[Dev Team] 3.0.3 Patch Released (Checksum d281)

[mial42]

That isn't what he means. People complaining about the empire-wide growth penalty are saying things like, "It's taking me ten years to grow a pop in the late game!" But that's ten years on a given planet. If they have fifty planets, all else being equal, each planet is growing a pop every ten years, or fifty pops in ten years. There's a pretty huge difference between "It's taking me ten years to grow a pop" and "It's taking me ten years to grow fifty pops". That's why he's saying the people making these comments don't understand the ramifications of the system.

1) When accusing people of misunderstanding a system being the basis for criticism, it's best to use direct quotes, rather than just your own impressions of what they think. You may be misinterpreting what others are saying and substituting what you think they're saying for what they're actually saying. This is doubly true if you (or @Ludaire in this case) accuse a large segment of the criticism of the new change ("I know the new growth system has been unpopular, but so much of that is because the effects are misunderstood.") of being because of misunderstanding, because then you need direct quotes from a large swath of users and not just one or two.

2) It's also a good idea to take into account hyperbole. When people say that they stop growing pops lategame, they probably (most forum goers have decent knowledge of the game and can understand basic mathematics) don't mean that empires literally stop growing. They mean that it feels like they stop growing and their empire stops progressing because of how slow it is. Likewise, if somebody were to write "it takes me ten years to grow a pop," it's reasonable to assume that they're talking about per-planet growth and understand that empire-wide growth is higher then that, and are just being imprecise. This is an informal forum, and people are not always going to be 100% precise about their language.

3) The way players interact with pops is all on the planet level/screen. They grow on a planet level, we provide them jobs via buildings and districts on the planet screen, we resettle them on the planet screen, we purge them on the planet screen, we look at them through the planet screen. The player does not directly interact with individual pops at all on the empire level or through an empire-wide screen. This means that a dramatic slowdown of growth per-planet over the course of the game will make the game feel much slower or even totally stagnant, even if total empire-wide pop growth doesn't actually go down that much (as it might not if you're rapidly adding new planets). If pops grew empire-wide, if we interacted with them on an empire-wide basis (meaning building buildings and districts, resettlement, purging, viewing, etc), then this change would probably have been better received. But that would be a completely different game, not Stellaris.

 

[Lidhuin]

Also, given that growth is still per planet, and growth is still soft-capped by pops (and not planets), a 50 planet empire will still grow faster than a 1 planet empire. There is no empire wide growth.

 

[Moonspring]

Resettlement should be slow enough to give time to build a new building (1 year or so). The problem is that having these times vary by empire type and modified by progression is idiotic.

We can build buildings ahead of time, tho.
If resettlement time didnt exist Pops moving away would not be a problem. Because they'd just move right back once the building is done. With building slots no longer relying on having Pops to support them, we dont risk buildings being destroyed by resettlement anymore either.

So, there really is no downside to having resettlement be instant, other then having to care about Jobs available more since auto-resettlement will be better.
Then again, lower cd on resettlement means more calcs being done. Is that then, the reason for the cooldown?

That aside, resettlement being slow means Pops stay unemployed for far longer time. Currently base rate is 5%, which means they'll be unemployed for 20 months on average. At a base energy rate of 6 (which is very low) you miss out on 120 energy in those months. Enough to pay the old resettlement cost on its own.

 

[Ludaire]

When accusing people of misunderstanding a system being the basis for criticism, it's best to use direct quotes

They literally quoted the main misconception directly in the very post you're responding to. And to be frank, I'd rather we not start quoting random people from other threads (likely pulling them out of context) to try to prove points.

The way players interact with pops is all on the planet level/screen.

This is essentially saying "All I can see and touch are trees, therefore the forest doesn't exist." Emergent mechanics exist, and they can be critically important in how a game works. The fact that you can't point to a number on the UI is irrelevant.

