stellaris does genuinely just have too many moving parts for the AI to keep up, unlike civ or galciv or even eu4. there are so many indirect modifiers, variations on what actions are best to take for different ethics/civics/species traits/etc., that it's just an impossibly complex problem. even if it were solved, the game would chug trying to play itself properly. the glavius approach of giving each AI behavior modes based on their current circumstances, budgets to aim for in each mode, and then letting the base game AI (which can at least target a resource and build for it) handle the rest is probably the best that can be hoped for
|
|
# ? Mar 20, 2021 18:34 |
|
|
# ? May 12, 2024 11:38 |
|
Sampatrick posted:no ai in any paradox game has ever been remotely good and it seems incredibly likely that they will never be remotely good because thats not a design priority for anyone involved in these games. the challenge in paradox games is always going to be the starting circumstances and then, in some games, some big event nation that spawns with essentially infinite units.
|
# ? Mar 20, 2021 19:23 |
|
Jazerus posted:stellaris does genuinely just have too many moving parts for the AI to keep up, unlike civ or galciv or even eu4. there are so many indirect modifiers, variations on what actions are best to take for different ethics/civics/species traits/etc., that it's just an impossibly complex problem. even if it were solved, the game would chug trying to play itself properly. Oh bother. Again, the most interesting opponent in Stellaris is a moving wall of death. How good is AI as an opponent has very little to do with the complexity of systems in a 4X game.
|
# ? Mar 20, 2021 19:31 |
ilitarist posted:Oh bother. Again, the most interesting opponent in Stellaris is a moving wall of death. How good is AI as an opponent has very little to do with the complexity of systems in a 4X game. i was responding directly to you and you're just re-asserting the same thing as your previous post instead of, like, engaging with the discussion. i don't think it holds up entirely because how the AI builds its planets, which diplomatic actions it chooses to undertake when, when it chooses to war and when it doesn't, its response to things like the crisis, how it moves its fleets during war, those all affect the "AI quality" that the player perceives. it's not just a function of fleet size many of these things matter a lot less in a traditional 4X in terms of player perception, compared to stellaris
|
|
# ? Mar 20, 2021 19:40 |
|
Wasn't clear. Many of the things you've listed (apart from moving fleets which is a chess-like task for AI) are exactly what I'm talking about: very simple calculations that don't require smart AI but a well-balanced dumb simple rule reaction. Let AI apply one of few build orders depending on a type of colony and it'll mostly be fine and sensible, and AI will have time to spend elsewhere. Give AI simple rules about declaring wars depending on its personality and it will be more interesting than Stellaris approach of giving AI complex considerations always resulting in it sitting there motionless. I'm talking about Crisis being the most interesting opponent because it's a perfect example. You have dozens of boring opponents with supposed personality and complexity that only exist to be tools on your way to the big bad dumb wave of enemies. ilitarist fucked around with this message at 20:14 on Mar 20, 2021 |
# ? Mar 20, 2021 20:12 |
|
I think you both might be coming at the same thing from different directions. Stellaris has a lot of moving parts, and all of an empire's parts really should impact how the AI reacts to things. But because it has arguably too many moving parts each of those parts when you take them all into account it all kind of comes out as a generic mush. So you need to either reduce the moving parts or start ignoring some of them (and increasing the individual impact of the remainder) when it comes to decision making. I feel the same way about a lot of the moving parts in Stellaris. There's a lot of cases where you have three or four distinct systems all bumping into each other when one or two better integrated systems would be smoother and lead to more impactful decisions.
|
# ? Mar 20, 2021 20:44 |
|
I mean no strategy game has a good AI and they are honestly all pretty bad. Now you can assume they all don’t prioritize it or more likely it just isn’t possible right now
|
# ? Mar 20, 2021 20:52 |
|
Imagine if every single AI in a game of EU4 played like florryworry.
