ImageNet Response

When I was using ImageNet, while it was funny to see the results, I felt very uneasy about this being what is actually used in the real world to label us. A lot of times I noticed the AI was very superficial; it labeled based on outfit or race or gender. There is enough problems already known with that (it labeled me foolish multiple times, as well as prophetess and heiress). However, I noticed that it also, many times, was also just completely wrong. It labeled me mestiza - a woman of mixed racial ancestry (European and Native American). It also labeled my friend batman and another friend a houseworker (on a picture where she is in a gown and tiara). It also could not differentiate between a piece of artwork and a real person, or recognize filters on my face. It would take two nearly identical photos or two headshots and give different labels. Where are these details even being pulled from? How are we being viewed out in the real world using this labeling system?