{The List} AI

[Kuciwalker]

Can't you do QA after you build the AI?

[okblacke]

Originally posted by skywalker
Can't you do QA after you build the AI?

Indeed, you'd have to, unless the AI were a truly separate entity. And you'd run the risk of completely nullifying the AI in the QA process.

I think it's safe to set aside any notion that Civ 4 will contain radial AI enhancements. If it could just manage workers better, improve governor AI, city placement and beef up diplomacy and trade, that'd be, well, respectable.

[ok]

[Kuciwalker]

Indeed, you'd have to, unless the AI were a truly separate entity. And you'd run the risk of completely nullifying the AI in the QA process.

1. why couldn't you freeze it prior to the QA process, then?

2. if you can't do number 1, why not just back up the AI and if you kill it, load the backup?

[MrBaggins]

Originally posted by skywalker

Quote:

Indeed, you'd have to, unless the AI were a truly separate entity. And you'd run the risk of completely nullifying the AI in the QA process.

1. why couldn't you freeze it prior to the QA process, then?

If you discovered a significant flaw in the AI, its development would have to continue through another QA cycle.

2. if you can't do number 1, why not just back up the AI and if you kill it, load the backup?

What do you mean? A single track of AI development?

That would develop a strain specialized to one circumstance... which will mean a suboptimal AI for where the circumstance is different.

[okblacke]

Originally posted by skywalker
1. why couldn't you freeze it prior to the QA process, then?

Because the AI would have "learned", for example, that it's important to place the Forbidden Palace outside the ring of influence of the Palace, and then in QA you discovered that that was imbalancing and removed that feature from the game. The AI doesn't know that.

2. if you can't do number 1, why not just back up the AI and if you kill it, load the backup?

You didn't lose the AI, you nulled it, rendered it incorrect. It learned what it learned by "observing" what happened under a certain set of rules, then you changed the rules. Get it? You'd have to rebuild. With every rule change.

You could (arguably) make it so the AI "learned" the rules as well as the best application, but that increases the work it needs to do by about a zillion.

What we fail to appreciate is how quickly we can adapt to a severe rule change like those introduced between, say, C3 and C3C. The easiest way to create a good AI ("good" AI? good "AI"?) is to create within a set of known rules: The more flexibility you demand of it, the huger the problem becomes.

Think of the difference between programming a computer to play tic-tac-toe and programming it to play Connect 4. And then think of writing a program that can do both, and you have some concept of it.

[ok]

[Peter Triggs]

Exactly. Think of the difference between CTP1's AI and CTP2's AI. Most players found CTP1's AI to be reasonably challenging for a non-cheating AI (although it's not as much fun to play against as Civ2's AI, which cheats). On the other hand, almost everybody said that CTP2's AI is too easy to beat. Why? Well I've always thought that one of the main reasons for this was that they added a new diplomatic victory condition to CTP2 but they didn't have the time to make the AI flexible enough to properly handle it and so it unbalanced the game.

[Nikolai]

Bumping this one, in case someone who have missed this have more ideas.

[alva]

Originally posted by Mark_Everson
FWIW, here's my list of what the AI needs to do in a civ-type game to get the breakthrough many of us want. Many of these have been mentioned already above. We're not going to get anything like this out of Civ4 :

AI that thinks in 'Levels' about strategy, from very high level down to small details. This is an Hierarchical approach

Levels will go from very abstract pictures of the world at high levels, down to pretty accurate copies of the world, for small things

Corrections will be made in the higher level AI to incorporate information from the more detailed AI levels below it.

* snip *

Isn't that what Soren (CIV3 AI programmer) is saying?

Apolyton: Does the AI make decisions on large scale, small scale, or both?
Soren Johnson: The AI makes decisions on many different levels. A leader AI keeps track of high-level concepts, like whether enough science is being generated or which civ is the biggest threat. A city AI decides which units to build for that area of the world and how to manage the city itself. The unit AI is itself split into multiple levels, first determining an ideal destination and then calculating the best possible path.

Furthermore, the different AI levels communicate between each other. For example, the unit AI controlling a tank might ask the leader AI which enemy city would be best to attack. After receiving a response, the unit AI would determine its own best path to the destination. Thus, the AI is able to organize a large group of units while keeping the discrete game decisions at a lower level.

[Mark_Everson]

Thanks, alva, I hadn't seen that quote. I think it's pretty clear that the AI in Civ 3 is better that in the previous games. Some of these points may be part of why that is so. What Soren describes is still not as sophisticated as I'd like, but it is certainly headed in the right direction.

I'd also like an AI that doesn't Cheat, but I certainly prefer a competitive cheating AI to a lame honest one.

[alva]

Full Soren interview

I'd also like an AI that doesn't Cheat, but I certainly prefer a competitive cheating AI to a lame honest one.

IIRC he says something about that too in the interview.
Something like "It's ok to cheat as long as you tell the players about it".

the interview has been on this site for a couple of years now.

[LDiCesare]

The unit AI is itself split into multiple levels, first determining an ideal destination and then calculating the best possible path.

