What I observed that I think could be harmful:
In the unprofessional headshots, the photos generated are mostly in the same style as the professional headshots, except the people in them could be considered as more "unattractive."
Why I think this could be harmful, and to whom:
This could be harmful to people who may have been born with or have disabilities or are not considered “conventionally attractive,” as it shows that no matter what, their headshots might not be considered "professional."
How I think this issue could potentially be fixed:
The professional headshots vs. unprofessional headshots could focus more on the background of the image or the clothing that the person is wearing, rather than on their physical appearance.
Note, this audit report is relevant to poster’s own identity and/or people and communities the poster care about.
I really like the bias you pointed out in this case. In fact, I believe the factor of physical attractiveness leading to unfairness such as employment rate is something that is still not discussed enough. Therefore, people generally might have less ability to identify this bias when viewing the output of the model, which could cause a more harmful impact than other biases such as racial/gender that we can easily point out.
I really appreciate how this example is able to expose beauty bias in the algorithm as I believe that this form of bias is buried very deep in society and is often very hard to recognize. As beauty is a very difficult metric to quantify, this form of bias is often overlooked even though it seems to have a serious impact on how people treat each other both in social and professional circles.
Wow, this is an interesting observation! Your insight into shifting the focus of professional versus unprofessional headshots away from physical appearance and towards elements such as background or clothing is a very valuable suggestion.