Forum adverts like this one are shown to any user who is not logged in. Join us by filling out a tiny 3 field form and you will get your own, free, dakka user account which gives a good range of benefits to you: No adverts like this in the forums anymore. Times and dates in your local timezone. Full tracking of what you have read so you can skip to your first unread post, easily see what has changed since you last logged in, and easily see what is new at a glance. Email notifications for threads you want to watch closely. Being a part of the oldest wargaming community on the net. If you are already a member then feel free to login now.

I pay for rules for games like Horizon Wars or Urban Manhunt because they usually have some art, some unique lore, and a lot of heart. I also buy rulebooks for long dead games for the lore. Even the rule book that came with Technolog/Robogear minis back in the day was a hell of a read.

I’ve bought rule books for smaller games before and been absolutely burned by the lack of lore or flavor. I’d probably have a huge Conquest: LAOK army if their stupid starter set rule book had included anything other than the driest crunch.

Well, that was about the reaction I expected from the community after reading the title.

I would also have to say no, I'd not be interested in an AI generated ruleset, even if AI tech got better at all.

That being said, I don't think it, or at least potentially future AI. has no place in Wargaming at all.

What I mean is, think of how massive in scale 40k is, how many models per army, how many kinds of permutations of lists you could possible have, how many varied matchups there are, the possibilities are nearly infinite.

While I don't think AI has a place in game design, I think it could be used to aid in game balance, running tens of thousands of simulated games, something it is completely impossible to have actual humans do, and determine the best strategies, units, and play patters and use that data to adjust the rules of your human designed game appropriately. So many people complain about army imbalance (even though I think GW does, in general, a pretty decent job with it), having an AI mathematically determine how strong everything is against the entire field and point them appropriately could finally end that issue.

The interesting bit would be what factors that might matter more to an AI than a human, like perhaps how fast an army would play, or an army being good only if you made the best possible move in any given scenario, which just wouldn't be possible for a human.

It reminds me a bit of the Starcraft II AI bots that long ago determined that mass blink stalker wins every game hands down because and AI has such fast reaction time that they can use the ability far better than any human could.

Anyhow, those are my thoughts on the matter.

 Tawnis wrote:
... having an AI mathematically determine how strong everything is against the entire field and point them appropriately could finally end that issue.
.

It won't, unless it comes up with a system that assigns points costs based on the actual forces on the table rather than just the contents of the army book.

Set points costs can never be even close to accurate, because how effective a unit is depends so much on what else is in the same army with it, what army they are facing, and the objective of the mission in play. The points cost can never realistically represent a units' true value unless it changes as needed to take those things into account... it will only ever be a rough average, and having AI assign that average won't change players' perception of that unit's value in their personal army.


Automatically Appended Next Post:

 Tyran wrote:

When you can find even GW rules hosted on a Russian site, what is the point of paying for rules?

Ignoring the legal aspect for a moment, if you like a product then there's value in supporting the entity that made it so they can continue to make product that you like.

 insaniak wrote:

 Tawnis wrote:
... having an AI mathematically determine how strong everything is against the entire field and point them appropriately could finally end that issue.
.

It won't, unless it comes up with a system that assigns points costs based on the actual forces on the table rather than just the contents of the army book.

Set points costs can never be even close to accurate, because how effective a unit is depends so much on what else is in the same army with it, what army they are facing, and the objective of the mission in play. The points cost can never realistically represent a units' true value unless it changes as needed to take those things into account... it will only ever be a rough average, and having AI assign that average won't change players' perception of that unit's value in their personal army.

That's partially true. However, with much more data accessibility you could point units, at the very least, differently for each detachment, which would help with that somewhat. If something has an incredibly high variance between being very powerful in some situations and terrible in others, that's a point for the actual people to come in and make judgement calls about what to do with it or perhaps it needs a rule adjustment? The point would be to save the human resources for those difficult discissions and use AI as a support system to handle the data collection/crunching and meaningless busywork.

Well, virtually properly configured AI could make billions of playtests in short time and adjust points pretty neat. But rules different things. And should be written by humans

 Tawnis wrote:
If something has an incredibly high variance between being very powerful in some situations and terrible in others, that's a point for the actual people to come in and make judgement calls about what to do with it or perhaps it needs a rule adjustment?

Nope, that's just how things work in these sorts of games. For example, a high rate of fire but low penetration weapon is super effective against weak, unarmoured targets, and next to useless against heavily armoured targets. If you point that weapon against the former, it's going to be overcosted against the latter. If you point it against the latter, it's too cheap against the former. Point it in the middle and it's 'wrong' all of the time. Complicate that further by having more players using one of those armies than the other... and the more granular the points system is, the worse that gets because the variance makes a bigger difference.

