> They're also free to circumvent the software and play the game without it [...] People shouldn't have to abstain from playing a game they bought because the game company doesn't have total access to their machine.
Imagine we're playing soccer. We both know the rules. However I don't really like how those rules restrict me from using my hands when I'm not a goalie. You could say that I feel like the rules shouldn't have total access to dictate what I can and can't do with my body. But you want to play soccer and the rules that go with it. Who should abstain from playing the game?
For the sake of this example, please don't suggest saying that the rules should be changed by the players to accommodate both. This isn't a possibility in video games for the players. They can either choose to play with the existing ruleset or not play at all.
If your suggestion is to just avoid players that don't want follow the rules, with the game providing this as an in-game mechanism, please be aware this doesn't really work and is very ripe for abuse. There was a case early on in Overwatch where they had a top player complaining that they were no longer getting into games successfully. An investigation revealed that they were so good at the character they were playing -- a sniper -- that their opponents didn't want to play against him/her and thus used the in-game avoidance feature to do so. This had the net result of the match-making system not finding enough players for him/her to play against anymore.
> They need to come up with better ways to detect cheaters. Methods that work on the server side.
But they are coming up with better ways to detect cheaters, and it's through these kernel drivers. Ultimately, differentiating between a regular player and a cheater is a matter of how many signals you can analyze. The quality of the model you can build for identifying cheaters increases as you add more signals. And with a good enough model you can reach a certainly level that allows you to comfortably start issuing bans. You can see it with the link you provided, where the researchers found a way to add more signals to the model by analyzing the network traffic patterns.
Unfortunately, server-side detection can only get you so far. If a player turns on a wallhack, what signal can the server use to figure out what's going on? If my LoL champion has a defensive ability that I can use to dodge attacks, and my cheat program can detect incoming attacks that'll hurt me and activate it for me, with it programmed to have a minimum time to activation in order to pretend the reaction time is human, what can the server do to detect this?
Ultimately, to answer these questions you have to start asking the client to assist you by providing more signals, but when the client is under user-control it's trivial to lie to it. Hence the need to move some of the anti-cheat code to driver-space.
By the way, server-side detection doesn't really account for the reality of how cheaters actually cheat. Fancy detection methods like dynamic Bayesian networks for statistically analyzing shooting accuracy in FPS's sound great in theory, but aimbots don't really snap to player's heads anymore. Instead they guide the player's aim when toggled on and only do this gently, so that even human reviewers have a tough time seeing what's going on. The intent there is to increase the uncertainty level of the model declaring the player a cheater, as companies fear banning players accidentally and will only issue bans once they are positively sure the player is a cheater.
I'm sure Riot and many other game companies are using as many server-side detection methods as they can, so what exactly do you expect the game companies to do when they run out of available server-side detection methods and still have a cheater problem to deal with that affects their bottom line?