Is it true that the AI cheats?

[Aliothkn]

If this is intentional (unlimited energy creds for market stuff) that's really bad way to boost AI. I think it would be more reasonable to go with direct bonuses on resource nodes (as example set mining station income to 6/8/10 instead of 2 according to difficulty), this way still leaves player with room for strategies like cutting off fat systems from AI to drain its economy.

 

[vortexsurfer]

The purpose of the AI is to give the impression of a competing space empire and to give the player a challenge that is fun, that is not too too easy and not too hard. Why does it matter how this is accomplished? The only thing that matters is how the player perceives it when playing the game. You want the experience of the other empires playing by the same rules but it doesn't matter if they actually do behind the scenes. It wouldn't matter to me if the AI was basically braindead and just spawning buildings, ships and fleets out of nowhere in a way that is designed to give the impression of producing them. It certainly doesn't bother me that it is given extra resources to compete with a player. I don't see how it would bother anyone.

Until we have actual AI then AI in a game like this will never be able to compete with a player that knows what their doing and is playing competetively. The solution is to make it play by different rules to some degree. That or you could take away the pause button, giving an edge to the computer that can think much faster than us. You can't both be upset that the AI cheats and that it is not challenging enough, that is completely counterproductive.

Yes, but once you know the fact that the AI just gets resources from thin air instead of building a functioning economy according to the game rules it totally destroys the illusion and breaks immersion (for me at least). This is why i stopped playing now. And this is why I stopped playing Endless Space 2 which has the same issues.

I have no problem with some "buffs" for the AI on higher difficulty levels (for example their energy grids produce more than mine). But an AI "playing" completely out of the game mechanics is a joke.

Why would would an "economic warfare" against an AI empire make sense then?

 

[Verx90]

The AI cheat , maybe but i know that The problem its that theyr selling on the market is not registered in the sistem , what they sold and the price adjustement that should come from it.

Im not so sure the AI actualy cheat (exept for the varius bonusses of resources production and special projects like pop modding ) .

BTW glavius too added a chat in the AI , all new starbases will autospawn for free 1 shipyard and 1 trade route ; he did that for the same reason paradox give cheat , a fast -notrealy risolutive cheat while he try to fix other issues .

In all x4 games AI have cheating , from spawning armyes (total war) to allmap exploration(actualy the majority of the x4) or simply resources bonusses , and the majority of the mod for AI add chats themself or alters the "gameplay" for the AI .

Sometimes they tell (glavius in patch note) somethims they dont . Stellaris AI was NEVER at a competitive level , this didnt change with this patch .

 

[Gratak]

Well whatever the AI does, the player can certainly use the insane prices on the market (2.2.2 only) for his own cheats... Over-produce minerals and food, sell it for insane prices to replace your energy production. Trade the extra energy you get directly to the AI for cheaper prices than the insanity on the market

 

[vortexsurfer]

I think the whole market thing was only introduced to make it easier for the AI. I mean, it couldn't balance 3 resources properly, now there's even more plus a deeper planetary economy/pop simulation. So, if you fuck things up (as AI or human player), the market comes to the rescue. Famines or energy crisis are a thing of the past, as long as you have one resource in the green it's easy to get the others. This takes a lot of challenge out of the game. As I wrote in another thread, it's easy to field large battle fleets with an agriculture economy, which is ridiculous and undermines the game mechanics. Plus, market ist badly implemented (infinite supply, no real price fluctuation)

 

[Verx90]

I think the whole market thing was only introduced to make it easier for the AI. I mean, it couldn't balance 3 resources properly, now there's even more plus a deeper planetary economy/pop simulation. So, if you fuck things up (as AI or human player), the market comes to the rescue. Famines or energy crisis are a thing of the past, as long as you have one resource in the green it's easy to get the others. This takes a lot of challenge out of the game. As I wrote in another thread, it's easy to field large battle fleets with an agriculture economy, which is ridiculous and undermines the game mechanics. Plus, market ist badly implemented (infinite supply, no real price fluctuation)

Guess im the only one that got in a game with food at 4 energy per units , minerals at 6 energy , alloys at 30 energy , consumers goods at 0.30 .

The fact its that the AI count only when she buy from the galatic market (and internal) but its not recorded (and dont alter the price) when selling , so the galatic market price can only increase .

So , yea if you are lucky , you can make a living with food surplus , or you may end up as in another of my gamewith food at 0.10 energy per units , because all the AI were cap with food , and i just selled it at 10000 evrytime i hit the cap .

