But they could easily change that by requiring everyone to submit their government ID. Look at Facebook's Real Name Policy [0], which was already not well received. Determining if the users are actual humans is easy, doing it without any loss of privacy is hard, because at in some part of the trust chain, a person has to provide proof to some trusted organization they are human.
Sure. But isn't what you are describing exactly the point of the sovereign/decentralized identity trilemma?
Anyway, the point is not even that Twitter needs to be able to identify its users. Twitter just needs to get rid of its bot problem. To solve that problem, they could simply charge their users $5 and promise to refund if after some vetting period (by their fraud detection and other users). The downside of this approach, of course, is that it would deter not just bots but also a large number of legitimate users who wouldn't be interested in paying to sign-up.
[0] https://en.wikipedia.org/wiki/Facebook_real-name_policy_cont...