Plus, we do interact with pops on an empire wide level. There's a count at the top of your screen. The species screen, including the setting of rights (which is how we determine who is purged) is either empire-wide or galaxy-wide. The planets and sectors screen groups pops by planet and sector, but still shows your entire empire. All the resources you get from pops, which is the reason you care about them at all, is empire-wide, while planet-specific stuff like amenities and housing are less critical.

 

[Bayushi Tasogare]

Well, I'm pretty much aware of it. So this is what I think about it:

1. blackbox is completely fine. For different AI levels you can just take AI with different games played (1000, 10000, 100000 for example). It will roughly reflect their skill. Anyway, it's much better than just buffing resources they get
2. Hmm, mods will suffer for sure. But if AI will be competent maybe mods won't be this needed anymore?
3. Well, you don't. If it doesn't research enough then you just add research points to target function. If it still doesn't prioritises it you add more weight until it behaves as you wish.

Beign said, I'm not sure if you've seen AI playing dota 2 on International. OpenAI outperformed the best dota player in 1-1 competition. It was shocking new for people who always considered AI as "bad opponent to train weakest of the players". But no more - the best human players tried to beat them and only succeeded in single cases via cheasing that doesn't happen in game normally (so AI couldn't learn how to counter it).

All in all, ANN is a step back in some sense but only to make 2 forward. I'm not sure PDX Has resources and/or willingness to do so, but it would be amazing if they tried. As I have shown above your concerns can be resolved with relative ease.

Blackbox is not fine. If you don't know how the AI is doing what it's doing, the risk of a Type III error increases drastically, and that means you have no idea how stable the AI is. It might be okay for this game, but what about next game? Also, by the nature of what we're discussing, you're going to need an AI for every possible empire.

And modding isn't always for balance, as others have said. How do you expect to allow modders access to the AI?

I... don't think you really are aware of what you're asking, if you think it's as simple as modifying a target function. Here's a question for you: do you want an AI that can build a stable empire, or do you want an AI that can win the game? Those are different goals, and thus different target functions. Also, what's the reward loop? Before you suggest that an ANN should be used, you really need to be able to answer basic questions about the expected form of that ANN.

I'm actually quite familiar with ANNs playing various games, though not Dota 2 in particular. Also, at the risk of sounding a bit terse, Dota 2 is not nearly in the same category of game as Stellaris. The problem space is nowhere near as large in Dota 2, or any game that I've seen an ANN applied.

To be frank, you have shown that you don't understand my concerns, not that they can be resolved. Machine learning algorithms have a fairly well-defined set of problems to which they can be applied, and there are both theoretical and practical issues with trying to create an ANN AI for Stellaris.

Re:

It's not just hype, as I told above we got buts way beyond anything valve could create in almost 10 years. And unlike strong AI in other games it doesn't just click a lot: it has its own machine strategies and IIRC prople learned something from the competition I mention. It also lead to Valve balancing the game.

All in all, ANN are really hyped but I do believe this is one of the areas they are more than applicable and history shows it makes great results

ANN is overhyped everywhere. I'm a little concerned that you believe that the type of game (i.e., problem space) is immaterial to the efficacy of ML algorithms like ANNs.

In the quant finance world, NNs (and machine learning in general) is the latest buzzword and it just drives me insane.
Regardless, it would simply be faster to program the AI to be 'good' directly.

You might compare it to a simple programming exercise like trying to find different ways to sum a list of numbers from 1 to N. There are high level languages, low level languages, various tricks and functions - but the fastest method that will beat all of them is if you can employ knowledge about the data to 'cheat.' In this case, the classic summation formula n(n+1)/2 that can crunch an arbitrary list in one step.

Similarly, we don't really have to simulate millions of games to know what techs are good and which are bad. We already know this information. At the very least, we can use that as a starting point. We don't need optimality for this problem, just 'good enough.'

Oh, the finance world suffers greatly from the lack of explainability 'feature' of ANNs, because the compliance/oversight becomes a nightmare. As I understand it, government entities don't like approving tools that they can't understand.

