|
Kitchner posted:I think it's just different experiences. In Civ it always felt like the other Civs were just different types of players all playing the same board game. When you play CKII, EUIV, or Stellaris, you feel if someone is nice or a dick it's because they, in the game, is nice or a dick. Some of the people you meet are basically only there to be nice and die to a bad guy. Nah, I have a decent bit of insight here and AI simply isn't a priority for most strategy game developers. There are many reasons for this, a big one is that there's simply not a good knowledge base in the industry about how to make strategy game AI. There's tons of people with experience coding FPS bot AI and they tend to be hired to make strategy game AI, but making a bot AI and a strategy game AI are completely different disciplines. There's other issues, like the fact that game developers don't play their own games, but I'm completely honest when I say that players being utterly dreadful when it comes to feedback about AI is no small factor. From the perspective of people who make the decisions, you basically always get the same amount of 'the AI is completely braindead and broken' because people will say this and only this no matter how large or small the issue they encounter is. Like you'll hear it just as much over AI literally ceasing to take any actions as you will over the AI, say, picking something that gives it 5 resources instead of 6. It's utterly toxic and completely discourages companies from giving a poo poo about AI.
|
#
¿
Jan 26, 2017 22:27
|
|
|
|
| # ¿ May 9, 2024 21:34 |
|
|
Baronjutter posted:How can we as players give better feedback on paradox game AI ? I'm really glad you guys give a poo poo. There's really just two rules: - Avoid generic, sweeping statements like 'Sector AI is stupid' or 'the military AI is brainhead'. It doesn't help us in the slighest and just contributes to developers not wanting to engage with the poo poo flinged their way. - Try to be as specific as you possibly can. 'The sector AI is bad at building farms, I regularly see my planets starving' is helpful. 'Here is a specific case where the sector AI is letting my planet starve' is even better. AI issues often come down to edge cases and the more we know about the specific case, the easier it is to locate the problem. The best thing you can possibly do is provide us with a save and a detailed report. If the issue is reproducable from the save, we can usually fix it in no time at all, whereas in the worst cases it could take weeks or months for the issue to be reproducable for us.
|
|
#
¿
Jan 26, 2017 22:54
|
|
|
GlyphGryph posted:Especially since: Properly responsive AI is often a really hard problem. You want to make the AI good but not TOO good. You need it to be smart, but also to be stupid - but the right kind of stupid. You often need to build multiple AIs for different difficulties... and if you want it to be genuinely good, you also want multiple AI's for different "personalities" as well so the players have to fight against various strategies instead of the same thing over and over again. It's a difficult and thankless job. I also happen to enjoy it a lot.  Even for me though, the toxic negativity gets really old sometimes.
|
|
#
¿
Jan 26, 2017 22:56
|
|
|
Rakthar posted:I'm messing around with Machine Learning stuff in both Keras and Tensorflow. I'm curious what tools game devs typically use to code their AI, and if there's any type of machine learning / deep learning algorithms they tend to prefer. Basically I'm trying to get a sense of which style of neural networks / ML algorithms they implement. It's also really neat to read your description of the issues from the designer perspective, thanks. I've gone into depth on this topic many a time, so I hope this doesn't come off as dismissive, but... as it stands, machine learning is a complete pipe dream for games of any real level of complexity. About the only games that can employ it outside of highly limited/selective implementations is fighting games, and even there it's in its infancy. The belief in technology and algorithms over simple iterative work is actually a huge issue with strategy game AI, people want easy solutions and they just don't exist. It's about hard work, a bit of design know-how and a thorough understanding of the game you're working on.
