JP
Sep 23 2026 - 10:43am 3 mins

In 2021, a professor of psychology at Stanford called Michal Kosinski downloaded one of the many basic open-source facial recognition software programs from the web and applied it to a massive set of images of people’s faces. Kosinski had collected clear, naturalistic images of nearly 1.1 million people from the US, UK, and Canada of all ethnicities: the largest sample, of 860,000, came from a US dating website, and the rest came from Facebook profile pictures. He trained this off-the-shelf facial recognition software on those images and asked it to predict the political orientation of the people in them by comparing them to another large database of faces that belonged to people whose politics were known.

Chance would give you a correct result 50% of the time. Humans, when they’re asked to say whether a face is conservative or liberal, do slightly better and get it right 55% of the time. The software Kosinski used was able to accurately guess the political orientation of the people represented 73% of the time. In statistical parlance, that represents a “large” effect size. Nor were the results country-specific: when the algorithm was trained on US faces, it could still distinguish between liberal and conservative faces in Canada with 68% accuracy, and in the UK with 67% accuracy. Some demographic markers are pretty obvious — in the US, white people, older people and males all lean slightly more conservative. But even when Kosinski excluded that data, the accuracy only dropped a fraction. Working off tiny visual cues like head pose (the roll, pitch and yaw of the head) or facial expressions such as surprise or happiness, the algorithm could still tell a person’s politics from their profile picture with 71% accuracy.

Our minds influence our faces. There are obvious ways in which our makeup choices, or our hairstyles, or our particular style of facial hair or jewelry also reflect our values. The assumptions we make about the politics or aesthetic preferences of someone in goth or emo makeup, or the clean-shaven young man with a perfect Fifties side-parting, are probably fairly accurate. But our faces also affect our minds.

“The younger Goodwin certainly cut a different figure in the public sphere from the one he cuts now.”
- Boyd Tonkin

In 2021, a professor of psychology at Stanford called Michal Kosinski downloaded one of the many basic open-source facial recognition software programs from the web and applied it to a massive set of images of people’s faces. Kosinski had collected clear, naturalistic images of nearly 1.1 million people from the US, UK, and Canada of all ethnicities: the largest sample, of 860,000, came from a US dating website, and the rest came from Facebook profile pictures. He trained this off-the-shelf facial recognition software on those images and asked it to predict the political orientation of the people in them by comparing them to another large database of faces that belonged to people whose politics were known.

Chance would give you a correct result 50% of the time. Humans, when they’re asked to say whether a face is conservative or liberal, do slightly better and get it right 55% of the time. The software Kosinski used was able to accurately guess the political orientation of the people represented 73% of the time. In statistical parlance, that represents a “large” effect size. Nor were the results country-specific: when the algorithm was trained on US faces, it could still distinguish between liberal and conservative faces in Canada with 68% accuracy, and in the UK with 67% accuracy. Some demographic markers are pretty obvious — in the US, white people, older people and males all lean slightly more conservative. But even when Kosinski excluded that data, the accuracy only dropped a fraction. Working off tiny visual cues like head pose (the roll, pitch and yaw of the head) or facial expressions such as surprise or happiness, the algorithm could still tell a person’s politics from their profile picture with 71% accuracy.

Our minds influence our faces. There are obvious ways in which our makeup choices, or our hairstyles, or our particular style of facial hair or jewelry also reflect our values. The assumptions we make about the politics or aesthetic preferences of someone in goth or emo makeup, or the clean-shaven young man with a perfect Fifties side-parting, are probably fairly accurate. But our faces also affect our minds.