|
# ? Mar 20, 2021 20:57 |
AI design in strategy games has to be the most thankless job in the industry. You get poo poo on by everyone, but the moment you make it so that your Total War spearmen reliably turn to brace a cavalry charge everyone just screams that the AI cheats and knows you're charging it and can effortlessly counter you and now your horses aren't as fun to play with anymore. I'm pretty sure there's a knife edge where the AI is just competent enough that people are neutral towards it and that's about the best you can do. Deploying units in a siege in TW:3K is like this. You can cheese out the AI well enough and get them to deploy everyone on the wrong side of the city, at which point people are like haha dumb AI! But if the AI set up their forces "well" it would just be accused of cheating to know what you were doing and now you can't game it. FWIW I think PDS games generally have solid AI that never takes me out of a game like Civ's AI sometimes does. Anno fucked around with this message at 21:26 on Mar 20, 2021 |
|
# ? Mar 20, 2021 21:20 |
|
Anno posted:FWIW I think PDS games generally have solid AI that never takes me out of a game like Civ's AI sometimes does. Yeah same, the AI was actually one of the things that really impressed me when I first got into Paradox games. Especially compared to other games in the genre.
|
# ? Mar 20, 2021 21:39 |
|
yeah for me EU4 strikes the right balance of general AI predictability and competence with plenty of challenge built-in with starting circumstances. Really enjoyed my Sun God run.
|
# ? Mar 20, 2021 21:41 |
|
*100k ottomans die in Siberia while Constantinople is besieged*
|
# ? Mar 20, 2021 21:49 |
|
Weirdly enough I’ve never had that happen. Not that it doesn’t but I only ever get competent sieve AI never idiot siege AI. Total war on the other hand that has some fun stupidity
|
# ? Mar 20, 2021 22:34 |
|
CharlestheHammer posted:I mean no strategy game has a good AI and they are honestly all pretty bad. Does Football Manager count as strategy? The AI is good enough there, even if the game is a glorified Excel sheet Anno posted:
I can't recall the last major war I've had in Europa univeralis where the AI didn't flood the most remote corners of an adversary to capture level 1 forts while happily abandoning their homeland. If a war hits a certain level of province scale, the AI does incredibly bad decisions. It's a shame because i swear EU2 and 3 weren't like this, neither was 4 for a long time (either that or I'm just remembering with rose-tinted glasses)
|
# ? Mar 20, 2021 22:47 |
|
Well I do think complication adds to it games are only going to get more complicated and the AIs weren’t great before ain’t gonna keep up
|
# ? Mar 20, 2021 22:52 |
Mans posted:I can't recall the last major war I've had in Europa univeralis where the AI didn't flood the most remote corners of an adversary to capture level 1 forts while happily abandoning their homeland. I think this is another tricky problem, though. A lot of people dislike the Total War: Warhammer AI because it won’t take fights on the campaign map unless it thinks it can win. Which is probably what you’d think a “good” AI would do, but makes for an uninteresting opponent for a human player. And like when CK3 launched it would just beeline your capital if it didn’t think it could take your forces in combat to try an capture your character and end the war. People hated that poo poo, but it was something that if a player did it they could feel smart and laugh at the AI. Not saying any of the combat AI is any strategy games isn’t prone to plenty of mishaps. Just that it’s incredibly incredibly difficult and can probably go from good to bad extremely easily if a couple things change.
|
|
# ? Mar 20, 2021 23:14 |
Anno posted:I think this is another tricky problem, though. A lot of people dislike the Total War: Warhammer AI because it won’t take fights on the campaign map unless it thinks it can win. Which is probably what you’d think a “good” AI would do, but makes for an uninteresting opponent for a human player. I think most people wouldn't really mind this if catching AI armies was easier/less frustrating. As it is the AI just force marches [i]just[i] out of range and you don't really have any tools to catch up to them.
|
|
# ? Mar 20, 2021 23:19 |
Staltran posted:I think most people wouldn't really mind this if catching AI armies was easier/less frustrating. As it is the AI just force marches [i]just[i] out of range and you don't really have any tools to catch up to them. Well yeah I’m sure people would prefer being able to catch AI armies more easily. I’m just saying that there are all sorts of AI design potholes arising from people thinking they want to play against a “smart” AI but not actually enjoying it in practice.
|
|
# ? Mar 20, 2021 23:27 |
|
Staltran posted:I think most people wouldn't really mind this if catching AI armies was easier/less frustrating. As it is the AI just force marches just out of range and you don't really have any tools to catch up to them. I mean, that's exactly what a player would do as well. It ties up your army and keeps their army as close to the objective as possible. I'm reminded of Heroes of Might and Magic 3 which had similar issues and where players would use chains of heroes to bait the AI to coming into range before charging forward and handing troops off until the leading hero suddenly had an army and could wipe out the enemy. And it's tricky because we don't often want optimal play, we want something that is challenging, but ultimately surmountable. We want the AI to make some level of mistake so that there's a level of optimal play that we can reach that is just beyond what the AI can do. And honestly, that's an incredibly difficult bar to hit, because what would be an obvious minor mistake to a player familiar with the game might be something a newcomer cannot recognize.