 

[Red-XIII]

The issue is that this is a strategy game, that a valid military strategy is to snipe the opponent's economic and production base, and so if the AI isn't playing by the same rules as the player, that strategy becomes invalid. Which is the point another poster made in a post now deleted - the AI should be fun to play against. Having the AI blatantly cheat and not follow the same rules as the player often doesn't post an interesting fun challenge - because the AI's economy is based on console hacks, it means that the interesting and fun strategies the player can employ are null and void.

That SOUNDS right and smart, but actually isn't quite there.

You're missing a HUGE counterpoint here.

There a different PI game called Hearts of Iron IV. And it rather clearly shows what exactly you are missing.

You only get fun strategic stand-off if your opponent is capable of putting up a fight strategically.
And it isn't, because it's "AI" it's at the very least AGES away from anything that could be compared to thinking.
We can only "teach" our computing tech to do what we know how to do, and in case that somehow isn't clear - humanity does NOT know how and why it is capable of thinking. But I digress...

All we need form that last part for our discussion is that AI is incapable of thinking. That includes thinking strategically. And that include making any of those highly valued strategic manurers of yours a challenge.

In other words if we can't teach it to counter human strategy (and we can't) our strategy becomes nothing better than an easy exploit.

Which is why so many games focus on forcing the player into "his own PvE game" and at best sprinkle this a bit with illusions of interactive strategy.

AI in HoI IV plays (or at least used to last time I checked) by the same encirclement rules as the player does.
Basically it means 2 things - once you're cut off from your supplies your armies gradually weaken to the point of non existence, and If you run out of the land to retreat to your armies are wiped out.

All it did for the player though is turn HoI into one hell of an boring game, unless of cause you're the kind of player that has fun steam-rolling your opponent instead of having fun through challenge.

Reasonable challenge means more than fair play when it comes to having a fun competition. And there are some rules which put AI too far from reasonable challenge.

Now, the fact that PI implemented a system which their AI is unable to use is a separate blunder, and I'm not trying to defend that, however the angle of this thread is plain wrong.

There's nothing wrong with AI "cheating" (considering that rules of the competition were never the same for AI and for players to begin with, I don't think the term is exactly applicable, but whatever...), though the HOW of it could indeed use some work. After all there ARE ways it could better imitate playing by the same rules as we do.

Also I don't remember PI saying that those economic moves everyone is so hung up on could be used against AI.
They probably didn't think much about it at the time, it's a recurring theme in stellaris - to not notice the reality behind all the imagination spurring crab that it's stuffed to the teeth with, so I'm guessing that focus on imaginary finally caught them where it could VISIBLY bite. But whatever... So yeah, they blundered, no questions there.

But here's a good one - why didn't any of YOU (I address the complainers with this) catch that?
Did you REALLY expect AI to be able to put up a fight in an economic war? REALLY?
I'm more than sure that you just swallowed that line of theirs without thinking.
Because I don't see how anyone would buy it after thinking about it.

So try thinking about it now maybe?
You guys are here because someone fed you a silly line without thinking and then you bought it without thinking either.
And you think that is a good enough reason to talk about trust issues and to blame them for breaking yours? Doesn't even qualify as a half-decent excuse from where I'm standing, let alone as a good reason.

P.S. It's usually not the fact that AI is cheating that's bothering players, it's the knowledge of it.
The only two possible solutions to that are either complete ignorance or understanding and acceptance.
It's up to you if you want to try to solve your own issues in a different way once your old way becomes impossible, or to just keep suffering.

 

[Jman5]

Do those AI mods really make them play "better," or just make decisions more optimally?

I know those sound like the same thing, but there are tons of little choices that can be made to get a little more if a resource, or get things done a little bit quicker, that make the AI more of a challenge, but don't actually make the AI "smarter," or more specifically, more situationally aware of what a good decision is.

One example I've seen an actual AI dev talk about in relation to 4X is tech trading in Civ games, back in Civ 4. It's easy to tell the AI that techs have a value, that you should try to get something close to that value in return, and that you shouldn't trade with someone who might use it to beat you, but you can't really tell it (at least with the understanding of AI at the time) how to know when it is a good idea dump technologies on a civ that has 0 chance at catching up to them for far less gold than it is worth in order to use that extra gold to rush an important wonder or building, but that is a value judgment a human player can and does do all the time, to great effect.

I also remember back when someone made an AI that could beat a grandmaster at Go comparing the number of moves that computer had to consider compared to what the AI does with Stellaris. While it was some pretty rough napkin math, I came up with pre-2.0 Stellaris having more than twice as many possible combinations of purchases you could make with just your starting minerals and energy than are possible in an average turn of Go, and that is arguably when the game is at it's simplest. That inevitably ends up ballooning that to much, much, higher, with a smaller proportion of "duplicate" actions that are basically the same thing in a slightly different order, and the Stellaris AI usually doesn't have a super computer's processing power to work with.

