Standard image of the new scientist

Emotion-sensing AIs are there – do they work and how should we feel?

Several companies have developed AIs to find out if you’re angry, sad, or excited. However, there are serious questions about their accuracy and the extent to which they should be used in public life


18th November 2020

By Chris Baraniuk

Some AIs try to identify people’s emotions based on their facial expressions

affective Fr.

RANA EL KALIOUBY was alone in her apartment and wrote a message to her husband. “How are you?” he typed. “I’m fine,” she typed back. Except that it wasn’t. The couple had been separated for weeks and she felt miserable. Had he been in the room, he could have read the emotions on her face at a glance. But it was miles away.

It’s a scene that could easily have played out during a coronavirus pandemic lockdown, when colleagues, friends, and even families were cut off from one another. But it actually happened 20 years ago, shortly after el Kaliouby moved from Egypt to the UK to study, leaving her husband behind.

It was then, she says, that she realized how blind technology was to human emotions. Since then, el Kaliouby has dreamed of building an emotionally intelligent computer – or as she calls it: “a mind-reading machine”. With so many relationships mediated through text or video calls these days, this is a technology that couldn’t be more relevant.

Today, Affectiva and similar companies, co-founded by el Kaliouby, claim to have systems that can recognize human emotions. The promises they make about the potential of this emotional artificial intelligence (AI) are staggering. Computers, it is said, will know if we are distracted while driving, angry emails we may regret, or when our sanity is beginning to decline. In fact, such systems already exist. But do they live up to their bill? And TU …