To your point, one of the things I worry about with apple news, for example, is that the underlying ai won’t understand _why_ I dislike a particular news story. Generally I downvote hyperbole, clickbait, and paparazzi content, but I would guess it’s more common for people to downvote things based on the substance of a story, e.g. downvoting a well-written, truthful article because the truth in that case is inconvenient for their political stance, or because it’s about a loathsome-yet-newsworthy person. A facial recognition system would have to know the difference between my face being frustrated at a high-quality story about the fires in the western states vs my frustration that apple news seems to find Andy Borowitz to be hilarious and relevant. Maybe possible, but lots of nuance under the surface.