That said, I have no doubt the AI will at times deliberately pick the less efficient option for reasons that have nothing to do with roleplaying a type of empire or limits to the software, but I'm also more likely to believe that, if this cheat is intentional, the most likely reason is as a band aid fix until they get the AI to work at the level of competency they want without imploding.

One of the problems I have in these discussions is it always feels like we're talking past each other.

When you guys think of improving AI, you think: How do you make an AI that can cope with all the bajillion edge cases out there so that it can defeat the tip top players. To do that you would need a super computer with deep learning software, experts at the game providing input, and an entire research team working with it.

When I say game developers could do a lot more to make a more challenging AI experience, I'm primarily talking about improving and adjusting the foundation of its existing decision making. Example: How do I spend my influence? This is already defined in the game files, but how it's defined can make the AI stronger or weaker.

What I and I think a lot of people want is an AI that has solid fundamentals from start to finish. Just give it 1 solid game plan that it can take and have decent run. A ton of edge cases become irrelevant when you make an AI that can generate a lot of resources and pump out a lot of ships.

I'm not asking for the kind of software that can defeat a grandmaster at Go. I want the kind of software that you buy off the shelf and can defeat an average joe player. What we have right now with Stellaris cannot do that.

 

[Typee]

So many of these 4X games have AI mods that prove this wrong again and again

Can you name me some? I'm genuinely interested.

 

[Slynx]

AI seems to be able to cheat 650 energy credits when reaching 0 credits with -49 energy, +6 minerals and +0 consumer goods (machine empire) monthly and a mineral stockpile of 49
AI that reaches negative income/ empty stockpile on minerals, energy, food and consumer goods cheats 1.1k energy credits
Link - Steintim

just out of curiosity - are you sure they just "create them from air"? cuz they may as well sell pops on a slave market (I've witnessed the strange behavior where ai will constantly sell it's pops one after another if it's energy income is negative. even if it meant it'll ruin some of his buildings and damage it's economy even more)

 

[Jman5]

Can you name me some? I'm genuinely interested.

Civilization V has some great AI mods. Civ VI Ai mods improves AI too, but doesn't do nearly as much because so much of the AI decision making is hardcoded. Endless Legend had a pretty good AI mod.

I would say Glavius' AI mod from version 1.9 was quite robust and I have no doubt his latest one will be too at the rate he is improving it. I remember I made a real effort at trying to beat it on the highest difficulty and it was a huge challenge. I never completed it so I don't know if I would have won, but I was a major underdog for centuries of game time. It really forced you to use a lot of diplomacy and craftiness just to avoid getting rolled over. The whole experience was super fun, which is why I so strongly push for better AI.

For turnbased tactical game, Xcom had a really great mod that improved the AI. BATTLETECH's roguetech mod also does a fair bit of AI improvements.

My experience is that whenever a game allows for extensive rework from modders and there is a big enough community, you get an improved AI mod. Some more extensive than others, but better is better.

 

[Foefaller]

One of the problems I have in these discussions is it always feels like we're talking past each other.

When you guys think of improving AI, you think: How do you make an AI that can cope with all the bajillion edge cases out there so that it can defeat the tip top players out there. To do that you would need a super computer with deep learning software, experts at the game providing input, and an entire research team working with it.

When I say game developers could do a lot more to make a more challenging AI experience, I'm primarily talking about improving and adjusting the foundation of its existing decision making. Example: How do I spend my influence? This is already defined in the game files, but how it's defined can make the AI stronger or weaker.

What I and I think a lot of people want is an AI that has solid fundamentals from start to finish. Just give it 1 solid game play that it can take and have decent run. A ton of edge cases become irrelevant when you make an AI that can generate a lot of resources and pump out a lot of ships.

I'm not asking for the kind of software that can defeat a grandmaster at Go. I want the kind of software that you buy off the shelf and can defeat an average joe player. What we have right now with Stellaris cannot do that.

Actually the reason I started with that question is so that we *wouldn't* be talking past each other.

Guess the follow up didn't help with that point; they're two factoids about game AI that I've always found interesting and have a tendency to blurt out in these discussions... though I'm usually better at using them...

Anyway, I do agree on some of your points, particularly that "fun" AI does not always mean "smart" AI. One thing that has frustrated me and a lot of people I know with some of the Total War games is the AI's perfect understanding of where they can place their armies to never have to engage in a battle they don't want to, creating infamous Benny Hill moments where you have to chase an army down across a huge chunk of the map until they finally run to a point they can't run any further.

