machine learning


Toward A More Emotionally Astute Machine

Read this great story in The New York Times today by CommonHealth contributor Karen Weintraub on the future of affective programming, or teaching machines to read facial expressions and emotions. Here’s the lede:

In a Cairo school basement, two dozen women analyze facial expressions on laptops, training the computers to recognize anger, sadness and frustration.

At Cambridge University, an eerily realistic robotic head named Charles sits in a driving simulator, furrowing its brows, looking interested or confused.

And in a handful of American middle school classrooms this fall, computers will monitor students’ emotions in an effort to track when they are losing interest and when they are getting excited about lessons.

All three are examples of an emerging approach to technology called affective computing, which aims to give computers the ability to read users’ emotions, or “affect.”

People are good at understanding one another’s emotions. We realize quickly that now is not a good time to approach the boss or that a loved one is having a lousy day. These skills are so essential that those without them are considered disabled.

Yet until recently, our machines could not identify even seemingly simple emotions, like anger or frustration. The GPS device chirps happily even when the driver is ready to hurl it out the window. The online class keeps going even when half the students are lost in confusion. The airport security system can’t tell whether someone is behaving as if he were concealing something or is just anxious about flying.

The piece details work by Rosalind Picard, founder and director of the Affective Computing Research Group at MIT’s Media Lab. Picard and her colleagues are developing technology to measure emotional arousal through skin sensors, something that might ultimately provide a kind of lifeline for people with autism or similar disorders who have difficulty making social connections and communicating emotions: Continue reading