C3C flavors can make the AI smarter!

[ZargonX]

That's some pretty good news, alexman. I think the critical component would be making sure to mix-and-match the flavors, hopefully allowing 2 AIs to "synergize" (sorry ) with each other, creating more powerful civs than if they were left to their own natural devices.

[donZappo]

Originally posted by alexman
donZappo, I thought that a 100% relationship between flavors just means that the AI is as likely to research a flavor tech as an unflavored tech. (so the tech gets 100% of its original value).

Am I wrong? If 100% means that the AI researches the tech 100% of the time, what happens if there are two techs at 100%? From my small debug test (giving everything a 50% value), I saw that the AI researched techs in the order it would if there were no flavors at all.

I think you misinterpreted what I wrote, Alexman. If it is set at 100% flavor relationship then the AI will pretend as if no flavor existed at all. What I was trying to say was that it would be a problem if you adjusted the flavors for a civ so that it would research a given tech 100% of the time, or bee-line towards a certain tech 100% of the time. You can give a civ a 20% increase in likelihood of researching a given tech and that will make it research that tech 95% of the time. Sometimes all the civ needs is a small nudge in a certain direction to make it very predictable.

[donZappo]

Alright, here's the results of my test with the flavors. I've only included the tests that I felt were relevant as I tried a couple of things before I came upon my standardized testing conditions.

The settings for this test were:
Debug mode.
Emperor Difficulty.
Archipelago map with 4 AI civs set to be isolated from each other and away from goody huts.
No settler building allowed (to speed up research).
Each civ was set to have no inherent civ traits and started with Masonry, Warrior Code, and Pottery.
The human settler/worker were set to be fortified from the start.

I ran each game 10 times until 3100 - 3000 BC and checked to see which tech each AI had researched. The resultes are included at the end of this message. I looked for a number of different possibilites for how flavors interact: I wanted to see if the flavor values were multiplied, if it took the highest or lowest values, if it ignored values that were set to 100 or 0 if there were other values present, etc. The most striking part of the results were what happened when you averaged the results together, though, and I think the chart really helps to elaborate that point. I didn't do any advanced statistical tests, so I can only assume that the values of the test are statistically "the same." If it's not as clear as I think it is, or if anybody has any other questions feel free to ask and I'll try to answer them as best I can.

[donZappo]

Chart:

[alexman]

donZappo, thanks for sharing your results.

You didn't say it anywhere, but I guess you gave all four civs both flavors 3 and 4?

Perhaps your results are right, but I have some reservations to accept them because it looks like you didn't take into account the probability for the AI to research Bronze Working without any flavors.

Each tech has a value in the eyes of the AI, and I suspect that flavors reduce that value, and not directly the percentage of the tech being researched.

For example, if Bronze Working has an AI value of 160, the Wheel has an AI value of 20, and Ceremonial Burial and Alphabet both have a value of 10, then the AI would research Bronze Working 80% of the time. With a 50% flavor, you seem to suggest that the AI would research Bronze Working 40% of the time, but I think the flavor would reduce the value of Bronze Working to 80, which would mean that the AI would research it 67% of the time.

I think the best way to check the effect of flavors would be to create some identical techs that have an equal probability to be researched, and then play with the flavors.

[donZappo]

Silly me, I forgot to include the baseline in that spreadsheet! It's been updated so that you can see how the AI reacts without any flavors being assigned. When a flavor is listed as either the civ doesn't have that flavor, or there is no corresponding relationship to the flavor that Bronze Working has.

And you are right about the values being affected. It seems like it could work either way, but it makes more sense for the values to be affected as opposed to the actual percentages. Are the AI values known values, or has somebody happened to figure them out? If not, then it seems like no matter which method is used to actually determine research percentage trial and error will have to be used to get the AI to behave exactly like desired.

I thought about creating identical techs but I was hoping that this initial survey of flavors would help to guide any future experimentation. At the very least, I discovered how big of an impact flavors can have on how the AI reacts, how touchy the AI decision model is to a relatively small change in flavor values, and that it appears that flavor values are in fact averaged when multipile flavors are taken into account.

[alexman]

Yes, your tests were no doubt a very valuable starting point.

As for AI tech values, yes, they are known, provided you know the number of turns it takes the AI to research each tech.

[donZappo]

Thanks for the link Alexman! That is a very useful set of values to have for figuring out exaclty how the flavor mechanic works. That must have taken quite a while to compile, so good job!

[donZappo]

Man, after playing with flavors a bit more I wish they had made it so that flavors would make a civ *more* likely to research or build something rather than just less likely. It makes it an awful lot harder to convince a civ that they should build, oh, a harbor or something since you have to flag every other building as being "avoided" as opposed to just flagging the harbor to have the flavor. This is going to make it a lot more complicated to make civs do what we want them to.