AI - human trade revealed

theNiceOne · January 15, 2003, 04:50

Hi.

I have now started a new thread where I rewrote the post. Please check: AI - human trade revisited .

bongo · January 15, 2003, 05:33

Impressive.

Hurricane · January 15, 2003, 08:09

Very nice. If you are correct it would mean that Regent is NOT a perfectly equal match between the human and the AI. May I ask you how you did the test?

theNiceOne · January 15, 2003, 08:35

Hurricane: I am right

I did the test by creating a scenario consisting of two islands each with one city. One belonging to Egypt (AI), the other to Rome (human). I placed one warrior of the opposite side close to the cities, so that the civs had contact from the start.

Both civs were set up to know all techs up to currency, but only one side knew currency, monarchy and republic.

I then started the scenario on chieftan and wrote down exactly what the AI needed for each trade, then restarted the scenario on warlord, etc. I did all this twice, once with the AI knowing the three techs, and once with the human knowing them.

I then tried this with changed AI-AI trade bonus settings and found that it made no difference. I also tried it with different AI cost factor settings, and found that it made the difference as written above.

I also extended the scenario by roading the human island and adding two wine (and a harbour to the city), while extending the egyptian island by increasing the number of cities to 30 (one with harbour) and roading the island.

I now found that the AI would give 300 gold for one wine on chieftan, but only 298 on warlord (300 / 1.008 = 298) and only 261 on deity (300 / 1.15 = 261).

After this it was a bit of thinking to correctly factor in the difficulty cost factor (easy) and the monarch wonder cost (a bit harder).

And you're right that regent is not a perfectly match, partly because of this, and partly because of the AI-AI trade bonus of 130. But then again, an average human player will more than outweigh those handicaps through trading alone by only trading techs during its own turn.

Hurricane · January 15, 2003, 09:03

Just one more check: did you write down what they offered (i.e. the "what would you give for this tech?") or adjust the deal to the most favourable deal for the human?

Can you try a similar scenario with 2 AI players and see how much their offers differ? I am under the impression that there is some factor for that too (i.e. egyptians would pay 300 for the wine, but the Germans only 290).

theNiceOne · January 15, 2003, 09:14

Hurricane, I adjusted it to the most favourable deal for the human each time, both when buying and selling.

Yes, I will get some more numbers with more than one AI. I did one additional test, btw. I first sold Egypt the wine, which made them gracious, but they demanded exactly the same gold for the tech afterwards.

Nor Me · January 15, 2003, 09:55

Interesting. I suspected that something like this was the case but have never tested it. I've noticed that this also depends on things like aggression. So, for example, I've been able to get an even deal with the French on regent from a scenario.

I think the tech price formula is a bit more complicated than that (In fact I've never seen it; if anyone knows it then post.)
I don't see that there's any reason why I can't plug my thread:Partial and pointless formulae: resource cost

Quote:

Originally posted by Hurricane
Very nice. If you are correct it would mean that Regent is NOT a perfectly equal match between the human and the AI.

Hurricane, you are assuming that the AI will give other AIs an even deal most of the time. I suspect that this is not the case.

I've heard that the AI-AI trade bonus only applies when one AI cannot pay the full cost. After all, it can only affect one side in a deal.

Datajack Franit · January 15, 2003, 10:00

Awesome

faded glory · January 15, 2003, 11:49

cool

Catt · January 15, 2003, 14:14

As always, very nice work theNiceOne

Catt

Hurricane · January 15, 2003, 14:43

Quote:

Originally posted by Nor Me
Hurricane, you are assuming that the AI will give other AIs an even deal most of the time. I suspect that this is not the case.

I've heard that the AI-AI trade bonus only applies when one AI cannot pay the full cost. After all, it can only affect one side in a deal.

Of course I know about the AI:s doing deals for discounted prices to other AI:s. But since the human player can do that as well I don't see this as a handicap for the human. What surprised me, however, was that Firaxis always has said that the AI should play by exactly the same rules as the human on Regent. As TNO has shown, this is not the case.

theNiceOne · January 15, 2003, 15:45

Quote:

Originally posted by Hurricane
Of course I know about the AI:s doing deals for discounted prices to other AI:s. But since the human player can do that as well I don't see this as a handicap for the human. What surprised me, however, was that Firaxis always has said that the AI should play by exactly the same rules as the human on Regent. As TNO has shown, this is not the case.

The AI-AI trade bonus is a handicap for the human if the human lacks enough money for buying a tech. The civ having the tech would sell it to another AI for the gold you can pay, but not to you.
However, IMHO it's a bigger handicap for the AI that it can never sell a tech for less than the AI-AI trade bonus adjusted value. If a few civs fall behind and cannot pay even after this bonus, then the human player gets to sell all techs to those civs for whatever they can pay.

When it comes to the AI-human bonus I found, this can be viewed as a cheat if you want to, but I think this AI bonus is still worth very much less than the trade bonus the human player gets by only trading during its own turn.

RonO · January 15, 2003, 23:23

Did the government of each civ matter? Logically it would make sense for a civ to ask more from a player under a form of government it 'hates' or less from a player with a form of goverment it 'likes'.