I urge you that writing AI is more complicated than writing one fold or series summation.


If writing AI is this easy and PDX has all the information then they aren't doing this because they are lazy? There is a huge demand for AI And the only thing that could stop doing this is huge costs (or bad management ofc, but as I see it Stellaris's one is quite competent).

Okay, here's the question I want to ask you: WHY ANNs? We've had the optimization technology for solving games like Stellaris for over 50 years: nonlinear programming. (If it wasn't for the amenity/happiness/stability interaction, you could use straight-old linear programs from the 1950s.) If you're going to hire someone so knowledgeable in ANNs to build an AI for Stellaris, you can just as easily hire a nonlinear programming analyst, and get a perfect solution, not just a heuristic one like the ANN would provide.

In conclusion, could you please give me an example of a 4X or grand strategy game to which an ANN AI has been applied? Your DOTA2 example is... not a very good one, in this context.

 

[mial42]

They literally quoted the main misconception directly in the very post you're responding to.

They didn't though. They put their own statement in quotation marks, which is not the same thing.

And to be frank, I'd rather we not start quoting random people from other threads (likely pulling them out of context) to try to prove points.

Then don't say things like "I know the new growth system has been unpopular, but so much of that is because the effects are misunderstood." If it's just one person you're responding to who has that misconception, respond to that person. If it's someone irrelevant to the conversation, don't bring it up at all. If you try and say that a large chunk of the unpopularity of the empire-wide malus in the wider community is because people don't understand it, you should back that up.

This is essentially saying "All I can see and touch are trees, therefore the forest doesn't exist."

If all you can see and touch are the trees, then the trees are relevant. The forest might also be relevant (I wrote a post defending the empire-wide malus when 3.0 first came out along those lines, although I changed my mind after actually playing a couple of games with the it. Don't respond to an argument I'm not making), but that doesn't make criticism of the tree-level effects invalid.

Emergent mechanics exist, and they can be critically important in how a game works.

And there's nothing emergent about the empire-wide malus. It's not a complex system that arises out of the interaction of simpler parts, it's an extremely simple formula.

The thing about the empire-wide malus is that (while it solves real problems, unlike logistic growth, which is terrible) it doesn't work well with the rest of the game. The UI still indicates (with the empty building slot icon, with the locked building slot icon, with empty district icons, with the upgrade capital button) that you should fill up planets. The AI (both other empires and planet/sector automation) is premised on constant, predictable planetary growth. You still get warnings for unemployment/low housing on all planets, not just ones you're actively developing (developing 4-5 planets at any one time is a perfectly valid strategy, but not one actually encouraged by the game). It doesn't work well with assembly in general and mixed growth/assembly empires in particular. It moved conquest from "strongest strategy" to "overwhelmingly overpowered compared to not-conquest." It incentivizes weird cheese to get around, like keeping around tiny empires to raid for pops. There's no indication of what it is, how it works, or even a vague flavor justification in-game. There's a bunch of smaller problems that have been beaten to death on other threads.

I don't want to start on argument on the merits of the empire-wide malus since that's not the point of my post, which is that you shouldn't accuse people who don't like the the malus of misunderstanding the system without actual evidence and that per-planet pop growth is a perfectly reasonable thing to be concerned about.

The fact that you can't point to a number on the UI is irrelevant.

It's completely relevant that the game's pop UI and pop-centric gameplay are all on the planet level. Somebody complaining that a planet will take a decade or more to grow a pop, and that this will only slow further (thus effectively halting this planet's development, especially if you're nearing endgame already) is a perfectly valid criticism, since they're interacting with that pop on a planet level.

Plus, we do interact with pops on an empire wide level.

Hence why I included the word "individually."

There's a count at the top of your screen. The species screen, including the setting of rights (which is how we determine who is purged) is either empire-wide or galaxy-wide. The planets and sectors screen groups pops by planet and sector, but still shows your entire empire. All the resources you get from pops, which is the reason you care about them at all, is empire-wide, while planet-specific stuff like amenities and housing are less critical.