On the other hand, IMO an important piece of "fun" AI is the appearance of acting like a player might, and not obviously being AI. Part of that is that the AI is allowed to *not* act intelligently, and make mistakes the player can act on, like sometimes moving their army someplace where they can be attacked by a larger force, but it also requires a bit of being able to act intelligently in the sense most people think of in regards of what they want with AI, or at least appear to act intelligently. Paradox, and a lot of 4Xs and Grand Strategies for that matter have tried to do this by doing their best to make the AI follow the same rules the player does, and sometimes giving it cheats to make it appear smart without actually being smart (like war opponents in EU4 knowing where an army is and where it is going for a month or so after spotting it to try and simulate a player making an educated guess.)

This bit of cheating however, if it is not a bug (and I honestly don't think it is), and knowing that it's there, makes it a bit too obvious that we are dealing with AI though, specifically one that can't bring itself to only play "okay" and not implode with a terrible economy.

(This might also be a patch to the fact that currently, and I think this is true in the beta, the AI never upgrades their non-capital buildings ever; I seem to remember Galvius saying that the only way he could *get* them to do it for his mod was creating an event that does it for them; fix for that would go a ways to making them more competitive at a less-than-perfect level of play)

So I don't think it's naive to ask PDX to have the AI play intelligently enough (whether that means broadening the decisions they can make or taking away the bad decisions they were allowed to make) to not require the ability to buy goods on the market for free. Even if that is usually the wrong thing to ask for.

 

[FlyingPhoenix]

That SOUNDS right and smart, but actually isn't quite there.

You're missing a HUGE counterpoint here.

There a different PI game called Hearts of Iron IV. And it rather clearly shows what exactly you are missing.

You only get fun strategic stand-off if your opponent is capable of putting up a fight strategically.
And it isn't, because it's "AI" it's at the very least AGES away from anything that could be compared to thinking.
We can only "teach" our computing tech to do what we know how to do, and in case that somehow isn't clear - humanity does NOT know how and why it is capable of thinking. But I digress...

All we need form that last part for our discussion is that AI is incapable of thinking. That includes thinking strategically. And that include making any of those highly valued strategic manurers of yours a challenge.

In other words if we can't teach it to counter human strategy (and we can't) our strategy becomes nothing better than an easy exploit.

Which is why so many games focus on forcing the player into "his own PvE game" and at best sprinkle this a bit with illusions of interactive strategy.

AI in HoI IV plays (or at least used to last time I checked) by the same encirclement rules as the player does.
Basically it means 2 things - once you're cut off from your supplies your armies gradually weaken to the point of non existence, and If you run out of the land to retreat to your armies are wiped out.

All it did for the player though is turn HoI into one hell of an boring game, unless of cause you're the kind of player that has fun steam-rolling your opponent instead of having fun through challenge.

Reasonable challenge means more than fair play when it comes to having a fun competition. And there are some rules which put AI too far from reasonable challenge.

Now, the fact that PI implemented a system which their AI is unable to use is a separate blunder, and I'm not trying to defend that, however the angle of this thread is plain wrong.

There's nothing wrong with AI "cheating" (considering that rules of the competition were never the same for AI and for players to begin with, I don't think the term is exactly applicable, but whatever...), though the HOW of it could indeed use some work. After all there ARE ways it could better imitate playing by the same rules as we do.

What?

I'm not saying that the AI should be as good as a human. I'm saying that the AI should be connected to the simulation/game mechanics, so interesting and smart strategies are possible.

It doesn't matter if the AI is making strategic maneuvers. What matters is that it is vulnerable to having its weaknesses exploited by smart gameplay.

 

[sicksock]

Yeah the AI cheats like gangbusters. Not just with economy either. They can spawn fleets out of thin air even without starbases.

In my last 2.2 game I captured all of a nations space and was methodically collosus'ing their planets and noticed that ever so often their last planet would spawn a construction ship, science vessel and sometimes a small flotilla (5 corvettes) dispute not controlling the starbase in the system.

As far as AI cheating goes, I wouldn't mind economy cheats so that the AI can keep up with the player and provide a challenge throughout the game. But when the AI simply ignores mechanics that bind the player it cheapens the experience in my opinion.

 

[Ariphaos]

Yeah the AI cheats like gangbusters. Not just with economy either. They can spawn fleets out of thin air even without starbases.