|
# ? Mar 20, 2021 23:30 |
|
My player competitive AI just reloads a save when it makes a mistake.
|
# ? Mar 21, 2021 00:03 |
Dirk the Average posted:I mean, that's exactly what a player would do as well. It ties up your army and keeps their army as close to the objective as possible. I'm reminded of Heroes of Might and Magic 3 which had similar issues and where players would use chains of heroes to bait the AI to coming into range before charging forward and handing troops off until the leading hero suddenly had an army and could wipe out the enemy. That would be really annoying in multiplayer too and people would probably complain about it if anyone actually played head to head campaigns. Games can be designed so that optimal play isn't really annoying to play against.
|
|
# ? Mar 21, 2021 00:14 |
|
I actually doubt they could realistically. Optimal play will always involve manipulating the games mechanics. This doesn’t just apply to strategy games either
|
# ? Mar 21, 2021 00:25 |
|
Dirk the Average posted:I mean, that's exactly what a player would do as well. And most armies in history. When people say that pitched battles were relatively rare and that most of war was sieges and raids, this is why- armies just generally did not accept battles they didn't think they could win, and finding a way to force an inferior force to fight was, like, half of all strategy. Though, I think armies tended to be more fragile things than Paradox games typically present them as. Do you think players would complain if every lost battle was a stackwipe? Lol. Though there were exceptions. One of the most interesting parts of Soldiers and Ghosts for me was the discussion of the Battle of Pydna, and the elaborate lengths Paullus went to to prevent his own men from forcing him into a battle he didn't want to fight.
|
# ? Mar 21, 2021 00:36 |
|
Staltran posted:That would be really annoying in multiplayer too and people would probably complain about it if anyone actually played head to head campaigns. Games can be designed so that optimal play isn't really annoying to play against. It's smart play. I've done it in SC2, admittedly a completely different genre, but feinting an attack with fast forces over and over again keeps your opponent's army occupied and penned in and allows you to expand behind it. Keeping your forces just out of reach, but also just close enough to be a major threat is not a novel concept.
|
# ? Mar 21, 2021 00:56 |
|
Staltran posted:I think most people wouldn't really mind this if catching AI armies was easier/less frustrating. As it is the AI just force marches just out of range and you don't really have any tools to catch up to them. Anno posted:I think this is another tricky problem, though. A lot of people dislike the Total War: Warhammer AI because it won’t take fights on the campaign map unless it thinks it can win. Which is probably what you’d think a “good” AI would do, but makes for an uninteresting opponent for a human player. I tend to complain about dumb poo poo like Stellaris but I have to give credit where it is due - Total War Three Kingdoms has the best diplomatic and army AI I've seen in the genre in a while. Someone there is doing great work. I even had an AI that I was overwhelmingly more powerful than surrender their entire Kingdom to me. It was only a few counties but it simply said "you got me beat, here's all my poo poo". I was at my standing desk at the time and I nearly fainted. AAAAA! Real Muenster fucked around with this message at 01:10 on Mar 21, 2021 |
# ? Mar 21, 2021 01:07 |
|
KOGAHAZAN!! posted:And most armies in history. When people say that pitched battles were relatively rare and that most of war was sieges and raids, this is why- armies just generally did not accept battles they didn't think they could win, and finding a way to force an inferior force to fight was, like, half of all strategy. And not just that, but there's a huge lack of information for both sides. Like you might know you've got about 12k combat ready guys with you, but you won't have a great idea what your opponent has in the field beyond "much bigger", "much smaller", or "about the same". It's probably bigger because the other guy probably wouldn't have taken the field unless he was reasonably confident himself, but still you don't know. Do you stand and risk taking that fight? Usually not, but sometimes your hand is forced. But neither that information disparity isn't present at all, nor are the factors that would make you stand and fight. Also large armies are ludicrously fast, so the game is all Napoleonic encirclement all the time.
