Inferring Someone's Personality Traits From Just Their Photograph
hubie writes:
A recent paper in the Proceedings of the National Academy of Sciences (PNAS) claims to have developed a machine learning model that can infer over 30 personality or psychological traits of a person from simply looking at a picture of them. They used deep generative image models to create photorealistic pictures of different faces and combined that with over one million judgements to infer physical traits such as age and happiness, but also personality traits such as trustworthiness, smart, liberal/conservative, Middle-Eastern, gay, and dorky.
One of the authors (Joshua Peterson) announced the paper in a Twitter thread. He noted:
Note that we study attribute *inferences* (impressions), which have no necessary correspondence to the actual identities, attitudes, or competencies of people whom the images resemble or depict. Put another way, our dataset not only contains bias, it deliberately reflects it.
He also pointed out that they can use their model to manipulate images by trait, so one could take a photo and increase its perceived trustworthiness without changing any of the other features, and he invites people to upload their own photos for a demonstration.
I'm sure my fellow Soylentils will agree that this kind of research will never be used out of context nor exploited in any untoward manner (oh, by the way, one of the traits is "electability").
I wonder how this relates to Resting Bitch Face?
Journal Reference:
Joshua C. Peterson, Stefan Uddenberg, Thomas L. Griffiths, Alexander Todorov, and Jordan W. Suchow,Deep models of superficial face judgments [open], PNAS, 119, 117, 2022.
DOI: 10.1073/pnas.2115228119
Read more of this story at SoylentNews.