Obviously empire-wide pop count and composition matters. It matters much more to actually winning the game then anything you do on any one individual planet. But you're not building buildings on your empire, you're building them on a planet. You're not seeing a pop grow in your empire, you're seeing them grow on a planet. Actual gameplay directly related to pops (that is, buildings and districts through the job system) as opposed to the resources they produce is overwhelmingly planet-based.

 

[Derp]

So what's the ramification that is being missed? A fifty-planet nation that grows fifty pops per decade is growing only one pop per planet per decade. That's only an acceptable level of growth if you think most planets shouldn't grow at all, which is an absurd idea. If most of your planets aren't going to grow, what's the point of going through all the effort to colonize them?

mature, specialized worlds don't need to grow anymore, so the pops they generate are free to resettle to colonies that do need pops

 

[maxirage]

This isn't the tile system. If you play with logistic growth there is a temporary planet softcap of around 35 pops, but after that it's 100+, which is effectively infinite. Planets in Stellaris are never really "done". Not anymore.

 

[general_de_golle]

Will we get a fix to AI priorities for building labs? AI seems not to build them at all till midgame, so it can be easily bitten by player and by endgame crisis.
So any fix or beta fix next week?

 

[Ludaire]

you shouldn't accuse people who don't like the the malus of misunderstanding the system without actual evidence

All I wanted to do was quickly offer my feelings on the patch and share my impression of the conversation. It wasn't even a long post with a bunch of analysis. It was just a quick sharing of my feelings. Is it really that horrible to express frustration with a change in the game or the state of the conversation around it without writing a dissertation with a dozen verified sources?

 

[AresOfThrace]

Hey, one question.

I haven't seen it anywhere in the notes, but has the issue been fixed that sometimes ambassadors that you sent into other empires as "spymasters"
have a chance of randomly being accused of crimes and the player is asked to leave them to the justice of said empire?

Unless I completely misunderstood the entire espionage-thing, shouldn't that be something that can happen only to an actual ambassador?

 

[Alfray Stryke]

Hey, one question.

I haven't seen it anywhere in the notes, but has the issue been fixed that sometimes ambassadors that you sent into other empires as "spymasters"
have a chance of randomly being accused of crimes and the player is asked to leave them to the justice of said empire?

Unless I completely misunderstood the entire espionage-thing, shouldn't that be something that can happen only to an actual ambassador?

Your spymasters are placed in the other empires under the cover of being ambassadors - completely intended!

 

[malakhglitch]

Hey, one question.

I haven't seen it anywhere in the notes, but has the issue been fixed that sometimes ambassadors that you sent into other empires as "spymasters"
have a chance of randomly being accused of crimes and the player is asked to leave them to the justice of said empire?

Unless I completely misunderstood the entire espionage-thing, shouldn't that be something that can happen only to an actual ambassador?

I do not think that is a bug. What do you think your spymaster's cover is?

EDIT: Ninja'd by @Alfray Stryke

 

[AresOfThrace]

Your spymasters are placed in the other empires under the cover of being ambassadors - completely intended!

That begs the question why I can build spy networks in genocidal empires then.

 

[malakhglitch]

That begs the question why I can build spy networks in genocidal empires then.

Just because they want everyone else dead doesn't mean they have to be rude about it. Now, kindly enter the chamber like a good xeno.

 

[AresOfThrace]

Just because they want everyone else dead doesn't mean they have to be rude about it. Now, kindly enter the chamber like a good xeno.

Not helpful.

 

[Tech Noir Synth]

I really think a lot of people struggle to understand this.

Yeah, an Agrarian Idyll is probably not going to compete with a Technocracy on a Galactic scale. It would be kinda weird if it did.

This is not a great example and not a great excuse either.

We are not writing "please take a look at balance" just to get the devs to make every single civic on par with each other. I'm not going to argue about some of the less significant changes. Not everything is going to be perfectly balanced, BUT

We have been asking devs to look at the balance because the past and current balance has been so incredibly bad that it hurts the game significantly for years!