|
# ? Mar 21, 2021 01:19 |
AAAAA! Real Muenster posted:I've been addicted to Total War: Three Kingdoms lately because it has a mechanic where if your army starts its turn in foreign/hostile/not friendly land, it has a -20% movement penalty. This means you can catch an enemy army that is in your lands much easier. 3K is indeed extremely good about a lot of this stuff. Gives me some hope that they’ve got a good balanced figured out over at CA right now, because I could sure use it in TWW3.
|
|
# ? Mar 21, 2021 01:23 |
Mans posted:If a war hits a certain level of province scale, the AI does incredibly bad decisions. It's a shame because i swear EU2 and 3 weren't like this, neither was 4 for a long time (either that or I'm just remembering with rose-tinted glasses) the AI has done stupid hellmarches for as long as i've been playing, shortly after eu3 divine wind launched, and they used to be much more crippling for them i feel like modern eu4 has done a good job of sanding the edges off so that the AI can make a few mistakes and still present a credible threat
|
|
# ? Mar 21, 2021 01:26 |
|
I think honestly that far more annoying than poor combat AI is poor infrastructure AI. I can live with the AI being pretty sub-optimal at combat and vulnerable to exploits and tricks, but what really put me off conquest in both Stellaris and Civ 6 is that when you actually take planets/cities, they're worthless. They're just so badly built, and when competent players could probably reduce their decision-making on city/planet construction to a flowchart that feels deeply dissatisfying. Stellaris has probably changed a fair bit, but back when I played it, my expansion flowchart was something like: If any resource in deficit and pops unemployed, build appropriate building or district. Otherwise, colonise. If Planets < 6, build a Balanced planet (Balanced planets should be self-sufficient and produce sufficient food and energy to feed and power themselves, then the rest should be minerals and goods/alloys.) [Once I have a core of well balanced planets, every other planet would be dedicated to one thing for simplicity] If Food < +10, and food stockpile not at max build a Food planet If Energy < +x, build an Energy planet If Minerals < +y, build a Mineral planet Else build a goods/alloys planet. And basically the only thing that would change as the game went on was the size of x and y. This is simplified a bit and I might be misremembering slightly, I don't think I've played Stellaris in about two years now. My gut feeling is that this sort of flowchart priorities system shouldn't be too hard for an AI to follow, but if it is, I'd much rather the AI just flat out cheats so that when I'd capture a planet I'd think "Oh, nice! I captured a forge world." instead of what I actually experienced, which was "Oh, I captured a trash world where nothing's been built and what has been built is useless and insane. Again."
|
# ? Mar 21, 2021 01:49 |
|
Reveilled posted:I think honestly that far more annoying than poor combat AI is poor infrastructure AI. I can live with the AI being pretty sub-optimal at combat and vulnerable to exploits and tricks, but what really put me off conquest in both Stellaris and Civ 6 is that when you actually take planets/cities, they're worthless. They're just so badly built, and when competent players could probably reduce their decision-making on city/planet construction to a flowchart that feels deeply dissatisfying. Yeah this is the worst thing about the Stellaris AI by a huge margin. It really drives home that they aren't really playing the same game as you, at all.
|
# ? Mar 21, 2021 02:00 |
|
AAAAA! Real Muenster posted:I've been addicted to Total War: Three Kingdoms lately because it has a mechanic where if your army starts its turn in foreign/hostile/not friendly land, it has a -20% movement penalty. This means you can catch an enemy army that is in your lands much easier. The 3k games do this by actually not really obsoleting units in the traditional way. Your units damage never really changes, beyond their experience. A spear in the hands of Imperial infantry is the same spear a peasant can get. which means that even though you'll win, you have to play well to win and not get scratched back. So because of this the AI is totally okay picking those fights knowing it can do damage. And since replenishment is pretty bad, that damage matters.
