An article in Nature reports on scientists who trained a deep-learning system with a large number of photographs of human retinas.  After training, the system was able to identify the sex of the people in the photos with >80% accuracy.  This is interesting because there is no known difference between male and female retinas.  Naturally the system can't tell us how it tells the difference.

via 52 Things I learned in 2022 via Kotke.org