AI decision making

[unmerged(9146)]

It may have been asked before and if the answer is somewhere in these forums I would appreciate any information where to find it but when the event comes up, how AI decides which option to choose? I would imagine it optimises some function of how the game evolved but then - I'm just playing a strange game where some weird things happen and one of them is that Russia never formed. I was surprised to see this so I wanted to check why (I thought maybe the conditions of the event were never met) and it turns out Moscow decided to turn down the option of becoming Russia. How does AI come up with a decision like that? It is a clearly dominated choice - you're getting to choose between being Russia (more VP, higher stability and more people in Moscow) or staying Muscovy (fewer VP and lower stability). I would think it's one of the dummy events where AI would always choose the former option. Clearly, it doesn't. How come?

 

[Syt]

I think there's an 80% chance (4-1) that the AI will choose the historical route in multiple chioce events. The other 20% you'll get alternate history, like Russia not forming, Castille not renamed to Spain, or the Ottoman Capital not moving to Constantinople.

 

[HisMajestyBOB]

Actually, the chance varies based on howmany options there are. Roughly, if there are 2 options: 85% and 15%. 3 options: 85% and 10% and 5% IIRC.

 

[nelly644]

It is always 5% for each a-historical option and the rest for the historical one

 

[unmerged(9146)]

I think there's an 80% chance (4-1) that the AI will choose the historical route

So AI just picks it at random? No consideration whatsoever for how it might impact the future development of their countries or what the consequences might be?

 

[HisMajestyBOB]

Yup, it just rolls a die.

 

[unmerged(5664)]

Originally posted by nelly644
It is always 5% for each a-historical option and the rest for the historical one

That's right. For two options, 95-5. Three options, 90-5-5. An four options, 85-5-5-5.

 

[Syt]

Originally posted by robin74


So AI just picks it at random? No consideration whatsoever for how it might impact the future development of their countries or what the consequences might be?

Nope, none whatsoever. Besides, the human player is nudged to go historically in most cases, too, as the historical option usually yields far greater returns than the a-historical ones.

 

[Castellon]

Originally posted by BarristerBoy


That's right. For two options, 95-5. Three options, 90-5-5. An four options, 85-5-5-5.

Not exact figures but close enough.

 

[unmerged(9146)]

Originally posted by Sytass
Besides, the human player is nudged to go historically in most cases, too, as the historical option usually yields far greater returns than the a-historical ones.

That's really why it surprises me that AI is just randomizing. I will go historical in most cases but I will certainly go historical if the counterfactual choice is worse in every respect than the historical one. I would expect AI to do roughly the same and randomizing does not reflect this process. I would expect AI to choose the option with the greater returns. I would suspect these choices to be historical in most cases, though ocassionally, since the rest of the game doesn't have to evolve historically accurately, AI would make ahistorical choice, not by a random chance but because it's beneficial under these specific circumstances. But there are some decisions where one option gives you only gains while the other gives you only losses and so one is clearly better than the other under any conditions. Like choosing whether to be Russia or not. I don't really see the point in making AI flip a coin in these situations.

 

[Garbon]

Originally posted by robin74
I don't really see the point in making AI flip a coin in these situations.

Well, then it would have to be programmed to have judging factor. I'm pretty sure it doesn't have that.

garbon

 

[unmerged(9458)]

Originally posted by robin74
That's really why it surprises me that AI is just randomizing.
<SNIP>
But there are some decisions where one option gives you only gains while the other gives you only losses and so one is clearly better than the other under any conditions. Like choosing whether to be Russia or not. I don't really see the point in making AI flip a coin in these situations.

In many cases the historical choice is not the best option, but nations did on many occations chose less than optimal solutions to their problems. In game terms it is usually quite easy to see what the best opion is, but that is mostly not the case in real life. Thus the random factor to decisions makes the game more realistic, even if it reduses the chance that the AI nations develop in a historical manner.

The in-game reason why the AI "flips a coin" is variety. If the AI always choses the the same options it makes the game much more predictable and less interesting.

 

[Winkelried]

Originally posted by Sytass


Nope, none whatsoever. Besides, the human player is nudged to go historically in most cases, too, as the historical option usually yields far greater returns than the a-historical ones.

That's exactly my problem with that system. Aside from making the decision a no-brainer, it implies that the world how it developed historically is the best possible outcome. Voltaire's Candide, ou l'optimisme comes to mind.

 

[Lucidor]

I read somewhere that there is an 85% chance of choosing A, and if A not is chosen, then there is an 85% chance of choosing B, and if that choice isn't chosen, then there is an 85% for C, etc... Ofcourse, if there are only three choices, then a roll against B will choose C.

 

[Director]

Originally posted by Winkelried


That's exactly my problem with that system. Aside from making the decision a no-brainer, it implies that the world how it developed historically is the best possible outcome. Voltaire's Candide, ou l'optimisme comes to mind.

No, it means that those rulers had what they considered good reasons for the actions they took. Unfortunately, EU2 doesn't model all of the factors that influenced their decisions - and we, with perfect hindsight, KNOW the consequences they did not.

 

[unmerged(9146)]

Originally posted by Nocuous

Thus the random factor to decisions makes the game more realistic, even if it reduses the chance that the AI nations develop in a historical manner.

I don't have a problem with a random element when choosing what's best added in order to simulate uncertainty of real life. But the random element cannot be too big. I just don't think allowing AI to choose _any_ course of action, no matter how dumb and inferior, is making the game more realistic.

The in-game reason why the AI "flips a coin" is variety. If the AI always choses the the same options it makes the game much more predictable and less interesting.

I don't propose that AI always chooses the same option. I propose that it chooses the best decision _given_the_circumstances_, and these will vary. How the game evolves is a function of much more than what choices countries made in the events. Sometimes Austria will conquer half Europe and sometimes it will be just one of the tiny German states somewhere in the Alps. I don't really understand why Spain couldn't take that into account when deciding whether to side with France or Austria in the Succession Wars. Add a random component to the objective function if you want. I don't believe that would make the game less interesting. Au contraire. If indeed AI simply sticks to the historical choice 95% of the time, I believe that making it actually optimise would make the game less predictable.

 

[unmerged(1047)]

what Lucidor said.... I've seen it in a lot of places that the AI has an 85% chance of taking each choice, alphabetically, and will take the last if it gets to it. Haven't really tested it myself, or even worked out how one would test this.

 

[DPS]

The AI doesn't even know how to handle naval attrition so it has to be given a cheat, and you think it can handle figuring out which event option is best? No way.

Besides, while some of the events have options that are clearly best, many events are not so clear-cut even to human players.

 

[Castellon]

Originally posted by Sheridan
what Lucidor said.... I've seen it in a lot of places that the AI has an 85% chance of taking each choice, alphabetically, and will take the last if it gets to it. Haven't really tested it myself, or even worked out how one would test this.

Here is the answer from a past post of mine I dug up.

Orignally Posted by Nocuous

Event choices :
2 : n = 90
3 : n = 95
4 : n = 100
The historical choice (A) has 85/n and all other choices has 5/n.

Castellon's expert analysis yeilds;

Therefore if:
2 Choices A= 85/90 or 94.4% B=5.6%
3 Choices A= 85/95 or 89.5% B=5.75% C=5.75%
4 Choices A= 85/100 or 85% B=5% C=5% D=5%

I hope this clears things up.