Close-set eyes, thin mouth, sharp nose — AI sums your face for political affiliation
A researcher claims he can divine the politics of individuals using a facial recognition algorithm.
The biometric algorithm, applied by researcher Michał Kosiński, was used on naturalistic images of 1.1 million people, and reportedly was able to correctly identify 72 percent of liberal/conservative face pairs.
That compares to 55 percent accuracy for human guesses and 66 percent for assignments based on a 100-question assessment, according to Kosiński’s paper.
Images were gathered from Facebook and dating sites of people (making them self-selected) in the United States, Canada and the UK. “Potential cues” of political leanings included face expression and morphology, and self-presentation.
Just 347,000 images were of people judged to be non-white, which the author holds out as evidence that the findings “apply to other countries, cultures, and types of images.”
Although it is positioned as a warning about mass surveillance and the potential abuse of face biometrics, the report has been met with skepticism if not condemnation.
Superficial physical identification has been used to ostracize people throughout human history. As recently as the late 1970s, Cambodia’s bloodthirsty Khmer Rouge regime used the wearing of eyeglasses to mark tens of thousands for state murder.
Technology investment publication VentureBeat called the study “outlandish” in the headline of an article about the facial recognition research. The article made sure to note in the headline that Kosinski is associated with Stanford University (he is an associate professor at Stanford’s business school). It links Kosiński’s work to discredited pseudoscience like phrenology.
He claims to have authored the first media article that warned against Cambridge Analytica, the firm Ted Cruz used during his failed presidential campaign to get detailed psychological profiles of U.S. voters without their permission using seemingly innocent Facebook surveys.
Kosiński’s work on AI and mass persuasion is credited with underpinning Cambridge Analytica. He has made presentations to the political and economic leaders of Russia. He maintains that biometrics-based AI is capable of discerning IQ, emotions and even one’s sexuality.