AI military strategy can we improve it?

[Nares]

Maybe I'm not doing a good enough job explaining unit attrition.

The AI will not engage in stalemates or worse odds when attacking, because it's defined not to in the XML.

It almost never gets good odds when attacking because friendly territory bonuses are so large.

The only way it can get good odds under the current XML defined settings is to have a tech advantage of 1-2 eras.

The result is that without a beastly ranged tactical AI, it cannot effectively wage wars at all, because it is defined not to in the XML.

This has nothing at all to do with 1UPT. The AI would appear just as inept under a stack ruleset because the tactical AI is so poorly applied to combat.

In fact, it would be worse under stacking because it would always compare against the best defender, which would always return sufficiently poor odds that it could conceivably never attack a enemy unit ever.

 

[Ramstorp]

The AI will not engage in stalemates or worse odds when attacking, because it's defined not to in the XML.

It almost never gets good odds when attacking because friendly territory bonuses are so large.

The only way it can get good odds under the current XML defined settings is to have a tech advantage of 1-2 eras.

So, why not reduce defensive bonuses? I weakened the stone throwing old lady in the city and made her range 1; gave AI higher unit production speeds; raised some other numbers here and there in the aistrategies folder. Now I get pleasantly killed after a couple of eras in some wonderful dogpile war. But I am not that good at playing the game anyway, so maybe it would be different for those who know how.

 

[Bibor]

I already did some theoretical work on how to improve the combat AI, right after Civ5 came out. I'm not a programmer tho so I cannot implement it.

http://forums.civfanatics.com/showthread.php?t=390641
http://forums.civfanatics.com/showthread.php?t=390909

 

[Nares]

So, why not reduce defensive bonuses?

It still won't engage at stalemate or worse odds. Literally it is defined only to engage in attrition combat (stalemate or worse) when on the defensive.

The change is as simple as adding an offensive weight threshold tag to the three attrition combat entires in the tactical AI XML. Obviously it's a rudimentary change, but it's a key component to increasing the AI's combat threat.

 

[lschnarch]

In fact, it would be worse under stacking because it would always compare against the best defender, which would always return sufficiently poor odds that it could conceivably never attack a enemy unit ever.

Who says that stacked combat has to take place in a 1vs1 system with always picking the best defender?

It still won't engage at stalemate or worse odds. Literally it is defined only to engage in attrition combat (stalemate or worse) when on the defensive.

The change is as simple as adding an offensive weight threshold tag to the three attrition combat entires in the tactical AI XML. Obviously it's a rudimentary change, but it's a key component to increasing the AI's combat threat.

First, this would mean that the developers don't know their own program, as they have increased the city defences lately.
Second, even if you make the AI attack against odds, how would that make it magically let its units fight in joint effort?
Third, and most important of all, my experience with the game tells me that the AI indeed does attack against the odds. It just doesn't attack with combined forces, but sends them here or there, finally ending being confronted as isolated units by the human player's combined forces.

 

[Nares]

Who says that stacked combat has to take place in a 1vs1 system with always picking the best defender?

That was how it operated in CIV.

However, the AI wasn't crippled by being unable to consider attrition combat when it was the aggressor.

First, this would mean that the developers don't know their own program, as they have increased the city defences lately.

The changes to city strength and healing were an extreme reaction to a particular complaint/strategy (namely Horseman rush, but Ancient/Classical era warfare in general). Also, the departure of Shafer suggests a change in leadership of CiV development, which could explain a non-continuation of the Shafer design philosophy or the implementation of immediate if heavyhanded changes.

Second, even if you make the AI attack against odds, how would that make it magically let its units fight in joint effort?

One of the changes in the Play With Me mod is to increase the flanking bonus because the AI actually does use its units in conjunction with eachother effectively enough to be passable.

Third, and most important of all, my experience with the game tells me that the AI indeed does attack against the odds. It just doesn't attack with combined forces, but sends them here or there, finally ending being confronted as isolated units by the human player's combined forces.

I believe what's happening there is that individual units are effectively turned back by checking for and finding unfavorable odds.

Most of the AI invasions I saw did start with some sort of unit cohesion. They broke down as advance units ran into poor combat odds and began repositioning (literally the command from tactical AI XML). You notice their offensive splintering, but the point at which the AI begins to splinter is much earlier than when you begin capitalizing on lone units, etc.

I'm not claiming to have "fixed" the AI. I just pushed it to the point where I couldn't successfully defend against an AI invasion with just the city attack and possibly an archer.

 

[Nares]

double post, please delete

 

[Ramstorp]

One of the changes in the Play With Me mod is to increase the flanking bonus because the AI actually does use its units in conjunction with eachother effectively enough to be passable.

Yeah, I suppose that is the situation in my case since I always Play With Me as well.
The AI would have many more units and therefore always higher odds.

How is that not more fun than having to play against an AI with guns when you still have clubs?

 

[Nares]

How is that not more fun than having to play against an AI with guns when you still have clubs?

Increasing the flanking bonus is just an alternative approach to addressing the same problem.

The hypothetical situation of an AI being 1-2 eras ahead is used only to demonstrate how powerful some of the friendly territory bonuses can be when stacked.

I preferred to drastically reduce the friendly territory combat bonuses rather than increase the flanking bonus because the human player can use the flanking bonus more effectively than the AI.

The best solution is probably some combination of reducing friendly territory combat bonuses and increasing flanking.