Diablo 3

Solving some of D3 and D4’s item and progression problems using machine learning and AI.

diablo4 - Solving some of D3 and D4's item and progression problems using machine learning and AI.
Loading...

First, a concrete example using the hot topic of D3's recent FB nerf. Deathwish has such a big multiplier it's been an overall balance problem for a while because it can be used with so many builds. If you like DW or don't main Wiz or know about DW, pls bear with me, it's more about the general ideas. So DW is just a basic dmg buff effect – it doesn't add a new animation or special effect, it's just a channeling damage multiplier that could be anywhere on gear and we know how they are with buff multipliers going willy-nilly to just about any slot.

But with changes to their code, they could change DW specifically in conjunction with FB for example. The sheet dmg multipler of Deathwish could quite literally change when paired with FB6. You would see a lower XXX% on DW when slotted with FB6 but when it's either cubed or slotted with any other builds, LOD, and other sets, where it's perhaps considered fine as-is, it would remain at that higher percentage.

This kind of per-combo dynamic power is nothing new to ARPGs and they could support it with lore and story if they used a system like that in general for all items. Various combos could be on the rise or decline with power levels based on the rise and decline of either the maker of that item, the spirits or entities that imbue them, the elements their made from, etc etc. The usual deus ex machina lore. Or, as they do know, just change it with no explanation other than "for balancing".

So in that specific FB example, they just knock down DW's multiplier when combined with FB6 and they're much closer to a solution without affecting other sets or builds that use DW – it's dynamic.

This dynamic system in general would allow them to balance so much better because so many of their problems stem from changing one item or set to fix one scenario but that change ripples through to so many other items and sets, throwing them out of balance. This happens with many classes, items and sets.

With a dynamic system where effects and power levels change based on per-case basis, they don't have that ripple problem. And players are already used to seeing the main sheet dps and toughness swing wildly w/ certain buffs and gear. The example of having DW buff multiplier change dynamically isn't an issue.

(Yes, I know, not going to happen, too much work, they've never done it before, game's in maintenance mode. Thanks for playing.)

Another solution, actually very doable, is just to use any one of the quality off-the-shelf or Cloud-based ML AI packages to auto-balance everything when changes are made. It has an upfront cost of shlepping in all the data, constraints, goals for the system, and all that but once it's setup, it ripples everything for you – for example, you constrain XXX to remain the same while setting end-goal of YYY to change, it does all the hard work of rippling the change correctly. It may even be overkill where classic modeling tools might suffice but this brings us to this next point.

Загрузка...

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. And again, they could back this up with lore so that it fits themes in the game. 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.

I would even go so far as to say, that an ML AI system could actually auto-balance every class and build in the game if there was no inherently contradictory itemization.

It could also,in D3's case, guarantee that every class and build could hit around the same GRXXX solo, and given the right itemization from the start, could also do the same for the various group comps, where permutational complexity (various class and skill interactions) is higher.

And think of this scenario – early in D3V, 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 on those lines, don't hold me to exacts here. Anyway, incoming attacks over 35% of max Life were reduced to deal 35% like normal but the LPS instantly refilled health / red globe back to max 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 or something.

Anyway – the ML AI system would have picked up on that progression disparity very early, in real-time relative to other builds and players. The balance team would have been on it much earlier with a patch issued sooner. The system is only as good as the data, skill design, and itemization allow it to be. But with ML AI, especially with systems geared for simulation iterations (like chess for example) these systems can quite literally play themselves to quickly find successful or weak in-game tactics, and weed them out or prioritize them.

With D3 or D4, and Blizzard's resources behind it, this kind of system could iterate millions of times through the game with simulations of every class, every build, every gearing combination, every group comp, etc. It could quickly find exploits and weaknesses. Modeling and spreadsheets aren't enough. And doing this before launch with D3 as example, would have found that FA/EA exploit. The cost for these systems is almost all up-front.

If you're unfamiliar with this kind of tech think of it like the ultimate bot in the specific case above – just running through the game with different classes and builds etc to find issues.

So TL-DR – dynamic per-case gearing power levels and ML AI in conjunction with in-game instrumentation could relieve our beleaguered friends from so many difficult problems. And if they wanted to intro a new class set and have it be above par and slightly more powerful – they could do this in a precise way to incentivise community without guesswork and a bunch of overbuffs or overnerfs.

Source: Original link


Loading...
© Post "Solving some of D3 and D4’s item and progression problems using machine learning and AI." for game Diablo 3.


Top 10 Most Anticipated Video Games of 2020

2020 will have something to satisfy classic and modern gamers alike. To be eligible for the list, the game must be confirmed for 2020, or there should be good reason to expect its release in that year. Therefore, upcoming games with a mere announcement and no discernible release date will not be included.

Top 15 NEW Games of 2020 [FIRST HALF]

2020 has a ton to look forward to...in the video gaming world. Here are fifteen games we're looking forward to in the first half of 2020.

You Might Also Like

Leave a Reply

Your email address will not be published. Required fields are marked *