In my last 2.2 game I captured all of a nations space and was methodically collosus'ing their planets and noticed that ever so often their last planet would spawn a construction ship, science vessel and sometimes a small flotilla (5 corvettes) dispute not controlling the starbase in the system.

As far as AI cheating goes, I wouldn't mind economy cheats so that the AI can keep up with the player and provide a challenge throughout the game. But when the AI simply ignores mechanics that bind the player it cheapens the experience in my opinion.

No, those are cases where you've caused ships to go into emergency FTL and that's how they return if they can't find a friendly upgraded spaceport. The same would happen to you if you were losing that hard.

As near as I can tell, the current situation appears to be a case of emergency AI support (giving energy to avert crisis) gone wacko.

 

[UltimateTobi]

I seem to remember Galvius saying that the only way he could *get* them to do it for his mod was creating an event that does it for them; fix for that would go a ways to making them more competitive at a less-than-perfect level of play)

Indeed he said.
The AI is also incapable of repairing once ruined buildings (when a slot closes due to Pop drop below threshold). That wasn't moddable and has been disabled until PI fixes this bug.

 

[SirAlexius]

Personally I can't wait for the day when developers can use deep reinforcement learning agents in grand strategy games. I feel like that would solve a lot of problems.

It would completely change the game to be able to actually observe what the AI does in order to learn how to minmax best and win at the game.

 

[Typee]

Personally I can't wait for the day when developers can use deep reinforcement learning agents in grand strategy games. I feel like that would solve a lot of problems.

It would completely change the game to be able to actually observe what the AI does in order to learn how to minmax best and win at the game.

I wouldn't like it. If you think human players can have cheesy strategies sometimes, you wouldn't believe the crap deep learning networks can pull off.
In a 4X playing against a believable well scripted AI that "acts like a human" is much better than an optimized NN.

Take DotA for example, one thing openAI did was turn on and off the ring of aquila. On when the enemy was trying to last hit to increase openAI's minions armor. Off when he was denying to decrease their armor and deny more easily.

A neural network in Stellaris would pull off insane micro stuff like that all the time. Constant job switching of pops to maximize circumstancial bonuses, constant rerouting of trade routes as soon as piracy gets too high to a route through new systems, constant ping pong game with fleets to stall you when you have the superior firepower and catch you when you have an inferior one, just like in old CK2. It would be a nightmare.

 

[Jman5]

I wouldn't like it. If you think human players can have cheesy strategies sometimes, you wouldn't believe the crap deep learning networks can pull off.
In a 4X playing against a believable well scripted AI that "acts like a human" is much better than an optimized NN.

Take DotA for example, one thing openAI did was turn on and off the ring of aquila. On when the enemy was trying to last hit to increase openAI's minions armor. Off when he was denying to decrease their armor and deny more easily.

A neural network in Stellaris would pull off insane micro stuff like that all the time. Constant job switching of pops to maximize circumstancial bonuses, constant rerouting of trade routes as soon as piracy gets too high to a route through new systems, constant ping pong game with fleets to stall you when you have the superior firepower and catch you when you have an inferior one, just like in old CK2. It would be a nightmare.

I get what you're saying, but it's not like it has to be one way or another. A developer can limit some of the more cheesy high-micro things an AI does by adjusting game rules or creating penalties for shuffling things around too quickly, or locking things down for a time. So let's pretend we have one of these ultra good deep learning AI's running in Stellaris doing the things you suggest and you're a developer tasked with managing it's cheesier aspects.

1. Constant Job switching: This could be mitigated by introducing something like -25% resource generation for first 100 days on a new job.
2. Being cute with piracy patrols: This could be mitigated by adjusting how piracy builds up. Perhaps instead of a fleet totally wiping it out on its first pass it takes several patrols through. So it's more about prolonged and consistent patrols, then just a 1-time quick sweep every couple years.
3. Attacking simultaneously from multiple directions with hit and runs: Definitely something tough to deal with, but the existence of the pause button helps the player keep up with the micro-intensive strategies a super computer might pull off. Having high sensor range intelligence on your border would help the player anticipate and counter this sort of strategy.

The nice thing about Stellaris is there isn't really much micro when it comes to ship battles so some of the crazier stuff like perfect micro in combat that you see in other games wouldn't be as big of a concern.

Beyond that you can also do stuff like put hard limits on the number of Actions Per Minute an AI can perform which forces it to prioritize, or create input delays that aren't noticeable to a human player but really cuts down on some of the shenanigans an AI with 10,000 APM can do.

 

[Todie]

Has anyone seen this happen in 2.2.3?

I spent 20 minutes looking for it today but couldnt confirm anything...