Mike, Regarding your comment on "I also don't know how the machine could separate authentic sad photos from poseur sad photos." The basic answer to this is that the researchers probably used a supervised learning machine algorithm. Essentially the photos are tagged by real people, and this is done at a large scale as in machine learning the machine gets better at the task with more experience.. So it's ability to detect 'authentic' vs. 'poseur' sad photos ultimately depended the response of the people who tagged the photos. thus, the machine is a reflection of the biases of the people it learned from....