- cross-posted to:
- [email protected]
You must log in or register to comment.
Your title here isn’t technically wrong, but it is misleading. It should be "white AI faces judged to be more real than white human faces. The way you phased it insinuates that white AI faces are judged to be more real than a diverse mix of human faces.
For anyone who doesn’t want to read the paper, they basically took an 60 white men and 60 white women, and showed them a whole bunch of white faces, half of which were generated by AI. It turns out that AI faces were rated as more human-like than actual humans, and they had some hypothesis why. Principally that AI, by its nature, generates images close to “average”, while real people tend to have features that are not “average”. The reason the study focused on white people is that most AI have been trained on white faces, so AI tends to do better with white faces.
Thats not AI so much as photography. You can’t run a study with wildly differing variables like that. They all have to have the same pose, background, color palette, focus, etc.
Aren’t these models trained and evaluated on making pictures of humans that look real? And ones that don’t look real are penalized.
So this seems to mean the training method is successful.
Nobody wants an AI that makes unbelievable pictures of humans…
Right? Like, I felt like I was missing the punchline here.
Ok but a lot the example images shown here that people are saying are AI images have clearly been edited. People are basically saying something looks off with the faces.
The other people just happen to have a feature that is in a slightly unusual place.
To me this seems like people conflate something feeling off with the image being AI generated
Ars Technica’s take is probably more pertinent to the title. Including the little mini-quiz at the top.
“AI’s idea of what constitutes a human face is the kind of face most frequently considered human by AI”
That human female 84% is pretty cool looking.
Confirmed: white people aren’t real.
Clearly the technology doesn’t work, and probably won’t ever work. I certainly can’t tell the difference between the faces, so how could a AI model hope to do anything differently?
That is not what this is about. This is about if humans perceived actual human faces more human than AI generated faces. The result was that humans perceived AI generated faces as human more often than they perceived actual human faces as human. So clearly the technology does work.
The technology works the same way at everything. It averages and smooths out the rough edges into a homogrnous lump.
