What will the AI research next?

[PattonTwo]

Originally posted by Bamspeedy
If you use the Civ3 Multi-Tool (save-game editor), you can find out what each civ is currently researching.

Where do you find the "save game editor"? My civ game only has a scenario editor.

Thanks.

--------------------------------------------------------------

Answered my own question. Here's the link if anyone else needs it.

http://forums.civfanatics.com/showthread.php?threadid=54689

 

[anarres]

Thanks Bamspeedy, that is what I assumed. A crazy piece of AI programming...

 

[WillJ]

Very useful thread.

Hmm, so the AI doesn't value techs leading to other techs? For example, steel and refining leading to combustion, and writing leading to literature, map making, code of laws, and philosophy? Interesting.

And I'm assuming that these numbers don't mean anything besides relating to the others? For example, worker job has a value of 1 because it has the lowest value, and small wonder has a value of 2 becasue it's twice as valuable as worker job (etc.)?

Also, do you know if a civ always researches the tech that has the highest value of all available techs, or do they sometimes decide to research a slightly less valuable one?

 

[alexman]

Yes, the numbers are relative, because the AI doesn't use them for anything but to compare one tech to another.

I initially thought that the AI always chooses the tech with the highest value (with variations between two AI civs being due to difference in the sizes of their economies and any UU), but it seems that there is a random factor involved. Still, the higher the AI value of a tech, the more likely it will be chosen.

For example, since Bronze Working has a value so much greater than all the other first tier techs, the AI almost always researches that tech first.

 

[WillJ]

Okay, thanks for clearing it up.

 

[Ribannah]

Originally posted by WillJ
Also, do you know if a civ always researches the tech that has the highest value of all available techs, or do they sometimes decide to research a slightly less valuable one?

There is a random factor involved in the decision.
What I do not know is whether there is a difference between AI civs as well. I could imagine it to depend on aggresiveness, favoured/shunned government etc.
But there are those times that all the AI civs are researching the exact same tech.

 

[Ambiorix]

Disclaimer : I've only just started to dig into the information that is revealed in the editor, so bare with me if these questions seem trivial.

- the era-advancement flag isn't listed in alexman's post. Is this not a factor for the AI then ?
- The example given of Bronze Working valued at 218 - doesn't this value also depend on the difficulty level (different cost factors) and map size (different tech rates) ?
- Strider mentions the factor of a civ being at war/peace. Did anyone look into that ?
- related to the above : is the shunned/favored government of influence ?
- the 'optional : divide by 1.5' isn't explained. Why divide by 1.5 ? It doesn't change the relative value of the techs, right ?

Thanks for the thread, alexman !

 

[WillJ]

Originally posted by Ambiorix
- the era-advancement flag isn't listed in alexman's post. Is this not a factor for the AI then ?

Originally posted by Ambiorix
- the 'optional : divide by 1.5' isn't explained. Why divide by 1.5 ? It doesn't change the relative value of the techs, right ?

When alexman says "optional," he means the techs that are optional to research to get to the next age. That should explain both of these questions. (And I was confused by the optional part at first too. )

Originally posted by Ambiorix
- The example given of Bronze Working valued at 218 - doesn't this value also depend on the difficulty level (different cost factors) and map size (different tech rates) ?

Maybe I'm mistaken, but I don't think the difficulty level has anything to do with tech values. As for map size, yes, that does affect the value, and alexman said it's 218 on a standard map (and with each civ producing 4 beakers per turn).

Originally posted by Ambiorix
- related to the above : is the shunned/favored government of influence ?

I doubt it, because I'm pretty sure the s/f government doesn't affect which governments are the favorite and least favorite of each civ to choose for themselves, but instead it affects their attitude towards other civs with that government.

 

[Ambiorix]

Thanks for the fast reply, WillJ, that indeed answers most of my questions.

I was thinking difficulty level plays a role in the presumption that research cost at Deity level (for example) is only 60% of what it is on Regent level. (Using the cost rates of 20, 12, 10, 9, 8 and 6 for the different levels). Not sure this presumption is correct, though. Can anyone confirm or reject this ?

If (or when) I understand the topic fully, I might have a go at that spreadsheet.