AI won't fix that. Yes, a computer program built correctly could potentially analyse game data faster and more efficiently than people can... but that doesn't change the inherent imbalances that are an unavoidable side effect of having variable army lists fighting against other variable army lists with variable objectives on tables with a variable amount, type and layout of terrain. The best you can ever do is find a workable average... and regardless of how much analysis is used to arrive at that average, players will see it as wrong based on their own specific experience on the table.

^^^^ What Insaniak said.

Balance is a Unicron and AI will not change that.

Would I buy AI written rules? Nope. Not only nope, actually, but frak nope.

Despite working in the software industry, I have a very strong dislike for any generative AI (and I've actually worked on some AI/ML backends, so I understand how those things work).

Are there spots where AI might be helpful in game design/development? Probably, but not with a general model LLM. And the time it would take to properly train a model is probably not something that most tabletop game companies are gonna have the resources to do.

Regarding balance... 100% agree that is a cryptid. So long as each side can have different builds of their armies, and those builds are open to multiple permutations, it's a fools' errand to try to seek "perfect balance."

I think it's important to make that point about LLM "AI" which is more like a very, very fancy autocorrect.

Like the idea that a LLM could do anything mathematical is wrong, they're quite bad at maths and it's because they're not built to do maths, they're built to simulate conversation.

It's always very obvious to me when a kid uses a LLM to do their physics homework because the LLM doesn't really understand anything about physics or mathematics, it just can create something that looks like what someone might write about physics, or what a maths solution might look like. But it's usually nonsensical, perhaps mixing terms from related but different fields and so on.

AI is a marketing term for these programs and I'm really against using it for them. I'm also fairly confident when they start charging what these things actually cost, people are not going to find them worth it. There's probably some use cases but nothing that justifies the insane costs of the current way it works, and I'm skeptical that they'll be able to get around the hardware issues they have to improve that end of things.

 insaniak wrote:

 Tawnis wrote:
If something has an incredibly high variance between being very powerful in some situations and terrible in others, that's a point for the actual people to come in and make judgement calls about what to do with it or perhaps it needs a rule adjustment?

Nope, that's just how things work in these sorts of games. For example, a high rate of fire but low penetration weapon is super effective against weak, unarmoured targets, and next to useless against heavily armoured targets. If you point that weapon against the former, it's going to be overcosted against the latter. If you point it against the latter, it's too cheap against the former. Point it in the middle and it's 'wrong' all of the time. Complicate that further by having more players using one of those armies than the other... and the more granular the points system is, the worse that gets because the variance makes a bigger difference.

AI won't fix that. Yes, a computer program built correctly could potentially analyse game data faster and more efficiently than people can... but that doesn't change the inherent imbalances that are an unavoidable side effect of having variable army lists fighting against other variable army lists with variable objectives on tables with a variable amount, type and layout of terrain. The best you can ever do is find a workable average... and regardless of how much analysis is used to arrive at that average, players will see it as wrong based on their own specific experience on the table.

You're correct, but also missing the point.

There will always be edge cases, but when you have an AI that can run millions of game simulations, averages will start to show up based on a given army's performance vs the entire field. It wouldn't just spit out win/loss and VP, you'd get average damage outputs, survivability, contributions to VP, the whole nine yards.

The system will never be perfect, but it would give GW FAR more data to work with on how to balance the game. That's all it is for, data collection. Maybe they work out a math algorithm that assigns a point cost to it, and maybe they do that all by people looking at the data, it doesn't really matter since we don't know how they come up with points for every new edition that launches as is.

My point is that more information can only help with game balance, but the final calls on how to use that information should still be done by people, and that a workable average is really what you want if you are trying to get into that magic 50% win rate that GW strives for. If half the time your a high rate of fire but low penetration weapon feels like its pulling a lot of weight in a match up, and half the time it doesn't, isn't that what they're going for?

No, I get the point, I'm just questioning the actual need.

And ultimately, given the sheer number of variables in 40K, I would question the likelihood that an engine could actually be developed to consider them all in a way that would produce measurably better results than the current system. It's nice in theory, but seems like a lot of work to set up, for minimal actual benefit.

LLMs aren't simulation engines, so moot point.

It may be possible to build such an AI, but it would have nothing to do with the current crop of "AI" everyone is talking about (and would be far more expensive).

I mean, LLMs aren't a viable option for creating a functional set of games rules, either, (at best, you would get something that looks like a set of game rules, but internal consistency and actual function would be unlikely) so the premise of the thread is already relying on new technology.

Me talking about specific ML algorithm and not a chat - bot most here talking about. Why should chat bot do balance game.also talking about complexity of game, it's not really that complex. Algorithm which balancing data center resources we use in company iirc use 700+ parameters.

