Would this trading system work for D4? It combines free trade with BoA

General Discussion
21

Took a few secs to look at the Deep Mind ActiBlizz SC2 stuff. They have the source up on GitHub, a white paper. A few Barney articles (I love you, you love me). ActiBlizz looks like they were after investor buzz with the AI machine learning keywords. Like a talking head on Investor news “ActiBlizz involved in AI and machine learning, news at eleven”.

With the SC2 AI project, as a joint project, looks like they started w/ Deep Mind staff around 17 and Blizzard staff was around 7 - that’s a total of 24 people some of them were part-time. First announcement of that SC2LE stuff was 2016 with nice fruit on the tree in 2019.

With SW metrics, whether it’s LOC or MH, the “applied” project doesn’t seem nearly as big as a AAA title like D4. D3 ROS has a credit screen that scrolls for 20+ minutes - hundreds of people involved. AAA titles easily cost $100-300 million to develop, they’re on console, their I18N, tons of contractors, voice acting, music, graphic artists, developers, QA, and so on.

I’m not saying that these AI projects aren’t impressive but just in traditional terms of cost metric (either ttl man hours, lines of code, or cost etc) - they aren’t looking to exceed a AAA like D4.

My main point was that AI projects were going to be a small focused team of experts. That seems to be born out with the articles, white paper, and interviews of the project etc.

Humerously, ActiBlizz has filed a patent for maximizing microtransaction revenue through matchmaking changes using AI. How fitting is that - corporate greed over game innovation. Of course. We should have known. They’re also using AI for moderation of Overwatch language, lol.

Maybe we don’t need AI for coming up with an accurate item ranking system? It might be overkill for just ranking.

Small sidenote on Gates was that Microsoft and OpenAI (the guys that did the AI with DOTA 2) have a partnership. So his tweet was kind of shameless self-promo a bit.

Anyway, I exceeded my 10 minutes of googling so I have to stop before I self-destruct.

Couple links:
SCII Deepmind paper -
https://tinyurl.com/rn9y7875

Source code (LOC is very small but there are additional sources)
https://github.com/deepmind/pysc2

(D3 ROS credits take 20+ minutes just to scroll by):
https://www.youtube.com/watch?v=VwKRCq0kXFU

https://digital.hbs.edu/platform-rctom/submission/activision-blizzard-atvi-using-machine-learning-for-videogames-development-and-community-moderation/

https://www.pymnts.com/news/merchant-innovation/2017/activision-machine-learning-algorithm-makes-videogamers-spend-more/

2016 announce:
https://deepmind.com/blog/announcements/deepmind-and-blizzard-release-starcraft-ii-ai-research-environment

2019 fruit on the tree:
https://deepmind.com/blog/article/AlphaStar-Grandmaster-level-in-StarCraft-II-using-multi-agent-reinforcement-learning

1 Like
22

Keep in mind that looking for specific legendaries in D4 doesn’t guarantee it will have the legendary affix you want. Legendary affixes won’t be tied to specific legendaries.

23

It would still be worth it since they could use it in other products too. I am not an AI expert, but it seems most of the developed stuff could be applied elsewhere too:
https://old.reddit.com/r/Diablo/comments/md74g6/solving_some_of_d3_and_d4s_item_and_progression/

24

From that reddit post:

And think of this scenario - soon after D3’s launch in 2012, wizards crushed the game using an exploit with Energy Armor / Force Armor, really low vit life total, and super-high life-per-second. Or something along those lines. Incoming attacks over 35% of max Life were reduced to only deal 35% like normal but the LPS instantly refilled health / red globe back to max. Wizards were instantly at max health all the time - their low life with high LPS was a problem the devs didn’t foresee. So wizards were almost invincible before it was patched to include extra wording where being hit with dmg equaling 100% of max life was lethal.