|
|
#
¿
Jan 26, 2017 23:00
|
|
|
Libluini posted:This reminds me, the AI in Galactic Civilizations II could do impressive poo poo like using the player to fight a proxy war for them. Why is no-one in the industry hiring those guys to make their AI? At least oddly machiavellian decisions like that made a game interesting, even if the AI ploughed you under. It's not like the really unfun AI in Lost Empire, where the AI-players relentlessly steamroll you if you're not min-maxing as hard as you can from day 1. In the first case, the AI being too good adds to the fun by just messing with you, in the latter case you're just stressed out trying to keep up with your robotically efficient adversaries. The Gal Civ AI is the work of one guy with (as far as I understand it) basically the same approach we at PDS have. Tons of people rag super-hard on the Gal Civ AI, btw, further emphasising my point above. The reason other companies don't do this is that the approach is basically 'leave me the gently caress alone for six months so I can play the game and do iterative work' and that doesn't go over well at most companies. I had an incredible amount of autonomy when I came in as AI programmer at PDS and fought to give that same autonomy to our newer AI programmers, that's basically why we have the AI we have today. Wiz fucked around with this message at 23:06 on Jan 26, 2017 |
|
#
¿
Jan 26, 2017 23:03
|
|
|
Rakthar posted:So current strategy game AI is mostly decision trees and hard coded behavior? PDS AIs are largely utility machines. Decision trees are not a very good way to code a complex AI because you end up with really rigid behaviour and if something breaks at one point in the chain it affects everything that follows. That's one of the biggest issues with the Total War battle AI, for example: It's a giant tree of decisions flowing down from higher to lower level AI agents, and one bad decision at the top will screw everything below. EU3 had a similar agent-based system that I tore out and replaced with the utility machine that runs the EU4 AI, which more or less became the prototype for HOI and Stellaris AIs. It's mostly the complexity. Good ML is all about defining good/bad outcomes and in a game with thousands of things affecting thousands of other things it's simply not doable. The computational issues are there too, but they're not nearly the biggest issue.
|
|
#
¿
Jan 26, 2017 23:10
|
|
|
Libluini posted:From the bits and pieces I remember, the problem with using machine learning to make game AI is that this only works really well if you're building a customized AI and then slowly feed it with knowledge about your game. In the end, you've expended tons of resources and years of hard work on this and get a really, really good AI for your game. Which will only and only work for this one game and immediately break when the devs try to port it into a future project. A search engine playing Jeopardy is a much more fertile soil for machine learning because it has clear win/loss states. Let's try to translate that to say, EU4. What's a win/loss state there? Well, clearly being annexed is a loss, so if the machine AI gets annexed it's making bad decisions. Only, which particular decisions led to it getting annexed? Was it raising stability, fabricating claims? Was it the alliance it formed? Furthermore, there are countries that are going to do much worse on average than others, so you'd have to filter for that as well. At some point you accumulate so much data, it's simply impossible to filter the useful stuff from the noise. You could have a crack team of AI experts going at it for decades and I still don't think it'd be better than something I can put together in six months with a utility machine and some heuristics.
|
|
#
¿
Jan 26, 2017 23:15
|
|
|
Libluini posted:Interesting, sounds like AI-programming is more art than science. Or to put it another way... the best AI programmers tend to also be hobby game designers.
|
|
#
¿
Jan 26, 2017 23:18
|
|
|
Sindai posted:There was a good article on behavior tree vs utility AI on gamasutra a while back: http://www.gamasutra.com/blogs/JakobRasmussen/20160427/271188/Are_Behavior_Trees_a_Thing_of_the_Past.php Yep. I highly recommend this article if you want a good understanding of one of the big issues holding back strategy game AI.
|
|
#
¿
Jan 26, 2017 23:24
|
|
|
Fun fact: The EU4 army AI essentially consists of one gigantic evaluation where each AI army asks itself 'which province do I want to be standing in' every single in-game day. It has no concept of strategy, no long-term plan, just a list of provinces ordered from 'most want to stand in' to 'least want to stand in' with dozens upon dozens of weights and special cases that have grown out of endless iterations. This has some drawbacks, but it also makes it incredibly robust, because it's not trying to follow some long-term strategy that might have ceased to be valid weeks ago. When the EU4 military AI breaks, it usually just does something like pick a poor fight or walk back and forth between provinces for a few days. Conversely, when the Stellaris military AI (which does have a layer of strategic thinking) breaks, it tends to freeze up completely. EDIT: Oh yeah, the one exception in the EU4 military AI is the naval invasion AI, which does try to follow a rigid plan, and is also the part of the AI most prone to catastrophic breakdowns.