 

[WillJ]

Originally posted by Ambiorix
I was thinking difficulty level plays a role in the presumption that research cost at Deity level (for example) is only 60% of what it is on Regent level. (Using the cost rates of 20, 12, 10, 9, 8 and 6 for the different levels). Not sure this presumption is correct, though. Can anyone confirm or reject this ?

I looked in the Editor, and you're right about that.

Even though difficulty level seems to affect the cost of techs for the AI, I don't think that really matters, since each tech still has the same cost value in comparison with the others, and the numbers alexman gives are only relative (meaning they don't mean anything unless compared to others). So, for example, if I'm not mistaken, Bronze Working might have a value of 109 in some situations. If it did, then Masonry would have a value of 11. So it all means the same thing, basically. If that makes any sense.

 

[alexman]

Nice to see a discussion going on here!
I have two comments:

1) As of a few patches ago, the research cost of each tech for the AI is the same for all difficulty levels. The AI gets a production bonus (or penalty) which exactly cancels the increase (or decrease) in cost due to difficulty level. As a result, in PTW, research cost for the human is different at each difficulty level. In vanilla Civ3, the research cost is constant for the human and changes for the AI though.

2) Remember, the tech values given in the first post do not depend on the actual AI research cost, but on the number of turns required for completion. That takes into account technology cost, map size, the economy of the AI civilization, and for vanilla Civ3, difficulty level.

 

[Qitai]

Alexman, the research cost changes is effective vanilla civ3 1.29, not PTW.

 

[Ambiorix]

@WillJ : actually, it does make a difference, since some techs use the #turns parameter and others don't.
Meanwhile, I also remembered that the actual tech cost also depends on the number of civ's that already discovered that tech.

If you'd want to create an exact prediction of the tech sequence in a spreadsheet, you'd have to take all that into account. To keep it simple, I'd suggest using a 'base #turns' as input. This would be the number of turns required (for the AI) to research a tech if this tech had a base cost factor of 1.
So if you see during a game that a civ discovers bronze working in 30 turns (std map, regent, ...) you'd have to divide this by the base cost factor (3) to get a 'base #turns' of 10, and use that as input parameter for further calculations.

Uhm, am I still making sense (It's early on Monday here) ?

 

[Qitai]

There is no exact prediction. There is a RNG involve in the choosing of what the AI research. This article only helps to determine the probable tech that AI will research.

 

[Ambiorix]

I know, I meant "exact, except for the RNG".

Concerning the RNG, just an intuitive remark : suppose you can research two techs. One has a value of 200, while the other has a value of 50. Would it make sense to say that the chance of tech one being chosen would be 200/(200+50) ?
I suppose that could only be verified by very extensive tests, as happened for the battle-percentages. Don't see myself doing 6 million tech-research tests though.

 

[Qitai]

Well, I like to see those result myself. Because, although my impression on what the AI will research next fits quite well with Alexman's research, there are times when I think AI seems to favour certain tech which do not seems to be explained above. Of course, my sample size is only as large as the number of games I played which isn't really that large a sample size.

 

[WillJ]

Originally posted by Ambiorix
@WillJ : actually, it does make a difference, since some techs use the #turns parameter and others don't.

It seems to me like all the things alexman listed are either constant, or are #/turns.

Originally posted by Ambiorix
One has a value of 200, while the other has a value of 50. Would it make sense to say that the chance of tech one being chosen would be 200/(200+50) ?

Hmm, interesting idea...

 

[Ambiorix]

@WillJ : what I mean is that for instance the governement factor (259/turns) will fluctuate, while for instance the Defender factor (198) will remain constant. Depending on the # of turns, government and defender factors may swap places on the wishlist of the AI.

Still think I'll use this concept of 'base number of turns'.

 

[WillJ]

Originally posted by Ambiorix
@WillJ : what I mean is that for instance the governement factor (259/turns) will fluctuate, while for instance the Defender factor (198) will remain constant. Depending on the # of turns, government and defender factors may swap places on the wishlist of the AI.

Still think I'll use this concept of 'base number of turns'.

Oh yeah, you're right about that.

 

[denyd]

This is a very tough worksheet to build. I tried to do this a couple of months ago and gave up.

I decided that when I had the time, I'd try again using a VB program. That way I could use lookup tables for the UU, Civ and Game Level portions of the equation.

I wish you good luck.

denyd