 Tyran wrote:
I think a better question is how many people even pay for rules when there are so many sources of free rules on the internet.

When you can find even GW rules hosted on a Russian site, what is the point of paying for rules?

Because I'm not a pirating oxygen thief, and don't believe I'm entitled to the content of a book without buying the damned thing.

Quite interesting to read some of the replies here, and as expected it is mostly negative.

Here is an extra (I will say very facetious) question to add to that:
Are you definitely not OK with AI-written rules, but in the same breath are you also OK with GW not crediting its rules writers and developers?

I have read some accusations that some of the recently released rules (notably the new Horus Heresy) are at least in part AI-written. I personally don't think this is the case (in that case it just needed an editor), but it does now open the door to this sort of accusation.

I'm absolutely not okay with them not crediting writers.

 insaniak wrote:

 Tawnis wrote:
If something has an incredibly high variance between being very powerful in some situations and terrible in others, that's a point for the actual people to come in and make judgement calls about what to do with it or perhaps it needs a rule adjustment?

Nope, that's just how things work in these sorts of games. For example, a high rate of fire but low penetration weapon is super effective against weak, unarmoured targets, and next to useless against heavily armoured targets. If you point that weapon against the former, it's going to be overcosted against the latter. If you point it against the latter, it's too cheap against the former. Point it in the middle and it's 'wrong' all of the time. Complicate that further by having more players using one of those armies than the other... and the more granular the points system is, the worse that gets because the variance makes a bigger difference.

AI won't fix that. Yes, a computer program built correctly could potentially analyse game data faster and more efficiently than people can... but that doesn't change the inherent imbalances that are an unavoidable side effect of having variable army lists fighting against other variable army lists with variable objectives on tables with a variable amount, type and layout of terrain. The best you can ever do is find a workable average... and regardless of how much analysis is used to arrive at that average, players will see it as wrong based on their own specific experience on the table.

I don't think that's as good an example, because you'd presumably average it out over thousands of games with random combinations of armies. An anti-infantry unit will presumably have a fairly average scoring against a random mix of infantry and armor.


Where I think it'll be more useful is catching unit buffs which are often easily broken. For instance a character that gives re-rolls to any unit within x range, will have a completely different outcome if the unit is Grots or Grey Knights. It'll also be easier for an algorithm (AI or not) to catch really obscure game breaking combinations of special rules and units.

Being able to calculate playtests in a huge data set would be really useful: 1000pts of X beats 1000pts of Y 50% of the time is perfect, whereas 900pts of X beats 1100pts of Y 60% of the time is broken.

 Da Boss wrote:
I'm absolutely not okay with them not crediting writers.

When you look back at some of the bile and vitriol directed at certain rules writers? It becomes more reasonable.

 Mad Doc Grotsnik wrote:

 Da Boss wrote:
I'm absolutely not okay with them not crediting writers.

When you look back at some of the bile and vitriol directed at certain rules writers? It becomes more reasonable.

I bet Mat Ward wished his name wasn't mentioned on a certain codex or two.

 Mad Doc Grotsnik wrote:

 Da Boss wrote:
I'm absolutely not okay with them not crediting writers.

When you look back at some of the bile and vitriol directed at certain rules writers? It becomes more reasonable.

This. When you know why they stopped attributing, it makes more sense.

 Valander wrote:

Despite working in the software industry, I have a very strong dislike for any generative AI (and I've actually worked on some AI/ML backends, so I understand how those things work).

Same, I actually work with AI, using Machine Learning for diagnostics, but I'm really uneasy about how much generative / LLM AI is being shoehorned in as a solution for everything given as pointed out it's just a giant autocompletion and is still famously but confidently incorrect often.

 Tyran wrote:
LLMs aren't simulation engines, so moot point.

It may be possible to build such an AI, but it would have nothing to do with the current crop of "AI" everyone is talking about (and would be far more expensive).

I agree here too, though I think AI as a term is in such a bubble now pretty much any software is called an AI. A proper simulation engine, possibly even based off of something like Total War, would be able to do all of the number crunching.

An LLM may be useful for more language based tasks like checking for ambiguous wording or words which have confusing/bad meanings in other languages.



It's also worth highlighting that 40K isn't all of wargaming, and that whilst it's a total mess where balance is impossible, lots of other games have closer balance and less variation to grind though.
For example a historical skirmish game probably doesn't have the ability for an infantryman to fist-fight a Titan, or for a unit to have multiple stacked saves, re-rolls etc. The difference between a guy with an axe and a guy with a spear is minimal.


Automatically Appended Next Post:

 kabaakaba wrote:
Me talking about specific ML algorithm and not a chat - bot most here talking about. Why should chat bot do balance game.also talking about complexity of game, it's not really that complex. Algorithm which balancing data center resources we use in company iirc use 700+ parameters.

