*** The Official MAGAposting thread ***

[MadMojoMonkey]

The facial recognition software isn't categorizing things like a human would. It's examining the color values of groups of pixels using various filters and algorithms. It's not comparing nose size and distance between the eyes, it's comparing the colors and patterns of pixels, which is a subtle difference that bears stating. It doesn't know what or whom is gay until it is fed a sample set of pictures and told which is which to begin with. Once that process is seeded, you can start to ask it which of those 2 piles it would sort a new picture into. Often it's not too good at first, so you seed it more and more until it starts to show reliable results.

There is always variability in the seeding process, which can be hard to quantify. The sample set of pictures may well need to be all taken with identical lighting and camera work in order for the program to function... at first... but the advances in AI are coming so fast that it's not clear if this will continue to be necessary... just seed it more and more pictures with different lighting and telling it which pile to sort those pictures into. Eventually, hopefully, it will be able to sort the pictures no matter what the lighting.

Who knows if this is a fluke or if it can be improved upon, though. If we told it to determine the height of a person based on a photo of their face, maybe it can and maybe it can't. I'd guess the 2 are unrelated, but that could mean the relation is beneath my threshold of sensitivity to relate them. Perhaps if I was more keenly observant of someone's pore-spacing, I'd realize that it's directly correlated to height.

[oskar]

The facial recognition software isn't categorizing things like a human would. It's examining the color values of groups of pixels using various filters and algorithms. It's not comparing nose size and distance between the eyes, it's comparing the colors and patterns of pixels, which is a subtle difference that bears stating. It doesn't know what or whom is gay until it is fed a sample set of pictures and told which is which to begin with. Once that process is seeded, you can start to ask it which of those 2 piles it would sort a new picture into. Often it's not too good at first, so you seed it more and more until it starts to show reliable results.

Not sure if I'm misreading you, but you can definitely teach it what an eye, a mouth, a nose... looks like and then let it tell you the relative distance. Smartphones and webcams can already do that. Facebook does it (or did it) to auto-tag people in pictures.
I wouldn't mind reading the paper if it's available anywhere, but unless this is verified by independent researches I'm not giving it too much credibility.