Coincidentally a good example of what the balance AI cant do. It could try to adjust the 35% number, maybe increase/lower internal cooldown on how often an effect can trigger, but it cant come up with something new in an attempt to balance things.
Easier/faster balancing? Maybe. But also lower quality balancing.
It is easy to see through data analysis if something is over/under-performing. It is a lot harder to see why. And even harder to see good ways to fix it.

Merely using data to tell the devs where they should focus their balancing effort is hardly something you need an AI/machine learning for. But yeah, Blizzard definitely should use player data to find imbalances. I assume they are, but they could probably always do much more.

In D4, they could even do this in real-time in-game with data coming into the ML AI system from player’s beating objectives, bosses, leveling up etc. The system could constrain progress to a chosen rate by changing the power level on various gearing on the fly. That gear would be chosen by comparative data - certain gear and skills, especially early, can be really OP. The ML AI system, being fed data, would change power levels in near-real-time to, again, keep progression rates in check. They could also have a manual step where someone on balance team has to click a button to approve ML AI changes about to take place etc. They could slow changes down to occur daily at midnight or 3AM or something or keep it in near-real-time once they’re more confident with it.

Honestly, if you change the rules of a game on-the-fly, or every night, then you no longer have a game. Since you no longer have anything that can meaningfully be called rules.
Blizzard also learned that lesson when it comes to hotfixes in WoW for example. Being cautious about too many or too severe balance changes too often.
Heck, a very recent post from them:

When a specialization is overperforming in raid and dungeon content, we prefer to wait to rein them in until the next major patch that brings with it all-new challenges, different gear, and other changes that shake up balance. This is because even if we’re improving overall balance, mid-tier nerfs can cause entire groups to potentially struggle with raid or dungeon content they could previously complete without issue.

1 Like
25

No. Actually you enforce the rules with the AI - intended progression speed or whatever.

26

As your reddit link said, it keeps changing the rules.

“Intended progression speed” is not a rule. That is a goal.
Nothing wrong, at all, with changing the rules of a game so it adheres to your vision/goal of the game.
But if the rules are in constant flux, then they are not rules.

If the players cant be sure that the items they equipped yesterday, works today… well, take a guess how many would keep playing.

27

Am I reading that wrong or is that playing gamble with a trade system? I prefer unidentified item trade over this system, really. How can you expect everything to work with pseudo randomization? That’s also completely linear power progress.

28

As I already wrote you before in the solo/group thread this is something I’d consider a rule. It would make the game far more engaging since players won’t feel punished in certain situations, just as they won’t be able to exploit the system.

There’s no problem with this if the margins are already defined by the rules before the Season start.

If we take the current D3 FB case as an example and we apply to it simple margin rules regarding power of 2, 4, 6 pieces, then if anything goes off the borders the AI would adjust it on the fly. It may turn out that you get a buff.

Items would always work. Their power is what is to change.

A player that was smart enough to exploit some mechanic would simply have to switch to something else if his nerfed build is not top tier anymore.

29

Honestly it sounds like a boring game I don’t want to play.

It sounds like a RPG about character building where none of my build choices ultimately matter, because the AI will always ensure that the same result is achieved.

Essentially being able to find a build that works for you and exploit the system is part of the fun of the game that is centered around building characters. Balance is important but it doesn’t need to be too tight and it definitely doesn’t need to be attempted in real time.

When the developer tries to take too much control over every last aspect of gameplay, you end up with a game that isn’t as much fun to play.

1 Like
30

Yes, that’s why there are the margins - how much each item’s power should extend.

That’s what real choice is - when every decision has the same EV. The Math stuff that we know from aRPGs aka I equip sword A instead of B since A has better stats than B and I feel good about it is simply an “illusion of choice”.

Note that I am not saying we go to first extreme (same EV) always, I am saying we adopt both.

Exploiting simply kills the intended progression aka such players would enjoy more a P2W server where one can buy all the stuff at day 1.

31

That’s an illusion of choice.

I don’t feel good about it because I’m not rewarded for being clever. My build will just be almost instantly nerfed by the AI if I figure out something that works well.

