It’s crazy the amount of lack of information about the power of machine learning in this context.
Enterprise Machine learning models have millions of parameters. World of Warcraft would have far less, even when incorporating PVE, PVP, dungeons, raids, open world, etc
You CAN balance it to an arbitrary degree of accuracy, that’s kind of the point of this technology.
It is a black box system where you don’t necessarily need to know how it all works but that it will output the desired results.
The tool CAN provide multiple solutions for the same result which allows the devs to pick which specific solution they choose.
For example, the tool might produce a solution where the spell power coefficient of
SW:P is 50% and vampiric touch is 25%, and an equivalent solution with SW:P = 10% and VT at 40%. This gives the devs the ability to say that we want VT to be thematically stronger than SW:P and choose that specific solution.
People really really really need to stop thinking that arbitrary balance is not possible in every form of content across the whole game because it literally is. Machine learning abstracts away all of the interactions and uses the data to compute the intended result. It’s kind of the whole point of the technology
Also, having arbitrarily balanced classes does NOT mean homogeneity. The point of the ML model is to eliminate human inability to measure compounding interacting systems (talents, gearing, rotation, borrowed power, etc) and still find solutions that fit these parameters.
It would actually give the devs more freedom to get creative because then they don’t have to try to anticipate all of these effects because they just have to press f5 on the computer to run the code and find the balance