Can AI Really Understand Human Emotion?

June 10, 2021

An article at IT News Africa makes quite an assumption about allegedly smart software. It purports to describe “How Emotion AI Can Make the World a Better Place.” Writer Jenna Delport is, of course, correct when she notes many people struggle to understand others’ emotions. But is the best answer really to hand off the interpretation to an algorithm? Delport writes:

“Researchers at Stanford University modified Google’s augmented reality glasses to read emotions in others and notify the wearer. The glasses detect someone’s mood through their eye contact, facial expressions and body language, and then tell the wearer what emotions it’s picking up. ‘Emotion AI taps into the individual,’ explains Zabeth Venter, CEO and co-founder of Averly. ‘If you think about facial recognition, which is a kind of emotion AI, I can pick up if you like what I’m saying by whether your smile is a smirk or a real genuine smile.’ Such nuances go deeper. Another example is polling: what is your favorite color? Maybe it’s purple. But did you say that enthusiastically? Did you hesitate? Did you just say it to say something? Did you even understand the question? We simply can’t get this level of context from the available surveys, sales data and the many other ways we try to understand humans through information. But through emotion AI, we can grasp incredible nuance.”

Incredible nuance is great—if it is accurate. The write-up acknowledges AI’s problem with bias, but suggests combating it is as simple as training AI on better data sets than what is simply (and cheaply) available on the Web. Instead, developers would capture cultural nuances by training their AI on localized data. They would also, we’re told, balance input from male and female perspectives. That sounds like a good start, but will eliminating bias really be that simple? It may be premature to make promises about AI’s new and improved emotional sensitivity.

Cynthia Murrell, June 10, 2021

Comments

Got something to say?





  • Archives

  • Recent Posts

  • Meta