Civ4 AI Survivor: Season Four

[Reverent]

I suspect there is a value for an AIs willingness to lose units. A long time ago I tried out an Always War game, and I did so on Settler difficulty to get a feel for it. I noticed that putting a single unit on defensive terrain made enemy stacks take massive detours to get around it, where as the regular difficulty AIs would have just plowed through it. In AI survivor it's pretty common to see raiding cavs or tanks attack enemy cities.

Interesting point. I think you are correct about Ragnar attacking out of his cities and thus losing his defenders (or should have been defenders) at poor odds. I may have to go through the leader XML file and see if there is anything in there that might define this tendency.

There is! Here’s Ragnar's Leeroy Jenkins gene:

• Base Attack Odds Change indicates how likely the leader is to think that his or her units can win a battle against an enemy unit. Higher numbers represent a greater likelihood of attacking nearby enemy units.
  

For comparison, the other insane warmongers, like Montezuma, Shaka, and Genghis Khan, have a mere value of 4 for this attribute. Peaceniks like Roosevelt, Bismarck, and Gandhi have this value set to 0. Meanwhile, Ragnar has a ridiculous value of 6. With such a high number, Ragnar’s AI ends up ridiculously overconfident in his combat skills and makes suicidal assaults.

Incidentally, Napoleon also has a value of 6 here. However, I don't remember him making the same sorts of suicidal charges that Ragnar does in my own games, nor having the same kinds of barbarian troubles. Besides Napoleon and Ragnar, every other leader in the game has a value of 4 or lower.

[haphazard1]

Thanks for finding this info, Reverent!

Maybe this is intended to represent Ragnar's Viking berserker heritage? "Insane overconfidence" would be a good decription of the berserker personality type.

[Fluffball]

Wow that's really interesting. Every time I learn more about the Civ4 AI, it seems that much more subtle and well-honed.

The 'BAOC' number would explain why Napoleon is basically undisputed in war as an AI ally to a human. If an AI were to get ahead in pretty much any way, the willingness to throw away troops in favor of results is an enormous advantage. And since Napoleon is subpar and Ragnar is absolutely terrible on their own, we see them get behind and then throw away troops they can't afford to lose in AI Survivor.

I can't count the number of times Sullla has said "peacenik X lacks the killers instinct", and now we know that means they aren't willing to risk troops in order to definitely capture a city, i.e. Roosevelt struggling against a dead Rome in the playoffs.

If we ever see an AI Survivor game were Napoleon or Ragnar have ungodly starts, that number might come in to play.

[Sullla]

One last reminder: the Season Four Championship game starts about three hours from the time of this post. Good luck to those who entered predictions and I'm sure that I'll see some of you on the Livestream.

[yuris125]

Who won then? Sorry I wasn't around for the stream

[darrelljs]

You missed a display of incompetence like none other.

Charlemagne

Darrell

[Brian Shanahan]

You missed a display of incompetence like none other.

Charlemagne

Darrell

It was like the game was all Montys. Seriously.

[Fluffball]

The beginning was competent. Both financial civs and Kublai got off to good starts. Darius was just screwed from the get-go, but he was doing some impressive teching.

Shame Stalin was such a let-down, since he's the one civ of that bunch that has proven the ability to actually win a game.

Edit: Another thought; it seems like all the AIs had basically the same settling abilities, so in the non-mirrored games where a civ is a total dud, I think it's likely they just can't find the right spots to settle. Given enough land and enough good land, all the AIs did totally fine at expanding. Barbs messed up Stalin, and I think things like desserts, mountains, and lack of space really mess up the AIs in the other games.

Willem is sort of notorious for being terribad at expanding, and yet he did fine here.

[haphazard1]

The near-identical settling patterns on the mirrored map were very interesting. I wonder if differences in which techs the AIs had when they finished their various settlers had an effect, as they valued different resources based on whether they could work them yet?

I agree with Fluffball that the early and middle game was generally very competent, with the exception of Stalin who just stalled out completely. He was slow to expand, and that allowed those barb cities to appear blocking some of his prime spots. But everyone else expanded, got some teching going, and generally did pretty well. The economic powerhouses were doing their thing and racing ahead.

And then...somebody flipped a switch and all the AIs dropped about 50 IQ points.

[haphazard1]

A thought about Rifling: it is a purely military tech, and it does not lead to anything that is not purely military. It is necessary to build some units at later techs, but Rifling itself is not required for techs like Assembly Line. So the economic/culture flavor AIs just will not touch it until they run out of other options to tech.

Some other important military techs also have some other kind of benefit associated -- a wonder or a building -- or are prereqs to techs that do have non-military benefits. So the economic/culture AIs will (sometimes reluctantly) research them. But not Rifling. It also does not help that there are so many other pathways to research before the AI gets bottlenecked and MUST tech Rifling.