What will the AI research next?

[Ambiorix]

Actually, the spreadsheet is coming along quite nicely. Probably be able to finish it this week, although I'll probably give priority to my QSC-submission.

The only real obstacle left is that I'm not really familiar with all the implications of UU's. Things like 'Greeks don't build pikemen since they have the Hoplite'.
The UU is the only thing that differenciates civ's research-preferences from one another (right ?), so it would be really, really helpfull if someone could list all consequences of the UU as in the example above.

Edit : never mind, I think I can see this in the editor... Great !

I'll post the spreadsheet here when it's ready - it'll take some tough debugging though, I guess.

 

[alexman]

Rome: Iron Working gives resource defender
Greece: Feudalism does not give resource defender (pikeman)
Babylon: Warrior Code gives non-resource defender
Persia: Feudalism does not give resource attacker (med. infantry)
Zulu: Bronze Working gives non-resource atatcker
India: Chivalry does not give resource attacker (Knight), Chivalry gives non-resource attacker
France: Gunpowder gives resource attacker
Spain: Navigation gives resource attacker, Navigation does not give unit (explorer)
Carthage: Bronze gives non-resource attacker, Feudalism does not give resource defender (pikeman)
Mongols: No change
Egypt: No change
China: No change
Japan: No change
America: No change
Iroquois: No change
Aztecs: No change
Germany: No change
England: No change
Russia: No change
Ottomans: No change
Arabs: No change
Korea: No change
Celts: No change
Vikings: No change

All this is from memory, so did I miss anything?

 

[Doshin]

Good work alex A very interesting read

One question though. Does a tech's 'value' decrease if its primary purpose becomes irrelevant. For example, if Shakespeare's Theatre has already been built is the 'wonder' value removed from Free Artistry?

 

[alexman]

No, as far as I could tell an obsolete Wonder does not decrease the value of the technology. However, it definitely decreases the AI's desire to trade for that tech.

The AI places a value on each tech for trading purposes, which (according to Soren) is actually different than the values listed above, but this is a whole new can of worms! )

 

[Ambiorix]

Alexman,
thanks for the list, I'll check it with what I got from the editor.

I'm having some problems classifying some of the units : a lot of units have their own flag in the editor (naval power, cruise missile, ..) and don't have the 'offensive' or 'defensive' attribute. Do all of these count as 'no attacker, no defender, no naval transport' ? I find it hard to believe that an ICBM is only valued as 2. Another example : it would seem logical that a helicopter (air transport) is valued the same as naval transport.
Just asking.

 

[alexman]

I know it doesn't seem logical, but that's what I found in my tests. It's possible I missed something, but I don't think so.

 

[MrWhereItsAt]

Originally posted by alexman
No, as far as I could tell an obsolete Wonder does not decrease the value of the technology. However, it definitely decreases the AI's desire to trade for that tech.

The AI places a value on each tech for trading purposes, which (according to Soren) is actually different than the values listed above, but this is a whole new can of worms! )

Curse that Soren and his mystical repository of information I can never find!

I was just about to ask whether the research priorities discovered here amounted to trade priorities as well.... This would help not only to choose a tech the AI was unlikely to be researching, but a tech that the AI would pay the most for!

 

[Ambiorix]

Decided to wait a day to unleash my Greek Hoplites, and finish this project first.


The resulting spreadsheet can be found below. Here's how to use it (it's quite a lengthy explanation, but the usage is quite straightforward - I hope) :

First of all, I decided NOT to work with an exact calculation of the number of turns needed for researching a tech, since that wouldn't be practical (map size, contacts, difficulty level, ...). Instead I'm assuming that you have a vague idea of how long it took the AI to research a previous tech. Let's say you think it took the Americans about 20 turns to researh philosophy, for example.

That's the start, now let's look at the spreadsheet :
column A : the tech
column B : the base cost of the tech
columns C through AJ : the factors that influence research-priority. Row 2 contains the weight of the factor.
columns AK through CH : the different civs, starting with a fictuous 'neutral civ'. This is a civ that has no starting techs and no unique unit. I just added that one if you want to see the influence of these factors on an existing civ's preferences.

Per civilization there are two columns. For instance for Americans :
column AM, row 2 : this is the 'base number of turns' - you need to fill this in, here's how : remember the example above ? Americans got philosophy in 20 turns. in column B you see the base cost factor of philosophy is 6. This means that the base number of turns is 20/6 = 3.33. Fill this in in AM2.
column AN, row 3 and further : if you want to calculate the probability of this tech, fill in 1, otherwise fill in 0.
column AM, row 3 and further : if column AN contains '1', the probability factor is displayed here.

So basically it's very simple : to stick with the example above, let's say that you want to see whether the Americans will be going for mapmaking or polytheism : open the spreadsheet, calculate the '3.33' as explained above (or just make a wild guess if you wish, this factor doesn't actually change the calculations that much), fill in the '1' for the mapmaking and polytheism rows (AN20 and AN21), et voilà : the magic numbers appear (50.41 for mapmaking and 6.41 for polytheism).

Now I just hope I didn't make any mistakes. If anybody wants to check the formula's, go ahead. I can provide more info on those, if they're not very clear (they're huge, but not too complicated).

All feedback welcome !