|
|
#
¿
Jan 26, 2017 23:29
|
|
|
Quixzlizx posted:How about "don't complain about the AI being broken when you're playing with mods?" Eh, depends on the mod. Like, if you're playing with some giant kitchen sink mod that reworks every area of the game your feedback probably isn't useful to us, but Blorg Pith Helmets aren't going to mess up the sector AI.
|
|
#
¿
Jan 26, 2017 23:31
|
|
|
Kilravock posted:That reminds me on how in older Paradox games where the AI would 'fall asleep' after awhile. You had to save then reload the game to get the AI to start performing actions like to do any diplomacy. What was the cause of that? Decision trees breaking down, most likely. The older paradox AIs were full of specialized AI agents taking temporary control to do a specific thing (like move a unit) and if that agent got stuck in its logic that part of it was then lost until game restart, when it would be reassigned back to the main AI agent.
|
|
#
¿
Jan 26, 2017 23:37
|
|
|
Baronjutter posted:My only problems with sectors was just that they didn't seem to construct aggressively enough. Unemployed pops, empty bonus tiles, tons of resources in the bank and a large positive resource balance yet they would seemingly only build one thing at a time sector-wide. This got seemingly fixed then robots did something fucky, that got seemingly fixed. People still angry sectors don't built planets "optimized" but that's a hard thing to determine since there's no correct optimal state, it depends on your priorities. I've basically completely recoded the sector AI in 1.5 to prevent unnecessary resource accumulation, and the new tooltip I added shows you exactly what they are planning to do with the resources they have. I want to add a 'prioritize happiness' setting if there's time, too.
|
|
#
¿
Jan 26, 2017 23:38
|
|
|
Arglebargle III posted:Have you followed that team that taught their AI to play Brood War on ladder? Yes. It's cool, but Starcraft is also a much, much simpler game for this sort of thing than a PDS title. It has far fewer actors and far less randomness and you can essentially achieve an optimal build order purely through brute force trial and error. Baronjutter posted:That's really great to hear. I hope along with endless vague complaints and semantical arguments you also take in the fact that you guys have made some really amazing games and your dedication and "actually giving a poo poo" are really appreciated. Oh yeah, no worries, I'm definitely feeling a lot more love than hate.
|
|
#
¿
Jan 26, 2017 23:45
|
|
|
Baronjutter posted:I'd just be happy with some moo style planet glassing via fleets. The "cost" should be severe diplomatic reactions from most of the galaxy plus the lost opportunity of invading and taking that planet. The biggest issue with that it simply doesn't mesh at all with a PDS warscore system. You'd be able to knock out giant empires in a single war, which as it stands would completely unbalance the game. I definitely want to get it into the game and I more or less know how to solve this conflict, but it's not coming in 1.5. Kilravock posted:Sounds good. My main complaint about Sector/Planet Management AI was that it would accumulate minerals and energy but would very slowly upgrade or build new buildings on tiles with pops when it could afford it, if ever. I have not messed around with slavery/purging/robots in sectors but I know plenty of people have pointed out problems there. Reading the dev diary's about it sounds like you are changing slavery/purging mechanics to be more automated, which if I am right would also make sector pop management easier too. Sectors no longer make any decisions over slavery or purging in 1.5.
|
|
#
¿
Jan 26, 2017 23:52
|
|
|
PittTheElder posted:Wiz you gotta stop it with all these 1.5 tweets, my hype is already too real. Pfft. We haven't even gotten to the *really* cool poo poo yet.
