Another case where AI does things we can’t explain: apparently male and female retinas are different enough that a computer can guess your gender just by looking at your eyes.
In May of 2021, researchers reported in Nature a machine learning model that can determine a person’s gender, with an 85% success rate, based solely on a retinal image.
Because deep learning neural networks are computational black boxes, no one (yet) knows how the model does this trick. One clue: the model performs significantly worse in the presence of injury or disease of the fovea, perhaps implicating that part of the retina as gender specific.
This is the kind of problem the growing field of explainable AI is meant to address. I find it ironic that we’re building subfields of computer science to understand the mysterious inner workings of software we didn’t write.
By the way, did you know this:
The retina is the only tissue in the body where neural and vascular tissue can be visualized simultaneously in a non-invasive manner.