If Microsoft’s current research pans out, software could learn to accurately gauge human emotion.
The company’s Project Oxford, a collection of APIs designed to layer “intelligent solutions” into developers’ software platforms, now includes an Emotion Recognition API, which studies faces in images and rates their level of anger, contempt, disgust, fear, happiness, sadness, surprise, or “neutral.” Anyone can test the API by uploading their own images.
If any of this sounds familiar, it’s because Microsoft performed a similar experiment a few months back with how-old.net, a Website that uses a machine-learning API (called Face API) to guess the age of people in photographs. While the platform wasn’t perfect—it mistook an image of a fortysomething Teddy Roosevelt for a 63-year-old woman, for example—it did provide hours of amusement for visitors.
However, emotion recognition produces more multifaceted data. The expression on the model in this Shutterstock-derived image from Dice Insights, for example, primarily returned “neutral” (with a score of 0.749, for those keeping count) and “sadness” (0.247), with the faintest hints of anger, surprise, and contempt.
(When presented with too little information, the system also declines to make guesses. As an experiment, we uploaded an image from the video game “Destiny,” in which the characters wear expressionless masks; the API returned an error message.)
Although the system isn’t perfect, it’s well worth checking out—even if you’re not an artificial intelligence specialist or developer.