Probabilistic trade negotiations

[Shadowhal]

Sorry if this has been suggested for. I occasionally lurk the forums and did not encounter it.

So, traditionally it seems to me that the AI assigns a certain value to a given trade good in a given situation and you need to match that to get a trade deal. In theory you can start high and reduce your gold count or resources until it accepts. In fact, in some 4x games (like FE: elemental) it even gave you dynamic information on what will be accepted and what not. It works, but in the end feels rather mechanical tinkering with numbers.

What if instead you get a % estimate of whether your trade partner will accept your offer? Say, you offer them silver for pearls and the interface tells you that the AI will be 90% likely to accept. AND you can only make x number of offers to an AI per turn (maybe determined by envoys?). So no low offers that you just keep repeating until they accept. The probability to accept could be influenced by existing trade deals, mutual relations , relative strength etc. (the usual factors). But, and that is a key point, there is a point of uncertainty in the process. You can boost your chances, but it will cost you more. Trading because a bit more of a conscious decision as to how badly you want something. Can you live with just a 50% chance of getting it? Or do you want to be 95% sure?

It could be an effort to program something like that and make sure it works well between AIs, but I just thought it could be an interesting twist. What do you think?

 

[dexters]

With Civ3, which is the one with the most open trades in terms of everything up to and including cities having a price, I tend to find AI will sometimes offer deals IBT not possible later (this could be due to them spending their gold and gpt elsewhere)

More importantly, when negotiating, I tend to find that unless they can't afford it, the price floor is usually 15-20% below my initial offer. So if you open with a price, you can't find 50% discount on something even if the AI had valued something 50% below your price. So I think the AI is looking at initial offers and then applying a set logic to their negotiations, rather than going with their own valuations.

This has been much harder to suss out in later Civ games since trade actually become much more restricted. And in Civ5, you may be correct that AI valuation is simply based on their own valuation and not the offer price.

I don't think probabilistic trade is a good idea as trade negotiations IRL usually takes years and is tied closely to grand strategies of each nation state and the interconnected established interests within a state with set agendas which is unaffected by changes in government. So I would tend to want the trades to be more rational than chancy.

That said, trade negotiations should be dynamic insofar as prices should be fluid and adapt to negotiating tactics and can change within a negotiation cycle to lock in a price within a range of the initial offer to prevent players from repeatedly inputting different prices to find what the AI will pay and adding back/remove the deal 1 or 2 gold to complete the deal.

 

[m15a]

Personally, I'd rather a simpler system that just tells me what the civ wants to pay - like Civ V but with a lot of the bugs/quirks ironed out. I want to spend as little time as possible trading (although I think visiting the leader screens is worth a few seconds as long as trading doesn't happen every turn).

This idea sounds like, in practice, I'd be attempting trades with high risk every single turn with every single civ to maximize my profits.

 

[dexters]

Personally, I'd rather a simpler system that just tells me what the civ wants to pay - like Civ V but with a lot of the bugs/quirks ironed out. I want to spend as little time as possible trading (although I think visiting the leader screens is worth a few seconds as long as trading doesn't happen every turn).

This idea sounds like, in practice, I'd be attempting trades with high risk every single turn with every single civ to maximize my profits.

Getting a deal from the AI is usually part of the fun

This is one area where Civ has regressed. And it's not a technology or AI issue, the devs simply focused elsewhere.

I think it's worth thinking about it, even if we don't agree on the solution. or some of us think there is no problem.

 

[KrikkitTwo]

What I'd like to see.

A diplomacy Engine that calculates

How much the deal is worth to them (according to game factors independent of AI strategy)
AND
How much the deal is worth to you (according to game factors independent of AI strategy)

They (And You) know this info, and they will only take a deal that splits it 50-50 (so that you both "officially" want the deal the same amount.)

Either one of you can offer more.. in which case the one offering more earns some diplomatic points. (real points with real effect)

 

[ftl]

In the trading, I think the most important thing is that it's not too boring to exploit. The player shouldn't choose between playing well (getting a good deal) or not being bored.

In civ 5, the system was both exploitable and tedious. There were a lot of resources to trade, you could often get the best deals by shopping around selling them one at a time, deals would expire and may or may not be renewed, so if you wanted to get the most GPT out of your resources that involved a *lot* of clicking and making offers.

I think this probabilistic approach would have a similar problem. Suppose I have a resource I want to sell. With this probabilistic trading, the way to go would be to set a price that's too high by a little bit, a trade that has like a 10% chance of being accepted. And I try offering it the maximum number of times to one AI, then the maximum number of times to the second AI, and so on, until I get lucky...

There's of course an interesting optimization problem in terms of what price do you ask for given how much trading you want to do this turn, how many AIs do you have to trade with, what's the probability of accepting... but the mechanics of playing through it would be *really really really* tedious.

