Illustration by Andy Potts


Sure, everyone enjoys tagging their friends in pictures. But do you know what FACIAL-RECOGNITION SOFTWARE really means — or where it can lead?


When investigators on an episode of CSI: Crime Scene Investigation fed a photograph of a missing man into a computer, the system identified him in less time than it would take most of us to stammer “And what was your name again?” to a vaguely familiar-looking person who greeted us at a party like a long-lost friend. Of course, that’s TV. In the real world, facial-recognition technology still can’t really match a random picture against a world of individual countenances. But it will before long, and not just for criminal investigations. Retailers, financial institutions and other businesses are already installing systems that automatically identify people to some extent using just their faces. In some cases, this is to provide better individualized customer service. In others, it’s to provide more effective, targeted advertising.

“It’s here, and it’s happening faster than we might believe,” says Alessandro Acquisti, a Heinz College at Carnegie Mellon University associate professor who researches privacy implications of facial recognition. To find out what could be done with off-the-shelf facial-­recognition software and cloud-based computing power of the sort anyone can get, Acquisti and a team took photos of strangers from an online dating site that used pseudonyms to protect identities. They were able to match one out of 10 to photos from searchable Facebook profiles. A similar test using offline photos taken of students with a webcam identified one out of three faces.

Facial recognition works by analyzing a digital photo of a face and converting it into data points — distance between the eyes, skin color and the like. An algorithm compares these points against a database of existing scans to identify the person. Identification isn’t as ­accurate as DNA or fingerprints — people tend to look alike — but it’s much easier to snap photos than fingerprint someone or obtain DNA samples. And almost all of us already have digital photos on file at driver’s license bureaus, if not dating sites.

Before you start wearing a hoodie, cap and shades like a high-stakes poker player, consider that Acquisti’s experiment wasn’t quite real-world. The webcam photos were all of students at one university. That geographical restriction simplified the problem dramatically. “If we had tried to do it nationwide, we would have had to deal with databases of hundreds of millions, if not billions, of images,” Acquisti says. And that’s not feasible for existing technology.