I feel likely people aren't imagining with enough cyberpunk dystopian enthusiasm. Can't an AI be made that doesn't inherently like people? Wouldn't it be possible to make an AI that likes some people and not others? Maybe even make AIs that are inclined to liking certain traits, but which don't do so automatically so it must still be convinced?

At some point we have an AI which could choose not to like people, but would value different traits than normal humans. For example an AI that doesn't value appearance at all and instead values unique obsessions as being comparable to how the standard human values attractiveness.

It also wouldn't be so hard for a person to convince themselves that human "choice" isn't so free spirited as imagined, and instead is dependent upon specific factors no different than these unique trained AIs, except that the traits the AI values are traits that people generally find themselves not being valued by others for.