Eminent psychologists condemn "emotion detection" systems as being grounded in junk science

One of the more extravagant claims made by tech companies is that they can detect emotions by analyzing photos of our faces with machine learning systems. The premise is sometimes dressed up in claims about "micro-expressions" that are below the threshold of human detection, though some vendors have made billions getting security agencies to let them train officers in "behavior detection" grounded in this premise.

A panel of eminent psych researchers have signed an open letter condemning these products as grounded in "outdated science." The authors cite more than 1,000 journal articles that show that facial expressions are complex and cannot be classified using the techniques promoted by the self-interested commercial peddlers of emotion detection systems: a scowl is not a reliable indicator of anger, nor is a smile a reliable indicator of happiness.

We have arrived at a long-overdue and extremely welcome techlash, in which Big Tech's promises about how it stores and manages our data are being looked at with the skepticism they deserve. But many of these critics are remarkably unskeptical about the claims Big Tech makes about the efficacy of its products, taking at face value Big Tech's sales-literature boasts about being able to detect and manipulate our opinions.

If we're ready to believe that Big Tech lies about its taxes, its infosec practices, its anti-harassment policies, its privacy policies, its lobbying activities, and everything else it claims about itself, shouldn't we also ask whether its products actually work?

It may well be that Big Tech is full of people who believe that its products work, just as the private equity world is full of money managers who believe that they can outperform a simple, low-load index fund. Read the rest