Civ games complexity

[Rasbelin]

Mark,

that's true, but I guess you hardly fool around with a calculator while playing Civ II. So you end up using your personal knowledge from previous experiences.

[LDiCesare]

*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.

Ever tried ICS? You never switch to offensive units, you just produce only offensive units (ok, some defenders, but they are just to defend the offensive units anyway).

The lack of caravans in the analysis is downright frightening. Trying to cover it, however, even with all the experience we may have about resources availability, trade formulas etc. would lead to really complex math indeed.

I want to point out that the number of variables is a real obstacle for the AI, but if they can be broken into categories, it becomes more manageable. In chess, you have one move to do, at most 16 possible pieces to look at, each with 0 or more moves. Go has even more possibilities and computer AIs utterly suck at this game.
In Civ, you have a choice for every unit, but do not have to choose which single unit to move. That makes for more computation but less intelligence. (I wish MOO3 still had IFPs so we could see if it gave more depth to the genre). You also have a choice for every city (what to build...), every other civ (diplomacy), and the combination of these factors is huge. Because every single one has big impact on the rest.

Quick formula of the number of options to be computed for a single turn for a despotic civ with one city (pop 1), one unit and knowing one other civ:
Civ: Tax/lux/science rates? (46 possibilites if I do the math right : 6/4/0, 6/3/1, etc.)
What to research? (usually around 4 choices)
City: Which tile to work? (20 possibilities + 1 for an entertainer)
What to produce? About 3 choices at the beginning of the game.
Whether or not to rushbuy (2 options: yes/no)
Unit: Which order? (8 moves + fortify or sentry or wait)
Diplomatic state: War, peace, make contact? If contact, threaten, exchange or offer tribute, and what? (at least a dozen possibilites, depending on the gold and tech of both civs)
Multiply: A conservative 46*4*21*3*2*11*12.
3 million+ possibilities.
Now in chess, you have 20 possible first moves, and a limit of about 120 possible moves in one turn (never reached, you'd have to be able to provide each piece every available move possibility - not considering pawn promotion).
Even in Go, you'd need a grid of 1700x1700 to have the same complexity for a single turn!!!!!!!!!!!!!!!!!!!!!!
The fact that some options are no-brainers in all games deosn't change the ratio much.

If you don't split that search space into bits, you have thousandfolds more possibilites than in Go, and the complexity goes increasing as the game proceeds, instead of lowering in both Go and Chess.

Now the real question is: Must one be silly to try to code an AI for such a game?

[Dauphin]

I think the hardest thing about human opponents in a game is predictability. With computers you know generally what they are going to do, but the reverse is not true

In a recent game of Civ3 against my brother, we had a our two homeland islands right next to each other. Rather than invade directly I sent a carrier fleet and invasion fleet all away around the world to attack him from the far side. I did this as it meant that he had to move the majority of his defencive units to a point furthest away from by lands making a counterattack against my homeland difficult.

An AI would not come up with such an out of the blue plan, or appreciate the consequences of such a tactic. It thus can't predict how an opponent will react. (Sure, you could programme such tactics, but there are a myriad, and the computer will never come up with new and inventive ones suited to a game and its situation)

[VetLegion]

Now the real question is: Must one be silly to try to code an AI for such a game?

good point.

You have shown with simple math that calculating all possibilities in a civ game is not really computable.

I dont know how to make a good civ AI.

But I think we should keep trying. Leave the ideals of pure, perfect, unbeatable AI and simply be pragmatic. Where we cant 'make' intelligence, we can 'simulate' it
(example is AI amphibious landings which are hardcoded, but are not a part of a 'bigger picture' or a long term plan)

I strongly oppose making game less complex just for AI sake, like they did in civ3, but it is a pragmatic thing to do.

[LionQ]

I guess "trial and error" is the way to make a good AI.

[Nikolai]

Don't you think that would take a bit long time? Not that I know anything about programming an AI, but...