|
|
#
¿
Jan 27, 2017 18:57
|
|
|
Baronjutter posted:It's a strategy game, the ships and planets are massively over-scale so you can see what's going on. What next, EU4 soldiers just look too big compared to the buildings on the map? Nah, EU4 soldiers are realistic scale. People got a lot shorter after the Napoleonic Wars.
|
|
#
¿
Jan 30, 2017 22:09
|
|
|
Baronjutter posted:Earth was the source of Titanic Lifeforms all along!! Maybe a post-apocalypse alternative to cockroaches? What would be the giant animal that's taken over though?
|
|
#
¿
Jan 30, 2017 22:52
|
|
|
Singing Sunflower posted:I'm going to take that as a confirmation that SC will be included in the Banks patch as the final unannounced feature. This is a logical and perfectly reasonable expectation and I will not in any way or form be disappointed. Sorry but no, Star Citizen will not be included in the Banks update.
|
|
#
¿
Feb 2, 2017 11:38
|
|
|
Nelson Mandingo posted:I'm fine with the other stuff in the paid expansion- and plan to get it. But rights and privileges really seems like something that should be in the baseline game. Almost all of it is. Only the special slavery/purging is paid.
|
|
#
¿
Feb 2, 2017 17:29
|
|
|
ModernMajorGeneral posted:Does this mean picking your species world type is no longer just cosmetic now? Frozen, not arctic. Frozen worlds are uninhabitable.
|
|
#
¿
Feb 2, 2017 17:44
|
|
|
Allow me to explain the biggest problem with something like peaceful annexation of planets through culture: Everyone loves it when they're the one getting planets flipped to them... but when it's the other way around? Yeah, no, not so much.
|
|
#
¿
Feb 2, 2017 20:27
|
|
|
Baronjutter posted:No one likes getting their planets invaded either, but it happens. It should be a thing that only happens with a huge big attractive empire versus some tiny little poo poo country with no chance. If you're already in that position as a player you've lost anyways. People are a lot less willing to accept losing planets when they didn't have a chance to fight back. It's the same with say losing territory through espionage/sabotage, things that you can only fight back against 'indirectly' tend to be very frustrating when they have a large impact on the player's experience.
|
|
#
¿
Feb 2, 2017 20:35
|
|
|
Psychotic Weasel posted:This honestly only ever seemed like a mechanic that existed in the Civ and GalCiv gemes that existed for the player to exploit the AI with - if it was happening to you then you were probably doing something wrong. Yeah, basically. It only works as long as the AI is incompetent at it.
|
|
#
¿
Feb 2, 2017 20:36
|
|
|
Office Sheep posted:You could use the faction system. If the neighbouring system has great happiness a faction that wants to make the planet change emipres emerges. If the faction is strong enough on a planet liberation only wars only empires can annex that specific planet. Yes, that is a fine enough idea, or the planet rebelling and defecting if the rebellion isn't put down. The planet defecting because you built a galactic rock concert stage though, ehh....
|
|
#
¿
Feb 2, 2017 20:58
|
|
|
StrixNebulosa posted:Dark times at Paradox: It was the best of times, it was the best of times.
|
|
#
¿
Feb 3, 2017 13:14
|
|
|
GlyphGryph posted:If you implement a culture system without giving a powerful counterplay system aka "a way to fight back" you have only implemented half of the system so yeah duh players dont lile half finished game systems. When I say 'fight back' I literally mean 'fight back' as in shoot lasers at it.
|
|
#
¿
Feb 3, 2017 14:40
|
|
|
RabidWeasel posted:I don't think this was addressed, Wiz did say at one point on stream that he wasn't going to embrace his (very happy and popular) egalitarian faction to gain fanatic egalitarian because he was happy with his ethics, so I assume that you don't get a 'free lunch' and have to abandon an existing ethic or something. Shifting to Fanatic Egalitarian would've removed whichever of Pacifist and Xenophile had the lowest attraction from my empire ethics.
