still much of a match to the best civ players.

This actually goes after trying to tell a chess player than civ can be as mentally challenging and complex as chess is, if not more. So arguments please.

[spartak]

Solver

have you allowed for the fact that the besxt chess computers are optimised for the exceedingly fast computers they ruun on while an AI has to be able to work on any old crap?

[Solver]

I don't think anyone would code an AI good enough still for civ...

[DrSpike]

The point is you can't program rules that are really effective for a vibrant game like civ.....because it usually depends on the situation, and AI cannot adapt.

In chess computers have the upper hand over all but the best humans because computers' short term crunching of variations is unbeatable. Computers still lack long term understanding, which is why they can be beaten by good players if they force them into long term positional battles.

But it still isn't easy, because tactics usually play a role in converting any positional advantage......and the computer will usually find some ingenious tactical ways to extricate itself.

In addition you can tell a computer what openings to play, and the top programs can play all basic endgames flawlessly. This last point is a key one I think.......the deterministic nature of the game allows the machine to be fed what it might have difficulty in working out for itself.

No such thing in civ.

[War4ever]

have to agree with spike, what makes civ so good is the human players ability to read situations and adapt to them through various playing styles, the ai , simply cannot compute this at this point.

all ais have flaws in them, even to take all the best civ players and mold it into an ai, wouldnt' work....

there are always loopholes and bugs ......

[DrSpike]

Originally posted by War4ever
have to agree with spike

Always said you were an insightful chap.

[Lemmy]

IIRC the chess supercomputers are a different AI then the civ AI's.
The chess AI is so good because it can check every possible permutation for the next few turns, and select the best move according to the outcomes.
The Civ AI does not compute ahead, it has to decide on the current variables on what is the best move, and then it often comes down to how well the programmers knows the game, to decide the best move. Coding a good AI for Civ isn't difficult if you take the chess approach and just try out all the possible moves....but then you'd have to find a computer fast enough to run it on .

A possible approach to create a good AI for civ, is to make it a case-based learning AI, and basically apply the best solution for each case that arises. If there isn't a solution yet for a particular case, there can be a simple fast routine to create a solution, which most likely isn't the best, but the outcome will provide data for future similar cases. This would mean though that the AI for each installation may evolve differently, and it would depend on the player actions.

[War4ever]

yeah i agree lemmy.....i stated similar thoughts in a different thread on this

[Lemmy]

just read your post

[Solver]

Yeah, an AI that could count on every possible move in Civ would require some IBM supercomp to run .

[DrSpike]

It's not the combinations.......hell, millions of possibilities can be considered no problem. Rather it is the inability for the AI to understand how those possibilities interact........what in the other thread Mark_Everson calls non-linearity in the decision space.

[Solver]

Yeah... because there's no way a group of programmers could predict it all.

So with the current technological base a perfect AI is impossible.

[Rasbelin]

Erhhmm... if you would visit DMOZ, you would know the "A mathematical analysis of Civilization II" writing by a British college maths student. Well, I have to say it's good reading and proves how Civ II uses algebra to calculate all triggers, etc.

http://www.adk24.ucam.org/civ2.html

[Mark_Everson]

Hey Ras, that's interesting. I did a similar analysis a long time ago to figure out what was best to build at what point of city development in civ2. Having this sort of info does indeed make writing a good AI much simpler. Essentially you do a cost-benefit analysis of each potential building or military unit, and pick the best. I don't know if Soren's approach was similar or not.

[CapTVK]

Mannon beating out god when it comes to happiness!? Yes, that's about right. Money, is the ultimate supply in civ it's sort of a mobile resource pool which can can be used any time, any place.

Although I have to say that this analysis only is valid if you start out with all the techs. Plus it doesn't factor in time and the value of sacrificing units to speed up production. Let the alone factoring the value of military units for defense and offense*.

Ultimate irony is that if you tried playing according to the true algabraic model you would quickly be toast because building settlers/caravans and cities/improvements is far more profitable than units.

*It's my belief that any prolonged war will never be profitable in Civ. The loss of units in the field and shifting production to offensive units is simply to great to be compensated by the gain in production/gold of captured cities.