There is little point in trying to theorycraft or try new builds in that system. There is no point to engaging with it at all.

You basically just made a grinding simulator, and I already play World of Warcraft which Classic is bound to have more interesting class fantasy than your game.

If players finding a highly effective build in your game kills the intended progression that badly, your game wasn’t very well designed in the first place.

and an AI isn’t going to save you.

32

Not to mention the effect you see in WoW when it comes to how builds perform. The effect probably has a clever name…
But, if Build A is actually 10% stronger than build B, and the community perceives build A to be strongest, all the players who care about their raiding performance switches to build A. Some players, who dont really care, still use build B. Unsurprisingly there also is a skill lvl difference between people who care about their raiding and those who do not.
So when you look at the raid data, build A will perform much much better than the actual 10% power difference it had over build B. Due to the skill difference, and of course also gear difference etc.
Now, you can try to weight the data for all that of course. Removing the effect of gear should be fairly easy. Removing the effect of how skilled players are, quite a bit more difficult.
But if the AI is not very good, it could certainly end up seeing “popularity” as an indication of imbalance, and overcompensate in rebalancing. Something human devs also do again and again.

Besides, those completely unbalanced builds hardly needs an AI to be found. Those should be easy to find either through simulation data, or actual player data.
It isnt like Blizzard would nerf Rat runs if only an AI could tell them how absurdly imbalanced it is. They know. They apparently just dont care.

1 Like
33

If all players select the more powerful item, that’s not being clever, that’s an illusion of being clever. And if you make your game too hard in terms of Math, you simply push normal players away from it or force them to copy/paste.

The point is you’d be free to play the way you like it aka real choice, not a Math simulator.

As I said, going to extremes is not the way - we should adopt both options and mix these so that players like you still have the theorycraft aspect, and others still have the real freedom aspect.

34

One small issue would be ladders - someone might get a top spot with a powerful combo. Then that combo would be nerf’d and the top spot would be hard to beat w/ a post-nerf weaker power level. They would have to color-code or note the itemization version on the side of each ladder spot or something. I think this is why Blizzard does seasonal balancing - they balance in PTR and a bit after but when season starts, it’s locked in. They don’t have to wipe ladders mid-season. It’s still a very interesting and powerful idea though.

ALso trying to think of games where they tie world events, God-like beings, and magic systems, etc to the items themselves in a dynamic way.

Like with a D&D type of system, your character’s alignment of Lawful Good or Chaotic Evil will have game consequences - a Lawful Good character couldn’t use a powerful evil weapon for example. Or an evil alignment couldn’t use a Holy weapon etc.

If you took that a step further with the AI system - you could make a certain class of items derive their power from essences or souls of specific beings or magical forces. For example, I might have a sword imbued with “Lord of Lies” essences.

If they had a World Event (as they intend in D4) where the “Lord of Lies” has ascended in power - then all weapons imbued with “Lord of Lies” essences would have more power during that Event. It would incentivise specific farming. Seasonal themes and balancing might actually be lore-based as well - power levels changed based on changes in the world.

The AI system would be really good at that kind of thing and this would address the “boring” factor that some have made - in fact, it could be more interesting than the non-AI system because it frees the devs to do all kinds of things. There might be dozens of factors that influence the power of your items and how they work. Theorycrafting might thrive with so many factors.

It would be more like a “living breathing” itemization system that changes based on the World and events around it. And perhaps only something powerful like AI would let them pull that off.

35

If it boils down to item A is simply more powerful than item B, then your itemization is designed poorly assuming we’re talking about what should be equal items.

Items should naturally be stronger or weaker depending on your build and playstyle. If I build a character that has a bunch of benefits when I crit, it shouldn’t come as a surprise when the crit item is better for me.

So long as things are designed well though, there should be no “Everybody use Stone of Jordan. SoJ is the best ring ever”.

Choice that wouldn’t matter because all it would determine is the spell effects flying across my screen.

Math also has a harder time accounting for utility. Characters shouldn’t be able to be reduced to simply their damage dealt/taken.

