AI (or probably more accurately, ML) is an efficiency technology. Surveillance is one thing it can make more efficient, sure.
And accuracy doesn't matter. If it's 99% and the powerful are whitelisted for human review, that's enough. That's enough to pull a mesh so fine over every human being on the planet, forever, that reaches deeply into them before they can even begin to form a personality.
The jump from individuals listening to individual phone calls as they were happening to what became possible in the decades after that might be smaller than the jump from that to the capability of transcribing anything anyone utters, even in their sleep, automatically, and adding it to a giant corpus of data to fish for suspects or assets or blackmail recipients in, in seconds. With then being able to act on those people as if they were a dataset (set a few flags, which systems or humans downstream heed without question, done).
It's a very grave threat, and I find it interesting that people decades ago who wrote about it called it that, but now that we're in it, we find it very easy to just shrug it off. But it's the same lethal threat to what it means to be human. Privacy is required for thought, without thought there's no humanity worth mentioning, QED.
Which I think makes it even harder to get people to understand how shit this timeline is, because everybody is dreaming about what it could do for them, not to them. And the people who have reason to believe to never be at the wrong end of these sticks, have no idea how it deforms them, either.
Humanity has proven its skill at dehumanizing humans in all sorts of ways, with all sorts of rhetorical and technological means, so yeah, sure, "AI" (which in practicality means "surveil everyone and get them to like/do what you want them to like and do", the rest are gimmicks and chaff) is "nothing new", but it's still like humans going from "always having had weapons" to "fucking nukes".