Such systems always live in a Star Trek kind of universe, where they simply await orders from humans, yet know vastly more than them. It's probably about our sense of goal: machines help achieving a goal, but not set it. While I personally also like to have control, I'm sure there will always be a Musk or Kurzweil around to advocate machine autonomy, and they will implement it no matter what. That's why such interfaces are unrealistic.

For a long time that was my approach to LoB backoffice applications: make the backend clever enough that there is no need for non-trivial user interfaces (in the sense of being non-trivial to implement). That in part works well but on the other hand there are situations when the user knows more, uses the thing every day and thus making something something that looks like sci-fi UI (or like something straight out of mid-90's OS HIG, with the difference being mostly about color choice) makes sense.