What’s Written on Your Face

Age divination is a neat party trick, but facial analytics, as it turns out, may have a more sober application.PHOTOGRAPH BY JUSTIN PUMFREY/GETTY

Earlier this week, as part of its annual Build Developer Conference, Microsoft unveiled a little experiment in machine learning: a Web site that can guess the age and gender of faces in a photograph. The creators of the demonstration had been hoping—“optimistically,” they wrote—to lure in fifty or so members of the public to help test the software. Instead, within hours, the site was struggling to remain online as it was inundated with thousands of visitors from all over the world (many of them, curiously enough, in Turkey). Facebook and Twitter were soon brimming with exclamations of amusement, mainly from middle-aged men identified as young women, and of offense, mainly from twentysomethings identified as postmenopausal. Not even the Mona Lisa was spared. She is thought to have been twenty-four years old when she sat for da Vinci, but oil paint can be terribly aging; Microsoft’s program put her at thirty.

Age divination is a neat party trick, but facial analytics, as it turns out, may have a more sober application. In a paper published last month in the journal Cell Research, a group of Chinese scientists demonstrated that facial features can indicate how quickly a person is aging better than any other biological marker, and far better than chronological age. The team collected 3-D photographs of the faces of more than three hundred adults of Han Chinese descent and scanned them for tells—features that changed predictably year after year. Noses and mouths, for example, tend to widen as people get older; the distance between the upper lip and the nose grows, whereas the distance between the eyes shrinks. After averaging this data, the scientists were able to identify how quickly an individual was aging with respect to her chronological peers. The differences were striking: the face ages of one group of forty-year-olds varied by as much as six years, with the divergence increasing over time. What’s more, after correlating their predictions of an individual’s physiological age with her blood biochemical profile, the researchers were able to show that their estimates provided a much better guide to over-all health than a date of birth.

“People age at different rates,” Jay Olshansky said. “You know this from attending your high-school reunion.” Olshansky is a professor at the School of Public Health at the University of Illinois at Chicago and a board member of the American Federation for Aging Research. Somewhat surprisingly, he noted, there is not yet a scientific consensus on how to gauge a person’s biological age. Various proxies have been proposed—levels of growth hormone or inflammatory markers, bone-density measurements—but none have proved entirely reliable. Olshansky suggests a simpler solution: perhaps people who look younger are aging more slowly than their peers. His idea isn’t wholly new—gerontologists have been studying the epidemiology of crow’s-feet and liver spots for years, and in 2004 a group of Danish scientists successfully associated looking old for one’s age with early death—but Olshansky is the first to develop a piece of software that connects the two. He worked with Karl Ricanek, a computer scientist and the director of the Face Aging Group at the University of North Carolina at Wilmington, to build a program that could scan the faces in photographs for signs of aging, then correlate face age with mortality.

Their Web site, Face My Age, launched last July. It does not try to guess at age and gender, like Microsoft’s experiment, but rather asks its users to supply that information and more (marital status, educational background, smoker or non-) and to upload photographs of their make-up-free, unsmiling faces. “We have a different specific model of aging for each gender of each ethnicity,” Ricanek told me. “For example—forgive me for this—women tend to age faster than men, because their skin is different.” The public has so far uploaded more than eight hundred thousand images, which, the two men say, allows their program to generate a reliable comparison for most people. “Now, I will be honest with you, we tick off about half of the people that submit a photo,” Olshansky said, with the smugness of a man whose face is seven years younger than it ought to be. “People whose face age comes in older than their chronological age, they all say, ‘It’s got to be a problem with your program.’ ” Still, the system can be fooled by plastic surgery or make-up. Last autumn, Olshansky ran a photograph of an eighty-three-year-old Joan Rivers through the program. She had a face age of fifty-seven. When Olshansky sent Rivers her result, he told me, “she said, ‘Thank you so much—this confirms the value of all that plastic surgery.’ She then died a couple of weeks later.”

So far, Ricanek and Olshansky’s technology has drawn interest from cosmetics manufacturers, life-insurance companies, and financial planners. But the question of who gets to collect and use information about consumers’ faces is a thorny one. As its algorithm gets refined, Face My Age is beginning to be able to see traces of mental conditions, too, including long-term depression and even early-onset dementia. On the one hand, Ricanek envisions a future in which a person’s bathroom mirror could make use of the technology to provide helpful nudges—reminders to apply sunscreen, hit the gym, or see a doctor. On the other, Olshansky notes that the algorithm’s ability to “pick up on certain red flags” of accelerated aging and its associated health effects could make an employer less likely to hire someone or an insurer less willing to underwrite long-term care.

The spectre of a future filled with face-age-based discrimination is alarming, particularly because there is not much that the speedy ager can do to stop it. “It’s not like you can transform somebody who is destined to die at sixty into a centenarian,” Olshansky said. “You have to have won the genetic lottery at birth.” Ricanek—who, as it happens, has an older face than his chronological age would suggest—takes a more sanguine view. He uses regular face-age analysis as a motivator to “order the chicken sandwich rather than the super-deluxe burger,” he told me, and actively encourages his students to look for crow’s-feet in a future spouse. Face age aside, he said, “early-onset crow’s-feet is an indicator of someone who smiles a lot. And that’s a good sign.”