As an anthropologist you are in theory studying humans from out of a shared cultural context and it is easy to argue that a machine which hasn't even the slight

As an anthropologist you are in theory studying humans from out of a shared cultural context and it is easy to argue that a machine which hasn't even the slightest semblance of a shared cultural context would not be capable of distinguishing a twitch from a wink with 100% reliability. Not all human gestures are universal either so even if you trained your AI to read faces it would likely be trained on a small subset of human cultures and would produce varying results. In regards to the greater question of knowing intention I'd agree that algorithms do a much poorer job of guessing my intention than me directly communicating it and I am annoyed that so much of our on-line communication is filtered by algorithmic social engagement that an encourages shallow interactions.