Stanford AI sexual orientation prediction study

Occurred: September 2017

Can you improve this page?
Share your insights with us

Two researchers at Stanford University have published a paper apparently showing that AI can predict someone's sexual orientation from a few facial images, prompting accusations of junk science, physiognomy, and shoddy ethics.

Stanford Graduate School of Business researchers Michal Kosinski and Yilun Wang trained a neural network on almost 15,000 pictures of gay and straight people taken from a popular dating website.

They found that their AI could predict the sexual orientation of gay men 81% of the time, in contrast to a human man, who would be right 61% of the time, suggesting machines have a potentially better 'gaydar' than human beings. 'Gay men' they found 'had narrower jaws and longer noses, while lesbians had larger jaws.'

The study garnered criticism from LGBTQ groups, who criticised the study as 'dangerous and flawed … junk science' that could be used to out gay people and put them at risk. They also felt the study was too restricted by only using photos that people chose to put on their dating profiles, and by failing to test a more diverse pool.

Meantime, technology researchers and commentators focused more on the technical details of the analysis, with some figuring the neural networks are picking up on superficial cultural signs such as the use of make-up, eyeshadow and glasses, rather than analysing facial structure. Others highlighted what they saw as poor ethics, including scraping and using people's images without their consent.

Kosinski later claimed the research deliberately aimed to demonstrate the power of AI and how easily it can be abused and misused. 'I hope that someone will go and fail to replicate this study … I would be the happiest person in the world if I was wrong,' he told The Guardian.

Operator:
Developer: Michal Kosinski; Stanford University
Country: USA
Sector: Politics
Purpose: Predict sexual orientation
Technology: Facial analysis; Computer vision; Machine learning; Deep learning; Neural network
Issue: Accuracy/reliability; Ethics; Privacy; Pseudoscience
Transparency:
Black box

Research study


Research, audits, investigations, inquiries, litigation

News, commentary, analysis