Use of AI for Balancing

Saiytama

In a conversation with a friend, he mentioned that MMORPG games usually do not have accurate balancing due to the fact that we are human and cannot analyze all possibilities. However, this does not entirely apply to Artificial Intelligence. It would be possible to train one to assist in balancing the classes and races in Ashes of Creation, so that it would be responsible for performing simulations and possibilities of builds, thereby providing a wider range of solutions for balancing.

I believe that within a reduced scope, a human could do this. However, when the number of possibilities increases significantly, not even the creators can handle the vast amount of possibilities and builds. In this way, we could avoid having overly unbalanced classes or races. I know that achieving 100% balance is impossible, but we could get as close as possible to this number by using Artificial Intelligence as a means to validate as many combinations of builds as possible.

Unknown

This content has been removed.

tautau

At least for me, a lot of the fun of the game is solving problems like that....figuring out how to maximize your build in a unique way to do the best at whatever your game goals are. To have AI do that for you would be to reduce your pleasure of accomplishment in the game.

Kinda like having AI do your cooking so you get a tasty meal but have no fun cooking. Or just reading book reviews instead of reading the book yourself. Or giving your significant other toys to play with instead of making love to them yourself.

Saiytama

Yes, I am not saying that AI should create builds or do things like that, but we know that often some classes are more used because their potential in PvP or PvE is much greater than others. Even within a rock-paper-scissors balancing system, I still think it's important to use AI to optimize and assist the balancing team. I don't believe this will take away from the players' experience, since our role will not be to balance but to create builds and find paths to success within the diversities the game offers us.

In my opinion, all classes and races should have the capacity to reach a common potential, and from there, the rock-paper-scissors system would work very well. For example, everyone could reach 100 power, where this 100 represents various experiences, items, builds, buffs, and so on. But the 100 of rock would be superior to the 100 of scissors, for example, because of the compatibilities and counters from one class to another.

ItsNoGood

Where/how would you source the data to train the AI?

Saiytama

We have the Ashes database to access various information, and they also have the damage/defense calculations. We can work on the idea of how to simulate and feed the AI so it can run the simulations as accurately as possible. I don't know how the team is currently using metrics to perform the balancing, but we could incorporate the same logic while feeding the AI.

ItsNoGood

I don't think that would be sufficient. To train an AI you typically need thousands upon thousands of examples. Let's take a look at what it would take to train an AI to balance just combat abilities specifically: In this case, the training data you need might look like a list of numbers corresponding to the values associated with each ability (damage, cd, range, mana-cost, etc.), and a corresponding value that could quantify how balanced the combat is as a result. You could generate the data manually for the ability values easily enough by randomizing them, but quantifying how balanced the combat is as a result doesn't seem possible to me. If you somehow did have an algorithm or simulation that could predict how game-balance would change based on tweaking numbers on abilities, balancing the game would be trivial without needing an AI model.

Saiytama

Indeed, but if it is possible to create and train an AI so that it can generate information based on numerous variables that we sometimes don't even consider, it would be very useful. I don't know how most of the market balances classes, but if we take games like POE and LOL as examples, they have countless build possibilities, but there are some that are simply unbalanced. I want to insist on this topic because Ashes of Creation is a game I am excited to play, and I wouldn't want 80% of the players to follow the same build pattern; that takes away the fun. Therefore, having good balance is something that allows us to play with what we really want without having to follow a specific path to be competitive. The AI part can even be removed; there's no problem with that, but I would like to help make the game as good as possible within the scope that we and the developers expect.

Depraved

Saiytama wrote: »

Yes, I am not saying that AI should create builds or do things like that, but we know that often some classes are more used because their potential in PvP or PvE is much greater than others. Even within a rock-paper-scissors balancing system, I still think it's important to use AI to optimize and assist the balancing team. I don't believe this will take away from the players' experience, since our role will not be to balance but to create builds and find paths to success within the diversities the game offers us.

In my opinion, all classes and races should have the capacity to reach a common potential, and from there, the rock-paper-scissors system would work very well. For example, everyone could reach 100 power, where this 100 represents various experiences, items, builds, buffs, and so on. But the 100 of rock would be superior to the 100 of scissors, for example, because of the compatibilities and counters from one class to another.

don't need ai to balance anything. why would I need a prediction of which numbers I need to adjust, especially when the prediction will be incorrect plenty of times, instead of knowing exactly what number I need to adjust?XD

Saiytama

It would serve to help with balancing, how this would be done would have to be discussed.

