It's a gross thing to consider but I wonder if the filter lightened the skin of black female applicants. These results are disappointing but unfortunately not surprising.
At the risk of sounding conceited, when it gave me (black male) a bad score, I knew something had to be wrong. I went and searched for photos of famous black males generally accepted as attractive and noticed it also gave them poor scores.
That gave me a hunch about what was going on, and lo and behold: taking the same exact photos, but processed to black and white and brightened, resulted in extremely high scores.
The model was essentially measuring how light your skin is.
So if the model was trained on who folks wanted to date (as a measure of attractiveness) the model could still be somewhat accurate, even though not accurately measuring attractiveness.
How would models work on rating who would be a good NBA player / runner etc. I wonder if just being white would get me downrated.
Terrible pictures don't make you more attractive on dating sites...
It's just a bad model. There's bound to be bias in the model but ML 101 would be randomize saturation and hue on your training set
That said, I wouldn't jump to conclusions on how things like saturation or desaturation play into ratings, a lot of instagram filters for example have an a desaturation effect which makes me wonder if folks like that as a look. I don't personally.
Very well could be a bad model, but sometimes there are underlying factors that result in these bad models from the training set. And no, it's no ML 101 to randomize elements of your input in the training set, though if you want a broadly applicable model you'd look to have inputs that include a variety of hue and saturation (perhaps that's what you meant - I have seen folks though manually start doing things like randomize hue in the set, where the ratings were on the original image).
https://www.e-ir.info/2020/04/11/critical-reflections-on-eth...
Well, light skin preference for females persists to this day in African populations and every other population.
It was also present -- and documented even in Africa -- before any European presence existed, so it's strange to say that Europeans imported it. And darker skin is usually preferred for men.
Do you have a source for that? I'm not disagreeing, only curious.
https://evoandproud.blogspot.com/2006/12/skin-color-preferen...
https://evoandproud.blogspot.com/2013/08/perception-of-skin-...
Another observation he's made is that black children are born light and darken, with a noticeable sex difference:
- Black males darken until eventually they reach adult darkness.
- Black females darken until puberty, at which point they are darker than they will be as adults. One of the changes they experience during puberty is a lightening of their skin.
Apparently still an issue. When two friends of mine, a black woman and a white man, got married, they had to find a specific photographer who specialized in simultaneously photographing black and white people.