"mega study"

  2646 distinct facial images of 969 individuals. Using deep learning, we find 82/349 personal attributes (23%) are predictable better than random from facial image pixels.

I don't think I need to elaborate on how ridiculous this is, but do want to add that this should be at the very top of the "irresponsible", unethical, not to mention incorrect uses of AI (deep learning) that people get concerned about.