Every three months for 5 years, he would send me some blurb that basically said "NAHWheatCracker is a great engineer. Sometimes he's difficult to work with." We would have a one-on-one to review the performance review. I would ask about the "difficult to work with" part.
My manager wouldn't say who said what. He wouldn't discuss circumstances. He wouldn't facilitate having a discussion with whomever for specifics. It could be an issue that came up once, a chronic problem, or a minor complaint on a bad day.
I wanted something specific to improve on. It wasn't hard to make assumptions, but assumptions aren't clear. Non-clarity drives greater wedges between people. I shut down in conversations lest someone say something to my manager.
The idea of it being made more dehumanizing by having an AI slop out a performance review seems even more depressing. That said, I think bad managers will ruin this process regardless of AI. You can't get much worse than being less than useless.