I'm skeptical, not from any domain knowledge but from knowledge about spurious results like this in machine learning models. E.g. ML models have been able to pr

I'm skeptical, not from any domain knowledge but from knowledge about spurious results like this in machine learning models. E.g. ML models have been able to predict if someone is gay, or a criminal, etc, from an image, to some level of accuracy. But what they are doing is stereotyping, they are not actually applying some kind of causal logic. It could be equally simple here: seeing stereotype of a high testosterone looking guy, the women's bias more often than not could lead her to think he's up for sex. It doesnt have to be some innate sense, it could just be a simple stereotype. I couldn't read the actual paper, but I wonder what the accuracy was. I bet there are lots of misses. Anyway, maybe the paper covers all this, but one conclusion could be that you can be correct more often than not if you stereotype. Doesn't make it a fair basis for decision, but it's predictive power is better than random.