Op-Ed: Can Software Be Racist?

There’s a widely watched video on YouTube, more than a million times, that claims HP face-detection software cannot recognize black people. Let's take a look at the arguments.

In the YouTube video, a dark-skinned male and a white-skinned female Caucasian both try to engage the computer's face recognition software that’s supposed to make a camera follow their movements The software does it for the woman, it doesn’t succeed with the man. However, judging from my own experience, skin color is most likely not the point here, rather shape and seize of several facial features. Far from being a software programmer or recognition specialist, I put this information forward on the basis of my extensive work with the latest Picasa software Google freely offers to photographers. The software took several hours to scan through about four thousand images, and the results were astonishing. It picked out and identified faces from Europe, India, Borneo, Afghanistan, Indonesia and Thailand; people with all kinds of skin colors and facial shapes. And while the software rejected most faces of animals, it did think of a few cats as being humans, just as it thought certain statues of Indian deities and some faces on currency represented people. Looking at all of it very closely, it seems that such software looks for certain ratios, distance between the eyes, distance between eyes and mouth and chin. Actually, facial recognition has become so good by now, I find it frightening that it has already turned up in mobile phones and is being widely used by police in many countries. Without realizing it, we’re heading for Orwell’s Big Brother society way faster than I had previously imagined. Combine facial recognition with the fact that most people have mobiles and use credit cards, both of which are constantly logged by computers, add the latest ID papers with smart chips in them … it’s almost as if we ourselves are chipped – similar to cats and dogs.