|
# ? Mar 21, 2021 02:11 |
Yea, EU4 AI is good enough I think mostly because EU4 doesn't have enough internal development for the AI to screw up massively. You can be reasonably confident the AI will use their development in sensiblish ways and field a near forcelimit army. In stellaris (and Civ) you realise you've won the game by the midgame when your economics leaves the AIs in the dust, and there's no challenge because of resource/tech disparity and no reward either because of it. This is odd, because the economic maximisation stuff is what you'd expect an AI to be good at - it all involves known information and calculations and planning. But plenty of strategy games haven't worked out or haven't bothered to get their AI to be able to use their economic systems as well as a human player. Nothingtoseehere fucked around with this message at 02:27 on Mar 21, 2021 |
|
# ? Mar 21, 2021 02:23 |
|
Reveilled posted:I think honestly that far more annoying than poor combat AI is poor infrastructure AI. I can live with the AI being pretty sub-optimal at combat and vulnerable to exploits and tricks, but what really put me off conquest in both Stellaris and Civ 6 is that when you actually take planets/cities, they're worthless. They're just so badly built, and when competent players could probably reduce their decision-making on city/planet construction to a flowchart that feels deeply dissatisfying. They also shot themselves in the foot by making cg and alloy jobs tricky for the AI to understand. They've redesigned a chunk of the planet management in the upcoming patch that should make it easier but that's not the only part of the game which has been written human easy but AI hard. The most reliable way to write good AI is to design your game with the strengths and limitations of your AI in mind.
|
# ? Mar 21, 2021 02:33 |
Nothingtoseehere posted:Yea, EU4 AI is good enough I think mostly because EU4 doesn't have enough internal development for the AI to screw up massively. You can be reasonably confident the AI will use their development in sensiblish ways and field a near forcelimit army. suddenly johan's insistence on leaving internal development mostly unexplored makes complete sense
|
|
# ? Mar 21, 2021 05:17 |
|
Jazerus posted:the AI has done stupid hellmarches for as long as i've been playing, shortly after eu3 divine wind launched, and they used to be much more crippling for them The obvious rule that'd limit hell marches would be a limit on how far you can send your armies away from your territory, with occupied territory extending that limit. For a small enough territory you could go for the jugular right away, for a bigger state you'd have to work your way in first. Incidentally a rule that makes a ton of sense from a historical perspective, plus it's one that could be used to differentiate different types of armies/eras/states/terrain. Like, at the level of terrain, anything flat and open would allow more movement, while rough and closed terrain would restrict it, creating "soft-impassable" areas where armies don't feel like striking through. Certain government types/tags could then get a bonus to specific types of terrain, like steppe nomads getting a massive bonus to cavalry army range in the flat and open category. Similarly, new tech could unlock multipliers to strategic movement, eventually allowing Napoleon-style "Let me just march across the Great European Plain all the way to Moscow" style movements. Obviously, the less two the ranges of two states overlap, the less they'd care about each other diplomatically, while if the range of one or both states include the territory of the other they'd start being counted as either friend or foe. I feel like the above would do a lot to limit the feeling of the AI being dumb, since the most extreme behavior would be curbed until you get to the point in the game where the wars are big enough that a single stack hanging out in Archangelsk isn't as big a deal, and it'd still favor far-ranging movement across open terrain which at least has some sense to it, not "Let's invade across five mountain ranges".
|
# ? Mar 21, 2021 06:42 |
|
enjoy this video i found https://www.youtube.com/watch?v=r_TplmqP7tE
|
# ? Mar 21, 2021 16:23 |
|
I did not enjoy it and am now thoroughly vexed, sir
|
# ? Mar 21, 2021 18:24 |
|
Sure would be a shame if EU4, perhaps the best grand strategy game of all time and a game long overdue for a sequel, got a sequel. There's plenty of interesting things to do with the formula and I'm interested to see where they go with it. We know there's a mapgame with a non-historical setting in development too, so I guess that's possible too.
|
# ? Mar 23, 2021 08:31 |
|
Vivian Darkbloom posted:perhaps the best grand strategy game of all time and a game long overdue for a sequel, got a sequel. There's plenty of interesting things to do with the formula and I'm interested to see where they go with it. We know there's a mapgame with a non-historical setting in development too, so I guess that's possible too. this, but for victoria 2
|
# ? Mar 23, 2021 09:40 |
|
|
# ? May 12, 2024 11:38 |
|
Vivian Darkbloom posted:Sure would be a shame if EU4, perhaps the best grand strategy game of all time and a game long overdue for a sequel, got a sequel. There's plenty of interesting things to do with the formula and I'm interested to see where they go with it. We know there's a mapgame with a non-historical setting in development too, so I guess that's possible too.
|
# ? Mar 23, 2021 11:15 |