The easiest example would be Spiritualists vs Materialists.

I have started playing the game around Megacorp. Synths were incredibly bugged as were many things when patch 2.2 came out and later on they received some needed buffs.

But ever since patch 2.2.5 Synth Ascension has trumped everything in this game by far. Spiritualists never came close to it aside from evasion destroyer shenanigans.

No one can deny that the edict capacity changes were a giant slap in the face from paradox designers towards Spiriualist players. Who here thinks that having more edict duration is a great bonus? It was already bad before Paradox decided to make the most important edicts permanent. So now the ethic Spiritualist is almost entirely useless. In my opinion, having so much excess influence is not needed and in my playthroughs I am very often at influence cap. But I presume this might be different for other players. Still, the changes to Edict cap were a direct nerf to Spiritualists which is obviously a change in the wrong direction.

Even though Synths received some needed nerfs, its still by far the best Ascension in the game. No one is asking to balance this game perfectly.
But Spiritualists are in such a bad place that it significantly hurts our enjoyment of the game.

We don't even have fun playing Spiritualists because you feel the reduced pop growth and reduced power every time you play.

We don't want to have to play a gimped empire just because we are playing Spiritualists. Thats why the community has made mods like Stefan's perfectly balanced mod, which adds significant changes to both Spiritualists and Materialists and the game has much more variety and is vastly more enjoyable because of it.

Its good to hear that you guys are finally listening and considering balance. But where was this consideration the last 2 years?

You obviously didn't consider balance a lot when you added Edict capacity. Otherwise you would have buffed the Spiritualists edict when you changed edict capacity. You are not doing enough to balance this game. Some simple number changes for Spiritualists Ethic like what has been done in Stefans perfectly balanced mod would have gone a long way. But you did nothing to the ethic.

 

[Sayakus]

AI works! But only if i give it 200% of grand admiral bonuses.....

 

[Pzixel]

I agree that good AI is tough, kind of like good pathfinding. It took Dwarf Fortress something like 10-20 years to fix pathfinding issues IIRC.

That being said, just glancing at the economic "plan" the AI have been given any player with just brief experience with Stellaris would instantly know the AI is going to lose 99% of the time (unless it roflstomps the player early on), because it intentionally targets for zero research up to the midgame and 50 research in the midgame.

Most empires start with about 50 research, so the AIs current plan is to produce less research than it starts off producing... Supposedly, this can result in the AI deliberately destroying research labs, even when it has consumer goods surplus.

Now mind you, it can be totally viable to go for an alloy-focused build to start with and expand super fast before switching over to research. But that's not even remotely what's happening here. Indeed, I can't tell what is happening at all, but it's not good AI, it's not decent AI, it's not even bad AI. It's just straight up terrible AI that can never compete, regardless of how many bonuses you stack - unless they're ravenous swarms or fanatic purifiers and it's still early in the game.

I agree - the problem is what to do about it. I think i proposed an idea that wasn't there before and it may be one which can turn the tide from terrible AI to not so terrible AI. Apparently people don't like it - they think devs should just make AI smarter from the thin air. But the fact is I don't think they know themselves why AI is doing what it's doing (unless they debug some specific situation very closely).

"Just work harder" doesn't work in real world - if something isn't acting as expected you need to pivot some decisions, make machine helping you and so on.

 

[Sayakus]

I agree - the problem is what to do about it. I think i proposed an idea that wasn't there before and it may be one which can turn the tide from terrible AI to not so terrible AI. Apparently people don't like it - they think devs should just make AI smarter from the thin air. But the fact is I don't think they know themselves why AI is doing what it's doing (unless they debug some specific situation very closely).

"Just work harder" doesn't work in real world - if something isn't acting as expected you need to pivot some decisions, make machine helping you and so on.

Glavius could fix AI issues, if modder can do it then what is the blocker for company?