That's not really different levels, that's different routines. Not taking the pathfinding in the initial evaluation of a target would be a mistake by the way, so I think they do some kind of rough pathfinding first, which would make sense in terms of 'multiple levels'.
The leader/city/unit ai's aren't the same as strategic levels of ai's. They manage things the way the player manages them, but not the way the player thinks them. So there is a difference in approach. In particular, Soren mentions no coordination of the various city ai's to produce consistent units among the whole civ, but maybe the term region he used ment that. I also think that the ai cheating on intelligence is something which is probably overdone in civIII and other games like Galciv (knowing resource locations for instance).

[Addled Platypus]

How about when you are allied with one Civ AI

You are at war with another Civ AI

The next thing you know, your enemy has all the advances you traded with your ally


talk about donkey punched

[Brent]

Have a pregame or editor setting whether to allow AI civs to build cities in inhospitable locations.

AI Civs should view trespassing differently on water than on land. Be able to keep your units from entering rival territory.

My AI controlled cities shouldn't build any more defensive units once they reach a certain number of them in the city.

Allow automated workers to build distant colonies. Make building roads toward other civs a priority for automated workers.

Let settlers find their own site to build on.

Have minor tribes fight eachother.

[Lawrence of Arabia]

introduce different long term and short term goals for the AI

At the beginning, all civs should be striving to win the game, and thus will do deals/actions that will put them in that position.

later in the game, relatively weaker civs will have a long term goal of 'survival' aka be everyone's *****, make deals that will more likely keep them alive, be more cautious etc.

short term goals can be goals such as 'acquire new cities' , 'build new cities', 'acquire coal resource on the other side of the mountain range' etc.

these goals can be random, and the AI wont get stuck on one short term goal. they try it a couple of times and if it doesnt work, they will switch short term goals.

short term can be defined as a goal which can be accomplished within the acquirement of the next technology (or next two depending on how much progress is made on the one currently being researched)

and to add more complexity, add maybe an 'era goal' one which would span the length of an era and have such goals as 'defeat neighbor,' 'become 2nd strongest military' , 'build three wonders', 'achieve x culture' etc.

[Addled Platypus]

Brent-
AI Civs should view trespassing differently on water than on land. Be able to keep your units from entering rival territory.


I am split on this aspect.

I do get mad at try to get around their territory at the triemene level...

But then there is the "territorial water" claims of nations in real life.

Look what happend to three ships of the British Navy ..
If you have not heard..Iran's navy captured Three ships, because
they were in "their water "without permission

[Addled Platypus]

I need to vent about "AI"

Just lost two Frigates that were built in a habor, to a Triemene

[Master Zen]

I have to start by saying that the weakest link in Civ has always been the AI. It is frankly terrible and the only reason the harder difficulty levels are hard at all are becuase of the cheats and handicaps that the AI is endowed with.

I won't even mention how bad the AI was in Civ2, but the Civ3 one wasn't particularly better. I still want to see the AI use combined arms effectively and especially, learn to do massive naval invasions instead of sending the lone transport with destroyer escort. And of course, learn to use artillery, somethign that is missing completely in Civ3.

[Nikolai]

DC: I have looked through this part of the List now, and I think it's not much more to do now. I'll make a search for "AI" now, but if I don'tr find anything more to add then, I guess it's finished for now. The AI list haven't been very active, but some ideas thankfully have come, now I only hope more activity will come before the final List.

[DarkCloud]

thank you nikolai- I'll get to work on adding this posthaste

[Addled Platypus]

Does anyone else notice if you pillage enough, the AI will stop making the same improvement?

[Nikolai]

I haven't noticed. Weird...

[DarkCloud]

From another thread:

3.9.0 Variable AI
A good way to simulate this might be to have a variable ai...so at random times for periods of 20-40 turns the ai of a civ will move up a difficulty level, and a t random times for a period of 20-40 turns the ai of a civ would go down a difficulty level. You could maybe tie this in with the first suggestion - during its golden ages the ai goes up one difficulty level and in its dark ages it goes down a difficulty level. All this would occur pretty much in the background and be fairly unnoticeable to the player (although dark/golden age notification for other civs as a popup/newspaper would be cool.)
-Comrade Raoul

[patcon]

1. The AI needs to respect borders. If I'm in their territory I get 1 "OK, I'll leave" and then it's "Leave or war!", but the AI gets 5 (or more?) chances to ignore my polite requests to leave, all the while moving deeper into my territory, until finally the auto-leave takes them all teh way across ny territory to where they were wanting to go anyway, and now they got there even faster. The Civ2 auto-leave where units got sent all the way back to the capital might help alleviate some of this abuse. (Although I will have to admit that I sometimes used that to redeploy my troops closer to home.)

2. The AI needs to recognize the value of trade networks. I have all too often had AIs with resources to trade who refuse to connect their road network to mine, even if I have gone to the trouble of building a road across tiles of open territory just to reach their borders. Then if I try to bring in workers to build a road in their territory to connect up, I am told to "get out". Also, the AI tends not to build harbors that could establish trade routes.

3. It's one thing to have an AI that cheats by knowing the whole map, but I think it is complete and utter bovine feces that the AI knows where resources will pop up, once they discover the appropriate tech. This leads to the absurdity of the AI establishing cities in tundra or desert tiles because oil will eventually show up there. Does anyone know if there is a time constraint on this, as in the AI won't go after a resource it can't use yet more than X turns before it discovers the appropriate tech?

[Nikolai]

Updated, check out in DC's thread with the word file, I chose to update directly there.