I think one of the goals of the trading system should be that if a human player wants to buy or sell a resource and don't really care who they're selling it to/buying from, there should be a pretty simple interface to figure out what the best deal you can get is.

Right now, in civ 5 if I have, say, 5 Iron that I'm not using and want to sell, the process to get the best deal is as follows:
1) Pick the most friendly AI. Offer 1 iron for 2 gpt.
2) If they accept, repeat the offer.
3) When they reject, go down the list to the next most friendliest AI and repeat.
4) If none are willing, try lowering the price. Repeat...

It's so boring that a lot of the time of the time I don't bother and just sell suboptimally.

A better the interface would be something like:
1) Select what I want to sell.
2) Select an asking price.
3) See whether any civs are willing to meet that price.*
*with the trading system well-enough designed so that breaking up a large trade into multiple small trades never helps you, so selling 5 iron all at once is as effective as breaking it up.

 

[dexters]

Trading optimally is a mini-game and can bore people who don't want to bother with it.

Trading is also part of your grand strategy. Give iron to someone who needs it so they can fight a war against someone else you don't like is a valid strategy, but there may be no benefit in gold or gpt from that trade.



Now, if a trading system allows for bundled bonuses (ie: AI values 2 instances of something they really need exponentially more than 1 instance and 3 instance exponentially more than 2) then that would also be an interesting system.

 

[ftl]

Definitely! I agree that it's not always about profits!

I just think that the interesting choices are at a higher level. Once a player decides "I don't care who to sell to, I just want the best price" (they've made that choice) then that choice shouldn't be tedious. It seems like it's a pretty common situation.

Obviously if the player is trading for some reason other than maximum profit (to help or hurt a particular civ, etc.) then that's a different option. I don't mean to say that that should be disallowed.

Ideally, once a player decides what they want to do, it shouldn't be too annoying, interface-wise, for them to follow through on their decision, if any subsequent decisions aren't worth spending brainpower on. I personally think that "just sell at the best price" is a pretty common thing for a player to want to do, though obviously it does not cover all the cases.

 

[m15a]

There are a lot of interesting possibilities for trading that will make it more interesting (and thus more worth spending time on) that don't have anything to do with probabilistic outcomes.

Personally, I just don't want to play a probability trade minigame while playing Civ. But supposing I did...For a probabilistic game to be fun, there should be a reward and a risk. The risk of losing shouldn't be that the player has to try over and over until they win.

 

[qwerty25]

Sorry if this has been suggested for. I occasionally lurk the forums and did not encounter it.

So, traditionally it seems to me that the AI assigns a certain value to a given trade good in a given situation and you need to match that to get a trade deal. In theory you can start high and reduce your gold count or resources until it accepts. In fact, in some 4x games (like FE: elemental) it even gave you dynamic information on what will be accepted and what not. It works, but in the end feels rather mechanical tinkering with numbers.

What if instead you get a % estimate of whether your trade partner will accept your offer? Say, you offer them silver for pearls and the interface tells you that the AI will be 90% likely to accept. AND you can only make x number of offers to an AI per turn (maybe determined by envoys?). So no low offers that you just keep repeating until they accept. The probability to accept could be influenced by existing trade deals, mutual relations , relative strength etc. (the usual factors). But, and that is a key point, there is a point of uncertainty in the process. You can boost your chances, but it will cost you more. Trading because a bit more of a conscious decision as to how badly you want something. Can you live with just a 50% chance of getting it? Or do you want to be 95% sure?

It could be an effort to program something like that and make sure it works well between AIs, but I just thought it could be an interesting twist. What do you think?

I think this idea is better than the current trading system where you just keep incrementally increasing the gold or trades until they accept it.

I'd add a small fixed gold cost to each attempt. That'll prevent a person from trying three times over and over at the 51% chance mark. So each offer you make might cost you lets say for example 5 gold down the drain regardless if the trade succeeds or not. If you keep making offers it'll drain your coffers. (we really got to remove this asking the computer player over and over times the number of comps in the map. @ftl) Though this doesn't fix everything.

Also I wouldn't give such a fine information down to the percentages just give the player a vaguish chance at the 0, 20, 40... 100% marks until later in the game when they have spies and such.

 

[historix69]

AI assigns a certain value to a given trade good in a given situation

In Civ5 your diplomatic rating with the AI strongly influences the trade values, e.g. having a long peaceful era of friendship usually yields the most Gold in selling luxuries. When prices go down the player knows that diplomatic relations are not optimal. So trade is a diplomatic barometer.

In Civ5 You can always ask AI what it takes to make a deal ... allthough there are sometimes better solutions AI will accept.

 

[Ryika]

I prefer having clear information about what deals the AI will accept and which they won't.

But I would love the AI to actually adjust their prices dynamically depending on factors such as scarcity, how much they actually "need" a specific Resource, etc.