What your describing sounds like a game where I would not care to experiment or try new things because it would ultimately feel pointless.

and if the core gameplay loop was not damn near the best thing I’d ever seen, I would quit in a heartbeat.

1 Like
36

Looking at the reddit post and the Blizz Starcraft Deepmind project, there are many eerie similarities.

This is part of the Blizzard doc:

The reddit post and comments talk about exactly the same thing but with Diablo 4. They would put APIs into D4 that the AI system would use for balancing and automated play testing. It’s quite interesting. With D4 they could create a Rogue character and give it certain weapons and have it literally play through the game, like a bot. They wouldn’t have to guess how it would do or have an intern playtest it. The system would do it.

Another thing is that the SC2 stuff is PvP oriented - it’s very complicated on the AI-side with self-learning, reinforcement learning and all that. The reddit post seems to distill that kind of thing with D4 and focus it on itemization and character scripting.

It’s almost like someone on the inside is trying to plant a seed on the outside.

37

I’d still be dubious about using an AI. Attempting to do real-time balancing really wouldn’t benefit the game and would likely actually hurt it in this case, and standard balancing can be done just with regular data analysis backed up by your devs actually playing the game.

Which the devs should be playing the game anyway, AI or no AI.

It feels a bit like a case of some people seeing something cool and decided they need to use it before asking if it’s something they should use.

Balance isn’t unimportant, but a perfectly balanced RPG is one that is bound to be boring and full of homogenization and if that’s not your goal then the AI isn’t really all that beneficial over standard balancing measures.

Using an AI to balance Starcraft would make a lot more sense, being that it’s a highly competitive game being used as an esport. You want player skill to be paramount in the game.

38

As long as perfectly balanced does not mean everything is equal, then a perfectly balanced RPG sounds good. In an RPG some combinations should certainly be stronger than others. I wouldn’t want an AI to be like “Uh, tried playing this build with 6 defensive skills and it was 99.523% weaker than most builds. Got to buff all these defensive skills with a billion reflect dmg!”. (and yeah, of course you could set restrictions to prevent no-attack builds from the balancing, but it is just an example)

RPGs aren’t even that different from RTS in that regard. Spamming nothing but marines or zerglings should not be a viable strategy in most cases, outside of very early rushes. Synergies matters as much as in RPGs.

Another thing AI balancing might fail at, is factoring in how difficult a build is to actually use.
If one build is completely passive, with pets killing everything while you are AFK, and another requires insane reaction speed and multi-tasking attention, well, a computer might are really good at both of those, but for us players, the difficult to play builds have to at minimum be a tiny bit better than the passive builds, to be viable. Not much better imo, but at least somewhat. Skill should be rewarded.
Which holds true for RTS as well.

39

It’s something we’ll see in the future for sure since it enriches the game tremendously.

Illusion of choice and standard dev’s balancing setting margins will still have their share, but AI will be the main driving factor.

Once the AI takes over you’d basically have solved the balancing and the trading due to players interacting with the in-game AI vendor. The devs team would then focus on the creation of interesting builds presenting the players with real choices instead of moving numbers in spreadsheets.

40

In the context of having an AI do it I think I would almost have to. Sure you could set some parameters like a 6 defensive skill build shouldn’t be on par, but then you run into stuff like AoE vs single target, elemental damage, focusing on a certain enemy type, etc.

and even then if every “viable” build basically progressed at the exact same rate I feel like you run into the problem I pointed out before where the player never really feels rewarded for being clever or coming up with new ideas.

Once you have one of the builds, there is basically no reason to ever try anything else. It basically puts us in a place where we have a more numerically balanced Shadowlands, where the main difference between classes is the order of the buttons you hit and the spell effects you see on screen.

As I’ve experienced with modern WoW, having the difference be a slight variance in gameplay just isn’t enough to keep the game interesting. I think RPGs do best when they’re specifically not perfectly balanced and the balance is there just to make sure nothing gets too insane or too weak.