In your face! Clever things science is doing with facial recognition
Facial recognition technology is booming with governments and corporations making big investments. Scientists, however, see other potential applications, both good and bad. Tim Wallace faces four of them.
1. Who art thou?
Is the lady with flowers, a marble bust by Florentine artist Andrea del Verrocchio circa 1475, the same woman as the lady in the window, a painting by Sandro Botticelli from about the same time? Such issues are of intense interest to art historians. The passage of time shrouds the identity of many once-famous subjects in still-famous artworks. The subjectivity of artistic licence, styles and conventions complicates even identifying the same faces in different artworks. To overcome this, art historian Conrad Rudolph, of the University of California, Riverside, has led a project called Faces, Art and Computerised Evaluation Systems – FACES, of course! – to test the value of machine analysis in identifying faces in portrait art. The project is ongoing but initial results have shown promise; and yes, computer-based analysis does support the view the two ladies, despite their distinctly different artistic personas, are one and the same.
2. Rare diagnoses
Taking a photo of a child’s face may one day be as normal a part of paediatric health care as the doctor listening to their lungs. Marius George Linguraru, of the Sheikh Zayed Institute for Pediatric Surgical Innovation in Washington DC, has led the development of digital facial analysis technology to diagnose genetic disorders in children. The technology could assure the earliest possible intervention for better medical care. It has proven highly accurate in identifying Down syndrome, the disorder caused by an extra chromosome 21, and the much rarer DiGeorge syndrome, caused by the deletion of a small segment of chromosome 22. While these conditions lead to characteristic facial deformities, even experienced clinicians can have trouble identifying them in young children, particularly from different ethnic backgrounds. Two other rare diseases affecting children, Williams syndrome and Noonan syndrome, are next in line for testing.
3. Love Match
Could a broad range of psychological traits and preferences be detectable in facial features? Two researchers from Stanford University believe so. To prove their case, Michal Kosinski and Yilun Wang used artificial intelligence to extract features from thousands of facial images posted on dating sites and created an algorithm to predict sexual orientation. They report the ability to correctly pick, from a single facial image, the preferences of men in 81% of cases, and of women in 71% of cases; even higher results (91% and 83% respectively) came from five images. The findings are controversial, with one of the stronger criticisms being that dating-site pictures may reflect stereotypes; concerns were not only false readings but correct readings being used to discriminate. The researchers admit to agonising about going public with their study but stand by it as a stark warning of the privacy threat from facial recognition algorithms able to detect intimate traits.
4. State of the artiface
Speculators and spies have long been interested in reading the thoughts of government officials. None are more scrutinised than those who set interest rates, as fortunes can be made by calling a rate change before its official announcement. Two Japanese researchers, Yoshiyuki Suimon and Daichi Isami, have used artificial intelligence to analyse half-second changes in the expressions of the Bank of Japan’s governor, Haruhiko Kuroda. Though Kuroda sought to remain inscrutably neutral at press conferences, the software identified fleeting signs interpreted as betraying his true underlying feelings. Those micro-expressions indicated pessimism about the economy, the researchers say, and preceded significant policy changes in line with that. Kuroda has since laughed off suggestions facial recognition could be used to reveal his innermost thoughts; Suimon and Isami have yet to reveal what artificial intelligence says about that.