Researchers' Homosexual-Men Detection Model Denounced as a Threat to LGBTQ People’s Safety and Privacy

Stanford Graduate School of Business professor Michal Kosinski and co-author Yilun Wang developed an AI system designed to demonstrate privacy risks by detecting sexual orientation from facial photographs. The researchers scraped over 75,000 dating profiles from the United States, processing 300,000 images down to 35,000 that met their criteria. All subjects were white due to insufficient data for statistical validity among minorities. Using a deep neural network and facial analysis software, the system achieved 81% accuracy in identifying gay men and 71% accuracy for women when given pairs of photos. When provided with five photos per person, accuracy increased to 91% for men and 83% for women. The study was intended to raise awareness about privacy vulnerabilities, particularly given the potential for misuse by governments in countries where homosexuality is illegal. The research faced fierce backlash from LGBTQ advocacy groups who called it 'junk science' that threatens safety and privacy. Critics argued it represented a revival of discredited physiognomy theories. Kosinski received death threats and required campus police intervention. The study was based on the prenatal hormone theory of sexual orientation, which many experts disputed as inadequately supported. The work highlighted concerns about facial analysis technology being misused for surveillance and discrimination purposes.