DESIGN: AI Improvement discussion

[Immortal Wombat]

The fact that you have to have Gunpowder in order to get Tank Construction (and since Tank Construction is also on the list, no one can aquire it by trade) means that that kind of situation usually doesn't happen.

[Solver]

Yes, but don't you think that any rule as strict as "never trade xx" is bad? It should be better to allow the AI to, again, determine comparative values...

[Immortal Wombat]

Yeah, probably. But it makes for a fairly good and very simple stand-in.

[The Big Mc]

no who in the right mind would give the ability to make a nuke to some one for a peace packed

[Dale]

St Swithin:

Which pathfinding would you recommend?

[ToonGoon]

I am neither a programmer or SLICer but I am enjoying these discussions so much that I did a search for more info.

There look to be a few good links here.

[Martin Gühmann]

Originally posted by MrBaggins
Measurements at a strategic level produce 'needs' or tasks at a lower (tactical or operational) level;

This is the problem with the CTP2 AI, it doesn't have top level goals like:

Conquering the World
Achieve the Gaia Science Victory
Build up an world alliance
Just survive to the end of time

Usually a human player has one of these goals, when he starts a game and is able to change his goal during the game.

All these top level goals imply lower level goals. To conquer the world you need first a production base for units, so you have to build some cities. These cities need buildings to increase production and population that will also increase production. You need a road network to bring troops to the front, you see only roads between cities of the AI but never roads to the front.

The current AI can't think on top level goals it just thinks on the very low level goals like sending an army to an enemy city, it checks for armies without goals 5 to 6 times wich new goal could be assigned. So the AI will conquer by accident the enemy if it has enough units to send into the slaugter. A higher level goal would be here conquer this particular civ, a lower level goal would be then: Conquer this particular city, now it should check how many armies are available should these armies rearrenged, are transporters needed, are transporters available, are enough troops available and so on.

So here is a list of goals that are needed to accomplished so that world conquest can be accomplished:

Build cities so that the civ can get new production bases
-Make cities grow so that the production output can be increased
--To make cities grow build farms and fisheries around the city.
-While city grows secure this city and make shure new cities can be founded
-To avoid that newly founded cities can be conquered some Defenders must be build before production can be switched.
-To accomplish above goal Settlers are needed so build settlers in existing cities.

Well it looks like this is a long list of things to do to achieve world domination. But this is stratigical thinking than moving some units through the area and hoping the AI gets an advantage.

-Martin

[Solver]

Or maybe go like this. I have a lot of good commerce tiles and trade goods in my land. I should go for Trading Posts. I have a ton of commerce, and don't need the cash. Build science improvements. I have a ton of science. Only one civ is bigger than me, but my relations with it are good, and it's not aggressive. So, I should try to go for science victory...

[Myrddin]

Most players would settle for an AI that is competent at going for a conquest/domination victory (as long as it was smart in using diplomacy to help this)

Even in Civ3, the AI doesn't generally head for other victories, except for diplomatic, where the second place civ does seem to try to get the UN to win

[Locutus]

Gee, Martin, why does that sound vaguely familiar?

[Solver]

Cill Wouter, now we can actually do it and anything else with the AI .

Almost...

[st_swithin]

I would have suggested the original fuzzy logic pathfinding system in C:CTP, but it got stripped out by the subsequent AI programmer, who deemed it too highly obfuscated to be of any use.

Always made perfect sense to me, though. I don't know how much of it is still resident in the CTP2 codebase.

[Locutus]

Fuzzy logic rulez!

[Solver]

Loc, you can understand it?

[LDiCesare]

1. Generate your enemy threat influence map (based on areas, or sites as I call them, of highest enemy threat).
2. AI determines which site to send unit to.
3. Combine enemy threat & terrain influence maps together (to get the influence of enemies and terrain obstacles).
4. Through the generated influence map into A* and watch it weave a path from source to target avoiding all other threats and obstacles (to ensure the unit reaches the intended site) till it reaches the target site.

This is the path I'm taking.

You'd have to re evaluate that every turn in case the opponent moves their units or loses some of them: The map totally changes in such a situation. That's also a problem with the regular path-finding, because a 12-stack on an isthmus can block a path, move away and open it again, making it possible to lure a path findign algorithm that takes into consideration dynamic things like units. This is VERY tricky:
Imagine the ai computes the threat map, and then decides to go some way. The player then moves their units to a place which was not defended, and the ai moves right into them. Unless it evaluates the path every turn, in which case units moving in and out of chokepoints can make the ai path oscillate.
How do you solve that? You have to consider that some changes to a path must be done, and too many changes to a path mustn't be.