 

[drewshark]

@Amorbix
Thanks so much for all the work you have put into this spreadsheet. I just was playing around with it and it should be a big help. The only thing that confuses me is the base turns. Is it simply a guess as to how many turns it will take the AI civ to complete? Is the best way to guess this to click on the same tech advance and see how many turns it will take me? How do I know at what rate he will be spending on science? I realize it will not change the results too much but I was just curious if there were other ways to find out his science rate?

 

[Ambiorix]

Originally posted by drewshark
...only thing that confuses me is the base turns. Is it simply a guess as to how many turns it will take the AI civ to complete? Is the best way to guess this to click on the same tech advance and see how many turns it will take me? How do I know at what rate he will be spending on science? ...

Glad you like it.
Guessing the pace of the AI is the weak link. Perhaps early on in the game it's not really possible, since the AI's will be trading amongst each other, and even if you monitor them very closely, you can't really be certain whether they've just discovered a tech or traded for it. Well, maybe sometimes you can.
But later on when things have polarised a bit more, you may be in the race with one or two other civs, and then it's probably more doable to monitor how many turns there are between two tech discoveries of one civ.
Remember that the 'base number of turns' is the number of turns it would take to research a tech with base cost factor 1 (column B in the spreadsheet). so if you've monitored a civ, you still need to divide the number of turns through the base cost factor.


I'm still thinking about that RNG. Would it be possible to set up a scenario where the AI has, say, three research possibilities of about equal value, and where reloading the game would guarantee a different RNG, and so perhaps a different research choice ? With a demo-version of WinRunner or similar tool, it should be doable to automate a load& save sequence of that game, and check the research path with some other tool (thinking of Spiderzord's PTL).

 

[denyd]

Ambiorix,

Great table,



I downloaded the raw numbers to a database and used the tech table to come up with a projected research order for each Civ based upon required techs and the tech value. I'm posting Spain for people to review. If it looks good to the number crunchers out there, I'll complete the rankings and post it for usage by everyone who wants it.

Since beaker production (turns to research) and the RNG aren't constant they are excluded from the calculation. It would seem that the beaker production wouldn't impact the ranking and the RNG is impossible to estimate.



Spain
Ceremonial Burial 0.00
Alphabet 0.00
Bronze working 212.53
Warrior Code 142.53
Writing 29.20
The Wheel 28.40
Iron Working 26.27
Masonry 17.40
Pottery 14.80
Map Making 46.13
Mysticism 12.40
Horseback Riding 11.12
Literature 10.56
Mathematics 5.20
Construction 12.28
Code of Laws 4.56
Philosophy 4.27
Currency 3.60
Polytheism 2.13
Monarchy 5.43
Republic 1.23

 

[alexman]

Originally posted by denyd
Since beaker production (turns to research) and the RNG aren't constant they are excluded from the calculation. It would seem that the beaker production wouldn't impact the ranking and the RNG is impossible to estimate.

Actually, turns to research makes a relatively big difference. As an example, take the two extremes: 40 turns for everything, and 4 turns for everything. Here are the rankings and values of ancient-age techs for a civ without a UU in the ancient age:

4-turn rate

Bronze working 268
Warrior Code 198
Map Making 108
Monarchy 90
Writing 90
Republic 86
The Wheel 86
Iron Working 86
Masonry 75
Construction 75
Horseback Riding 70
Mysticism 70
Code of Laws 66
Currency 66
Mathematics 66
Ceremonial Burial 66
Pottery 66
Alphabet 64
Philosophy 64
Polytheism 64
Literature 48

40-turn rate

Bronze working 210
Warrior Code 140
Map Making 50
Writing 32
The Wheel 28
Iron Working 28
Masonry 17
Construction 17
Monarchy 13
Horseback Riding 12
Mysticism 12
Literature 10
Republic 9
Code of Laws 8
Currency 8
Mathematics 8
Ceremonial Burial 8
Pottery 8
Alphabet 6
Philosophy 6
Polytheism 6

As you can see, the rankings are considerably different in these two extreme cases, especially for optional and government techs.

 

[EdwardTking]

I have the Oracle and the Hanging Gardens so I held off researching Music Theory and Steam Power that would
obsolete these wonders.

Does the AI similarly consider the obsoleting factor
of technology?

 

[alexman]

No, the AI with the Great Library gladly researches Education.

 

[Qitai]

Throw in the variable that you research different tech at different rate and that complictes things!!!!

 

[DaviddesJ]

Does it matter whether the AI can actually build the unit? E.g., if the AI has Iron Working, but no source of Iron, will it still place a high value on Feudalism for Pikemen?

 

[alexman]

No, it does not matter.
Will you come up with an RCP-like strat using this article too now?

 

[Nor Me]

Originally posted by alexman
No, it does not matter.
Will you come up with an RCP-like strat using this article too now?

I've been wondering how to make the best use of unbuildable UUs for weeks now.

I take it you, of all people, would interested in only religious AIs beelining to Monarchy.

 

[alexman]

That's a great idea, Nor Me. A "priest" unit, available only to Religious civs with Polytheism, for example?

It would be flagged as a defender, would require uranium, and would upgrade to the Guerilla, to ensure that the unit is never actually buildable.

Religious civs with Mysticism would almost certainly go for Polytheism then. Nice thinking!

 

[SirPleb]

Originally posted by Ambiorix
The resulting spreadsheet can be found below.

Ambiorix, I used your spreadsheet while playing GOTM21 and it is a great help, thank you!!!