Physiognomy is a 19th-century pseudoscience. Why can’t we stop practicing it?
longreads.com
longreads.com
E.g. https://www.gsb.stanford.edu/faculty-research/publications/d...
I'm actually afraid the problem of physiognomy is not it being a pseudosceience but it being a severe threat to privacy and equality and a potential source of false positives (as personality is not a 100% product of a body but many people would argue).
The study on predicting sexuality wasn’t looking at physiognomy per se, but at photographs of people. One of the criticisms of the study was that instead of predicting sexuality based on facial features it showed differences in how people of different sexualities photograph themselves, eg. using camera angles more associated with how people with a specific gender photograph themselves.
Neither of these studies convince me that physiognomy isn’t just prejudiced guessing.
Edit: grammar.
From the article I linked: "Each student in the picture completed a Czech version of the Intelligence Structure Test that uses various types of tools to measure the different types of intelligence. The images were close-ups of the students’ faces, which featured a neutral, non-smiling expression, and did not wear jewelry or cosmetics, Business Insider reported.
The raters took their time rating each photograph for either intelligence or attractiveness. Of the raters, 43 women and 42 men judged photos for intelligence, and 42 women and 33 men judged them for attractiveness using a scale of 1 to 7 (1 being the highest score, 7 being the lowest possible score). The researchers then averaged the intelligence and attractiveness scores each student received.
The findings revealed both men and women were able to accurately evaluate the intelligence of men by just viewing the facial photographs. Men in the photos with a higher IQ were perceived as more intelligent much more than women in the photos who also had higher IQ scores. In both sexes, a narrower face with a thinner chin, and a larger, prolonged nose characterized the predicted stereotype of a higher IQ, while a rather oval and broader face with a massive chin and a smallish nose led to a prediction of low-intelligence."
Note the lines "Each student in the picture completed a Czech version of the Intelligence Structure Test" and " The findings revealed both men and women were able to accurately evaluate the intelligence of men by just viewing the facial photographs." So the study did actually measure peoples' IQs using a standardised IQ test, and compared the results of the tests to peoples' assesment of others' IQ based on their faces, showing a correspondence between the IQ people estimated based on faces and the actual IQ scores.
Here's a study that directly shows a correlation between sexuality and facial anatomy: https://www.researchgate.net/publication/270290521_Facial_St...
That paper is severely flawed, first of all in the way they trained their classifier to recognise peoples' sexuality from images of their faces.
What they did is they collected ~35k pictures of gay and straight men and women, all users of a dating site in the US. Half of them were gay and half were straight.
Right off the bat, that breaks the fundamental assumption in machine learning, that of distributional consistency. In short, machine learning works under the assumption that the distribution from which a training sample is drawn is the same as that of unseen data- i.e., of the real world. If the sample distribution is not the same as that of the unseen data, a model cannot be guaranteed to predict anything about unseen data.
But, in the Deep Gaydar paper's sample, the distribution of gay and straight men and women was uniform: 50% straight and 50% gay people, while in the real world the prevalence of homosexual preference is estimated to be around 10%- the Deep Gaydar paper itself cites a 7% prevalence.
So the classifier trained to identify peoples' sexuality from images of their faces, in the paper, is completely incapable of identifying anything about the sexualities of people outside its training sample- which is the majority of people on earth. Which invalidates their claim that it's possible to tell someones' sexuality from an image of their face. At best, what they can claim is that they trained a classifier to identify the sexuality of people in their training sample from images of their faces. But nothing more.
And no, that does not suggest that "physiognomy is not pseudoscience". Quite the contrary- sneaky data manipulation like that is the hallmark of crackpots and charlatans.
Um, what? That sort of came out of nowhere, and is then left hanging, as if every reader should obviously agree with it.
I mean it’s pretty clear that people with Down syndrome have a very unique look and this is not the only genetic disorder that has very defined an unique facial charetarisrics.
Wait, what?