Kilion

It sounds very logical when one keeps the concept of "balancing" as general as just using the term. But what exactly would such balancing look like?
What does "balancing classes and races" mean exactly?

Should the goal be to nerf physical damage dealer classes like Rangers and Rouges when a server has a lot of magical damage classes like Mage, Cleric and Summoners, because those are weak to Rangers and Rouges as per the Rock-Paper-Scissors design?

Because to my knowledge it seems that the balancing of combat power is exactly that: a Rock-Paper-Scissors System as described in the Wiki (see Link)

The balancing within that system will be done during the Alpha 2 and the Beta Phase. And the balance is supposed to topple when certain roles on a server become more common than others. More Casters means Phyiscal Damage Dealers will become stronger, more Physical Damage Dealers mean Bruiser become stronger, more Bruisers mean Casters become stronger.

Saiytama

I fully agree that everyone should have a class they are strong against and another they are weak against, but I still think it is necessary and also innovative to use something like AI (it could be something else, we just need to think of something that helps the developers, I use AI as an example because it has evolved a lot and has many applications). So, this mode of balancing, even within the Rock-Paper-Scissors system, would still be welcome since being strong against class X does not mean being unbeatable.

Let's be honest, there will be many items, various classes, and races, so there are countless possibilities that could take many years and a very robust team to balance well. Therefore, the balancing model I am proposing is not about making everyone equal in strength (without counters), but rather ensuring that within the Rock-Paper-Scissors system, everyone has a chance to win even when fighting against a class that is strong against theirs (in Brazil, we say something like "having fingers," referring to the player's PvP skill).

Kilion

That is exactly what the data gathering during the Alpha is for. Intrepid has stated that they will collect a sh!tload of information on everything during the Alpha phase to see what works as intended and what does not.

Seeing how the world event system is dynamic, it can be assumed that there already is some sort of Virtual Intelligence in use to make that work and to me that sounds like they are already on it.

Smaashley

It would be useful for AI to only balance metrics and complexe equations that summarize in damage. For that, Intrepid should buy a quantum computer

Vegaron

I cant imagine it is that easy, though i know nothing about Ai programming

Input from Experienced Pvp Players might be better. Only Problem with that, is that in a Class-Locked Game, people are usually very biased in their class assessment due to Ego reasons, very vocal with the weaknesses of their own class and blind to their Strengths. But AoC putting so much on transparency and Player Feedback&Interaction, maybe itll be better

Ayeveegaming1

Smaashley wrote: »

It would be useful for AI to only balance metrics and complexe equations that summarize in damage. For that, Intrepid should buy a quantum computer

arkileo

This seems like one of those things that could be done, but would take so much time and effort to develop that it would be cheaper to do it the old fashioned way.

AI is good at objective things, like math. So if you wanted to equalize DPS output across all classes in ideal conditions, then yeah, AI could do that somewhat easily. But you could also do that by writing a simulator, which is much simpler than fiddling with AI.

It's the subjective parts of balance that I think AI would really fail at. Take this hypothetical for example:

Say the bard had an AOE silence in a 15ft radius for 1s. For starters, what do you balance that against? You could calculate how much outgoing damage was prevented, sure, but a 1s 15ft radius stun is going prevent just as much outgoing damage. Obviously a silence and stun aren't balanced 1 for 1. In a silence you can still move, but what's the value of moving? The AI needs a number to balance against, how do you quantify being able to move?
Maybe you can balance it against something like a tank's wall. The wall breaks ranged LoS so they can't attack until they reposition, and maybe that takes about 1s so it's effectively like a 1s silence, right? Let's pretend that's true for a second. People hate feeling helpless, and a silence makes you feel helpless. A wall is a wall, you go around it lol. Mechanically they both limit damage the same amount, but the silence feels worse. So the forums are spammed with posts about the bard silence being OP and how much everyone hates it, but the wall doesn't get a single comment. They're objectively balanced, but a human factor makes one feel like it's not balanced. How does an AI grapple with that?

Balancing is as much about numbers as it is about vibes. If something feels balanced, then it's balanced. I don't think an AI is going to understand vibes. Even if it could, AI needs to be trained, as others have said. I haven't checked, but I'm fairly certain there isn't a balancing dataset that exists, let alone one that takes into consideration player sentiment.

Caeryl

AI is dumb as hell and shouldn’t even touch the development process

It can’t think of anything, it can only regurgitate information