To most people, AI means a LLM.

But I don't think an ML (or at least a Neural Network, which is what I assume you mean) is the right tool for the job here either. Partly because in order to train it you'd need to have another algorithm generate enough inputs and good outputs, and then you may as well use that algothim.
NN's are great for something small in scope for detecting if an image contains a cat, and it's pretty easy to train given enough data, but where would you get the data for units in a wargame?

Unless you mean some other form of ML.

Not okay with no writing credits.

Also, further up we seem to be talking about using a computer Model to help with balance rather than a LLM. However, there is a limit to what computer modelling can do, and there is a reason no one has done it yet.

Absolutely not. No AI art, no AI writing. It's Theft, and it's scummy as hell.

What about using AI to alter photographs?

Like removing a vehicle from a photo, or adding a street lamp?

Herzlos wrote:

 insaniak wrote:

 Tawnis wrote:
If something has an incredibly high variance between being very powerful in some situations and terrible in others, that's a point for the actual people to come in and make judgement calls about what to do with it or perhaps it needs a rule adjustment?

Nope, that's just how things work in these sorts of games. For example, a high rate of fire but low penetration weapon is super effective against weak, unarmoured targets, and next to useless against heavily armoured targets. If you point that weapon against the former, it's going to be overcosted against the latter. If you point it against the latter, it's too cheap against the former. Point it in the middle and it's 'wrong' all of the time. Complicate that further by having more players using one of those armies than the other... and the more granular the points system is, the worse that gets because the variance makes a bigger difference.

AI won't fix that. Yes, a computer program built correctly could potentially analyse game data faster and more efficiently than people can... but that doesn't change the inherent imbalances that are an unavoidable side effect of having variable army lists fighting against other variable army lists with variable objectives on tables with a variable amount, type and layout of terrain. The best you can ever do is find a workable average... and regardless of how much analysis is used to arrive at that average, players will see it as wrong based on their own specific experience on the table.

I don't think that's as good an example, because you'd presumably average it out over thousands of games with random combinations of armies. An anti-infantry unit will presumably have a fairly average scoring against a random mix of infantry and armor.


Where I think it'll be more useful is catching unit buffs which are often easily broken. For instance a character that gives re-rolls to any unit within x range, will have a completely different outcome if the unit is Grots or Grey Knights. It'll also be easier for an algorithm (AI or not) to catch really obscure game breaking combinations of special rules and units.

Being able to calculate playtests in a huge data set would be really useful: 1000pts of X beats 1000pts of Y 50% of the time is perfect, whereas 900pts of X beats 1100pts of Y 60% of the time is broken.

I design for a similarly complex game (Magic the Gathering). We can collect those type of win rates from our online clients. Data like “decks containing this card had a 54% win rate.” The data is fairly accurate because of the large number of games played. In theory, a simulation engine could generate that data before we print the cards, and we could make balance adjustments. Such an engine would be really resource intensive to create, and Magic is already a game that can be played digitally. I have to assume that a Warhammer simulation engine would be a lot more difficult to create. I can’t imagine that the investment would be remotely worthwhile.

 Platuan4th wrote:

 Mad Doc Grotsnik wrote:

 Da Boss wrote:
I'm absolutely not okay with them not crediting writers.

When you look back at some of the bile and vitriol directed at certain rules writers? It becomes more reasonable.

This. When you know why they stopped attributing, it makes more sense.

Ideally there would the be some way of acknowledging all of their writers and developers without assigning blame for any particular codex flaws.


Automatically Appended Next Post:

 Lathe Biosas wrote:
What about using AI to alter photographs?

Like removing a vehicle from a photo, or adding a street lamp?

I know a couple of photographers who consider this to be more like an advanced photoshop tool. One of them with use the AI and then go back and clean up with regular methods. The other will use the AI and then sell the client a janky-ass image that looks more off the closer you look at it. But for part of a slideshow, it works fine.

I do want to make one thing clear. It is extremely likely GW is already using AI for its writing. GW already owns a massive amount of text on its own lore, rules and art, so scrapping the internet for content isn't needed.

It will be the same for any large company that owns large amounts of text and data on which to train AI.

It is the smaller independent writers and developers that are stuck as either being unable to compete or being forced to use AI. But for corporations? AI is here to stay, specially corporations in games and entertainment industries.

 Tyran wrote:
I do want to make one thing clear. It is extremely likely GW is already using AI for its writing. GW already owns a massive amount of text on its own lore, rules and art, so scrapping the internet for content isn't needed.

Given the sharp drop in quality of both technical writing and lore, that is obvious

Right now I'm playing with AI to make my photos I took look like grainy "1970's black and white spy photos."