Discussion: Paying upfront / monthly payment

[Blurb]

A relevant point is that upfront costs are *far* easier for the AI to handle. And who doesn't want a competent AI?

Is it really easier for AI to handle? I heard the same point touted in favour of removing the tile system, which was allegedly very hard for the AI.
Much as I loved to see tiles disappear, AI performance in 2.2 speaks for itself regarding this.

 

[PK_AZ]

You can make the system work for the player. The problem is the AI. It has been an issue in every Pdox game that uses this approach. People always complain the AI builds too much of the wrong stuff, filling its queue with useless things that it won't prioritize better stuff over.

Can I hear more about which games use similar system, and what exactly kind of problem happens in them?
Also, wouldn't that problem be solved if AI use very short queues? I think literally something like:
OnFactoryBuiltOrProjectFinished()
{
if(FreeFactories > 0)
BeginNewProject();
}

 

[OrigamiPhoenix]

I also want monthly edicts and payments. It will heavily reduce the micro.

The problem is: what happens if you at some point can't afford the monthly cost anymore (e.g. because of a destructive war). Does the project then fail, even though it is almost completed?

Why not simply have it stop and resume once you can afford again?

This. It should automatically postpone the project and give an alert that it has done so. Postponing the project yourself would also make life easier so that you can stockpile if you think keeping up construction isn't worth it right now.

 

[BigPoppa1111]

Can I hear more about which games use similar system, and what exactly kind of problem happens in them?
Also, wouldn't that problem be solved if AI use very short queues? I think literally something like:
OnFactoryBuiltOrProjectFinished()
{
if(FreeFactories > 0)
BeginNewProject();
}

Head over to the HOI forums for almost any iteration of the game. I believe every iteration of the game used(uses) a production queue system. It as common to be steamrolling the enemy, wondering what in the world they were building, only to open up as that country and realize they were building garrison units in Timbuktu.

The problem with your approach is it assumes factories are more or less static. For instance, AI fills available factories. Player conquers one of AIs planets. AI factory count falls below what is required. AI has a hole to dig themselves out of. Fine. AI stops queuing more production until it builds down to that threshold. In the meantime, AI researches IWIN button. AI should queue it now and make it top priority, but in your system, it won't produce it for 10 years until it clears its queue.

So what you end up with, is having to code in some logic allowing the AI to overbuild with certain priorities, which opens up to the typical queued building priority problems. The root of this problem is prediction. You ideally want the AI to produce an item as close as possible to the point at which it first needs it. In war, this might be war material. So you code it that way. But then you have an AI that has a +100% production output building at 99% completion for 10 years while it produces war materials. The AI isn't good at predictive behavior. Which is why a save then pay system is easier for the AI to handle. It's not fool proof, but it allows it to better react to game situations than pay as you go does.

Just walk through the different game play scenarios and ask yourself whether either scenario offers a benefit to the AI. You will find that save first is overall better until the AI can better predict the future.

 

[The Founder]

Something worth noting in Stellaris is that for any project (buildings, districts, ships, megastructures, terraformation, etc...) the payment is entirely done upfront.
This has the advantage of making ressourses management easier because you won't have huge fluctuations in your income (several months with negative income and now knowing exactly when enough things will finish so it's positive again for instance).
On the other hand it lead to strange things like not being able to start a gigantic megastructure project over 5 years because right now you lack 10 minerals for the total cost of the layer.
Realistically there is no reason you have to pay upfront too because for those you don't pay the service, you give directly the ressourses necessary for the execution. Services like what enclave propose make sense to be payed upfront.

So I wanted to know if some players think that having a monthly cost for projects (with a tooltip showing the total cost, the cost/month and the total time) instead of paying upfront better ?
Would being able to start things you can't pay right now but know you could in the long run be interesting in the game ?
Or on the other hand do you think it's useless complexity (with the risk that an inexperienced player starts too many projects and end having all of them locked because of a lack of ressourses) and the upfront payment model is good enough for Stellaris ?

That is just not something you can change so late in the game. This part of the economic model is decided very early on.

The Galactic Market has made this largely irrelevant. If you can not buy 10 minerals, you propably should not start that megastructure anyway.

The only place where I sometimes miss this is Influence. It would be nice to be able to take up a "Influence Loan" if you just ran out of Influence and the Galactic Market bidding started. Or a very important resource edict ran out.

 

[permeakra]

Might it be that the ai is here capable to take the monthly trade deals into account, but some players are not?

Certainly.
Ai in general is better with detail that are individually simple.
.... Ok, this sounds wrong.