[Odin]

A good AI is possible, but you would a computer the size of Los Angeles to use it.

[Panag]

Originally posted by Odin
A good AI is possible, but you would a computer the size of Los Angeles to use it.

hi ,



a simple updated modern PC shall do , most MAC's do excellent

have a nice day

[Shogun Gunner]

This argument has raged on in many forms in many threads. More complexity = more expensive.

AI Complexity and sophistication is more expensive due to:

• Design Time
• Programming Time
• Playtesting time
• Software iterations to correct errors, flaws, bugs
• Higher system requirements to the end user
• A different type of "cost" possibily manifests itself in a steeper learning curve/micro-management requirements in the game - but not necessarily

I'm all for more complexity and a more effective AI. However, I have already resigned myself to accept that the AI in cIV will not be markedly better than Civ3's AI. What signficiant changes in the programming world have developed in the last few years?

I would argue none, and coupled with new game design elements and features that cIV will bring, the AI has to be taught a new system. If the system is rushed for release, this feature will suffer the most. Developing an effective AI is an art. It's part programming, part testing and part experience. All these require time.

So, I accept the fact that cIV will be best enjoyed in multiplayer game, not single player.

[DrSpike]

Bumptastic.

[Shogun Gunner]

I'm a good forum user. Instead of creating a new thread on a subject covered previously, I try to find where it was discussed previously.

I'll do the bump-a-rama thing.

[shimmin]

I think a heuristic-based AI could be programmed to play the ICS well. In fact, much of its strength even for human players is that most of your options are discarded from the start as not being worth another settler, or another offensive unit.

But this does illustrate one of the challenges facing the AI programmer. The best strategies won't be known until good human players get ahold of the game and give it a more thorough playtesting than it ever could receive in production.

If the game could be shipped as a multiplayer-only game and then the AI written for it a couple years later, then we might see strong AI.

Given a good plan, computers can figure out the best way to implement that plan. Coming up with the plan is what you need a human for.

[Protra3211]

Love Chess

One of the best tbs game ever. Still Civ games takes more of my time and is a deep thinking game like chess. Have to plan and think many moves ahead or lose for lack of vison.
The AI in some Civ games can be very weak like in Call to Power2. But with modding and a new goals the AI can at times out perform human player.
Have used my chess skills to update CtP2 AI thinking. Set goals for more units to patrol its borders. Moblise forces guick to hem in human player without stripping its defense away. In chess active defense takes away offense by using a hemming in strategy. Ai's in Civ game should seek to control more open areas this way the AI stays ahead by boxing you in .

[General Ludd]

The biggest challange is to make the AI roleplay, otherwise you end up with some annoying power-gaming, super-sprawling munchkin like the Civ3 AI, and the game becomes a mundane number crunching exercise rather than an atmospheic strategy game where you lead a civilization from the dawn of time.

[Shogun Gunner]

Originally posted by shimmin

But this does illustrate one of the challenges facing the AI programmer. The best strategies won't be known until good human players get ahold of the game and give it a more thorough playtesting than it ever could receive in production.

If the game could be shipped as a multiplayer-only game and then the AI written for it a couple years later, then we might see strong AI.

Given a good plan, computers can figure out the best way to implement that plan. Coming up with the plan is what you need a human for.

Which is why I have tempered my expectations. Since cIV is very likely to be a signficiant departure from Civ3, just like every new version of Civ has done in the past, then AI has to be retaught with the new rules, features and functions. What carries forward from previous civ versions in this regard? Anything?

I hate to be pessimistic, but Firaxis had an opportunity to fine tune the AI with all the patches, updates, mid version releases (PTW, C3C) of Civ 3, but yet the performance of the AI still had significant, unresolved problems by C3C patch 1.22. It seemed to me that the method for adapting the AI and bug patches was to fiddle with the numeric values of variables (resource scarcity) and game elements (power of FP for example).

Again, I hope I'm wrong.