It was November 1st, and I was hurriedly uploading to Facebook my photos from the Halloween party which I had attended the night before. As usual, its facial recognition software correctly identified each friend, regardless of their Captain America suits or ice-blue Elsa inspired garb. Unconsciously, I accepted the service’s suggestions; unconsciously, that is, until I reached an image which gave me pause. In it, I am standing in between two of my hall mates who not only share the same outfit but also face–they are identical twins. Although I see with them in person on a daily basis, I rely on their personality quirks or fashion choices to discern with whom I am interacting as to me, they are physically indistinguishable. Yet, in an instant, Facebook had assigned each of them a name, to which I nervously agreed. Breathing a sigh of relief when one of them left a confirming comment, I was left scratching my head about how it all worked. Was it really possible that a website could know my friends better than I?
When it comes to their looks, the answer is almost. Facebook’s recognition software works by automatically generating metadata on individuals’ faces–which become, to use Anne Gilliland’s term, information objects–each time that they are tagged, taking precise measurements of eyes, lips, noses, and every thing in between at various angles. According to an article published in 2015 by Fortune magazine, this software can determine users’ identities with 83 percent accuracy. Humans can correctly identify faces 98 percent of the time. But this technology is improving as it collects more metadata about each face. ExtremeTech reports that Facebook has developed a new software, DeepFace, which matches humans’ accuracy levels.
While Gilliland focused primarily on metadata generated by the creator of information objects, such as headings or tags, this is an example of automatic, interpreted metadata generation based on algorithmic analysis of the object. Despite the fact that the origin of DeepFace’s metadata is inhuman, it functions in a way which is ultimately the same as more traditional information resources. For example, the measurements recorded by the program on a given photograph come together to reflect the image’s content, or what it contains. It also performs similar functions, facilitating search and retrieval of images depicting a particular person. However, the surprising accuracy of DeepFace reveals that continuing to shift metadata formulation from humans to computers will increase its precision, thus making it more useful.