|
|
#
¿
Feb 4, 2017 16:26
|
|
|
Milky Moor posted:lol didn't wiz even say they were loathe to show these things because of this exact response from theorycrafters Never fails.
|
|
#
¿
Feb 6, 2017 07:56
|
|
|
Talkie Toaster posted:I hope hope hope to god that they don't have this ascension path be 'You are actually in a simulation!'. That was the thing that soured me the most on No Man's Sky. It's just a silly little reference.
|
|
#
¿
Feb 6, 2017 17:16
|
|
|
Gimmick Account posted:Wiz, would you consider giving Prethoryn-infested planets that have been bombarded into a barren state the 'Terraforming candidate' modifier? Since I assume that certain essential life-enabling assets such as a magnetic field or underground water deposits would still be there. Sure.
|
|
#
¿
Feb 6, 2017 18:22
|
|
|
Kitchner posted:So which of the traditions is the "I want to kill and purge everyone" tradition? I appreciate the purity one was basically genocidal maniacs only, but it's a very popular game style and it doesn't really seem supported by any of these. None, we wanted to make the traditions less ethics-specific. There is something else coming for this but we're not ready to talk about it yet.
|
|
#
¿
Feb 8, 2017 20:32
|
|
|
RabidWeasel posted:
My team is awesome, that's how. And yeah, we haven't shown nearly all the cool poo poo just yet.
|
|
#
¿
Feb 8, 2017 21:06
|
|
how the hell do you have so much great poo poo coming at once? This is the most frustrating patch wait ever.
|
Baronjutter posted:And this is just the first major DLC. Wiz do you guys have some longer term plans? General ideas/priorities for the next few big game-changing expansions? Of course. You can get an idea for some of them here.
|
|
#
¿
Feb 8, 2017 21:10
|
|
|
I mean seriously, we have five different modes of genocide in Utopia. I think we have the evil bit covered even without a Tradition for it.
|
|
#
¿
Feb 8, 2017 22:25
|
|
|
Kitchner posted:Please don't get me wrong, I totally appreciate and admire you commitment to genocide simulation, but traditions are something everyone is going to have to adopt (or I guess don't adopt them and deliberately cripple yourself?) and I guess I'm not even so much as looking at numbers and specific effects but more how it's all written. It seems every race ever will adopt at least two tradition trees, and if you're a "kill everyone but me" kind of play through, what do you pick? Expansion, Supremacy, Prosperity, Exploration? Harmony if you also want to be nice to the pops you're not exterminating? Supremacy is outright about conquering so I don't see how that wouldn't fit into a kill everyone playstyle.
|
|
#
¿
Feb 8, 2017 23:13
|
|
|
Tarquinn posted:So, if you only eat shroom people and plantoids, does that mean you are a vegetarian? Eating plantoids is canonically vegan in Stellaris.
|
|
#
¿
Feb 9, 2017 17:42
|
|
|
ChickenWing posted:my uncle works at stellaris and he let me play the game before it comes out and there's a race of pikablu I told you that in confidence!!
|
|
#
¿
Feb 9, 2017 19:35
|
|
|
Baronjutter posted:Technically it's all about the ability to suffer without consent, not their type of biology. A consenting human could make cannibalism vegan, breast milk would be vegan. Honey isn't vegan because the bees didn't consent to have the fruits of their labour stolen by MEN WITH (smoke) GUNS. Honey is theft, and slavery. Intelligent plant people, not vegan unless they give their express consent. Yes but on the other hand eating plantoids is canonically vegan in Stellaris.
|
|
#
¿
Feb 9, 2017 20:02
|
|
|
|
| # ¿ May 9, 2024 21:34 |
|
|
Rakthar posted:This is really weird to me, like honestly. I can't tell the joke if it's a joke. Okay, let me phrase your question for you properly: Is there in fact mechanics for determining whether the consumption of a particular alien race is or is not vegan in the next Stellaris update? What do you *think* is the correct answer to that?
|
|
#
¿
Feb 9, 2017 20:13
|
|