Nor Me · January 16, 2003, 09:29

Quote:

Originally posted by RonO
Did the government of each civ matter?

I've done essentially the same test as theNiceOne and found that this and aggression were the only obvious things that made the price different between civs.

In fact, I recently did a test on some randomly selected saves where I pillaged my resources found the lowest price each civ would charge from them.
From that I think that these two factors are more important than attitude or any other bilateral factor (so it's not worth being nice to civs in the hope of a better deal, at least for resorces.)

Quote:

Originally posted by Hurricane
Of course I know about the AI:s doing deals for discounted prices to other AI:s. But since the human player can do that as well I don't see this as a handicap for the human. What surprised me, however, was that Firaxis always has said that the AI should play by exactly the same rules as the human on Regent. As TNO has shown, this is not the case.

I was suggesting that the 103.9% might affect AI-AI deals. Only if other factors make the deal favourable would the AIs trade with each other at all.
This would be an advantage to the human who knows that a luxury-luxury deal at a 4% disadvantage is always a good idea.

Then the only advantage the AI would have over the human on Regent is in being given discounts by other AIs which the human would have to pay the full price for.

Nor Me · January 16, 2003, 10:42

I've just witnessed a mere 110.0% price increase between cheiftan and deity. Maybe this is not quite right. In case anyone wants to know, I was French and they were Iroquois.

Nor Me · January 16, 2003, 11:12

I did another test. I created two scenario which differed only by a modded civ. In one, the Iroqouis had the least agression level, were European and favoured Republic. In the other, they had the highest agression level were American and shunned Republic. I was a French republic.

When everything was in my favour I got AI-human bonuses of:
Chieftain 100.0%
Regent: 100.0%
Deity: 105.0%

When evreything was against me:
Chieftain: 100.0%
Regent: 109.0%
Deity: 122.0%

Yes, they are all exact percentages. I haven't tested these factors individually yet.

vmxa1 · January 16, 2003, 14:28

The question I have is that the developers have said all along that the game does not know the AI from the Human. If that is so how can it treat one party differently? Could not these discrepancies be accounted for by other factors.

wilbill · January 16, 2003, 15:08

theNiceOne, Thanks for this very helpful information.

Catt · January 16, 2003, 15:08

Quote:

Originally posted by Nor Me
[. . .]

When everything was in my favour I got AI-human bonuses of:
Chieftain 100.0%
Regent: 100.0%
Deity: 105.0%

When evreything was against me:
Chieftain: 100.0%
Regent: 109.0%
Deity: 122.0%

[. . .]

Curiouser and curiouser. Good work NorMe and theNiceOne!

Catt

Theseus · January 16, 2003, 23:44

Well done, indeed.

I always wondered why tech leads seemed SO much more important / rewarding on higher levels.

It bugs me a little that this was never reported by Firaxis, especially at Regent, but I also think it's very, very elegant. I've never liked the *outnumbering* approach to the AI higher levels, but like the stuff that just makes the game harder. Kudos, Firaxis.

theNiceOne · January 17, 2003, 03:27

RonO: I didn't check this. Both the AI and the human were in despotism when the tests were done.

I didn't really try to check everything that affects tech trade cost, only how difficulty levels affect it.

So there are more testing that can be done here.

The base trade cost (ignoring number of civs that have the tech and number of civs total) of a tech is probably (my guess) computed something like this:

[(Research cost * A + B) / CF + C] * AH + D

CF = Cost factor: 1 on regent an below, 0.9 on monarch, 0.8 on emperor and 0.6 on deity.
AH = AI-human bonus as described in the first post.

Where factors that don't affect reserach cost, but affect trade cost (like possible wonders, new resources, attitude etc)are either of type A, B, C or D:

Type A: Factors that are multiplied to the research cost. An annoyed seller could have a factor of type A = 1.1

Type B: Factors that are added to the research cost, and affected by cost factor.

If government availability is a cost factor, then it must be of type A or B, since the cost of republic was directly propotional to CF and AH when no other factors applied.

Type C: Factors that are not affcted by cost factor, but are added before the AH multiplier is applied and thus affected by this. As described in my first post, wonder availability is of this type.

Type D: Factors that are added after all multipliers, and therefore constant regardless of research cost and difficulty level. I would guess that there aren't any such factors, but I don't know.

This got a bit technical, but my point is that there are still a lot of researching to do before we can predict what an AI civ will pay for a tech. I didn't mean to find out everything, only that it is a (until-now) hidden difficulty-dependant modifier that affects the price.

theNiceOne · January 17, 2003, 03:39

Quote:

Originally posted by Nor Me
When everything was in my favour I got AI-human bonuses of:
Chieftain 100.0%
Regent: 100.0%
Deity: 105.0%

When evreything was against me:
Chieftain: 100.0%
Regent: 109.0%
Deity: 122.0%

Yes, they are all exact percentages. I haven't tested these factors individually yet.

Good job, Nor Me. I've done some testing as well, and found that values may differ from my initial numbers. It looks like this modifier may itself be modified

I will look more into it when I get the time.