What people often do is to some extent antropomorphise AI. AI isn't a fool or a human. It is a set of pattern-matching code, executing some actions in response to conditions. Nothing more. Since the code is executed by usually fast computer, AI has excellent reaction time, and since it is executed by a number crunching machine, AI can be programmed to match against fairly complex criteria. For example, calculating optimal distribution of pops to get maximum resource outcome is fairly trivial, if the AI devs actually bothered with it (but there might be valid reasons to bother with something else instead)

The problems AI always face are two-fold
- It cannot plan. If you see AI planning something, you are sadly mistaken: it simply was coded to react this way.
- It is set in its ways, and the target criteria a set by devs, who might have limited perspective on situations

These flaws were what devs of chess-playing AIs fought their entire existence. When computer and human are going play chess, the first thing human does is to make research on the strong and weak sides of the AI and then plan to exploit it. Since chess is a game with very wide decision tree, the devs always had to cut off many branches from considerations and usually used non-adaptive estimation functions to judge what resulting position is better. And this bitten them in the ass again and again.

I suspect that the current Stellaris AI is a simple rule-based code with elements of RNG-driven variations, that recognizes the patterns it was set to recognize and executes corresponding actions, with no estimation of outcomes. Doing something more complex would take a LOT of time and is not worth it until game rules are set in stone. Making the game friendlier to such AI is simple: shift model so AI could do a lot of small decisions with fairly obvious criteria and finally make criterias for optimization. However, what it definitely can be programmed to take into account is small details to annoying to human players to bother.

AFAIK, somewhen in 1.7-1.8 time Wiz once said that it is fairly easy to make AI ravage player's space-based resource production (which early-to-mid game at time could be about 2/3 of total production) by tens of small raider fleets. It would frustrate players to no end, so it wans't implemented. But the protencial to make AI extremely hard to deal with was there.

 

[The Founder]

Certainly.
Ai in general is better with detail that are individually simple.
.... Ok, this sounds wrong.

It is a truth that AI is by nature, leagues better at multitasking.
And also leagues worse as planning.

AFAIK, somewhen in 1.7-1.8 time Wiz once said that it is fairly easy to make AI ravage player's space-based resource production (which early-to-mid game at time could be about 2/3 of total production) by tens of small raider fleets. It would frustrate players to no end, so it wans't implemented. But the protencial to make AI extremely hard to deal with was there.

Oh yeah, that would have been trivial indeed. They could not use the AI's multitasking ability without making the gamepaly worse to unplayable for human players.

You really just need to play a few rounds against the AI in Planetary Annihilation on a multi-planet System to see it for yourself.
The AI can handle 5 planets with 50 squads each as easily as 1 planet and 1 squad.
Meanwhile we humans struggle with one planet and a blob.
If there was not the option to have planetary patrol orders for the airforce out of the factory, the games human/AI game would be unplayable.

 

[Acheron]

I remember a SPIEGEL article, the author basically said that Artificial Intelligence is a misnomer, all we have are virtual sieves stacked upon another, very many sieves in cases but still nothing too be to proud of.

 

[Dorian Ertymexx]

That is just not something you can change so late in the game. This part of the economic model is decided very early on.

The Galactic Market has made this largely irrelevant. If you can not buy 10 minerals, you propably should not start that megastructure anyway.

The only place where I sometimes miss this is Influence. It would be nice to be able to take up a "Influence Loan" if you just ran out of Influence and the Galactic Market bidding started. Or a very important resource edict ran out.

Oooh, I can see this being a grand Diplomatic thing. Like asking a "Favor", and then owing whomever you took it from.

 

[The Founder]

I remember a SPIEGEL article, the author basically said that Artificial Intelligence is a misnomer, all we have are virtual sieves stacked upon another, very many sieves in cases but still nothing too be to proud of.

To properly differntiate, scientists now talk of "Artificial General Intelligence" when speaking of the Skynet stuff. Particular preventing the Skynet stuff.

 

[Zergor]

I am not that worried for the game's AI (if done correctly).

People have said it: AI doesn't plan but it's not necessary.
If the system was a monthly payment system the only thing the Ai would have to do is to calculate if adding a new monthly payment would make it's economy colapse at some point or not.
Simple example:
AI lose 50 ressourse X / month and has 800 and will gain 50 instead 50 in 10 months because something costing 100/month will finish.
It wants to do something costing 50/ month during 10 month.
It will calculate that it can't do it or go bankrupt in 8 month.
In two months it will calculate that it's possible again and will do it.

Ok this doesn't take into account some fluctuations like war, new starbases and events but it would be a good estimation.

Plus with a monthly payment system the solution to reaching 0 is to automatically slow everything so that equilibrium is reached.
A 200 gain of X with a 400 use of X for projects would mean that at 0 all those project would go at 50% speed (so 50% monthly cost) and it would work.

 

[The Founder]

If the system was a monthly payment system the only thing the Ai would have to do is to calculate if adding a new monthly payment would make it's economy colapse at some point or not.

Then it would come into Deficits because of some other issue (Occupation or loss during war, running out of a bonus like Edict) and the AI was still bleeped.

Recovering from such a blow in a streaming economy just does not work that well. Pay out front is a tried and tested approach.