New Tool Show The Inaccuracies of Emotional Detection AI

The use of AI to detect your emotions is a $20 billion industry that encompasses everything from focus grouping ads to assessing the faces of job candidates in interviews. But scientists claim the premise behind this tech—that one's emotions can be read directly from the contours of one's facial expressions—doesn't actually hold up to scrutiny. A new web app from researchers at the University of Cambridge is designed to demonstrate how wildly inaccurate these systems can be while educating people on the extent to which they are already in use across a host of different applications. The tool, called Emojify.info, taps the user's webcam and invites them to try to fool a standard emotional recognition AI with their own expressions. While the system is fairly successful at identifying straightforward smiles or frowns, it often labels even wildly exaggerated facial contortions as "neutral" if they don't fall within the strict bounds of the six emotions it's trained to recognize—happiness, sadness, fear, surprise, disgust and anger. Even so, the expressions the user is making presumably have no correspondence to their emotional state at the given moment, demonstrating the face isn't always a direct window into feelings. "Emotion recognition technology has critical flaws baked into it," the creators of the tool said in a video introducing the tool. "Emotion recognition is based on a disputed theory that there are six basic emotions common to everyone. But many researchers believe this is too simplistic. It ignores context, differences between cultures and that emotions just may not fit into neat categories at all." Another game within the app is designed to highlight how the technology fails to differentiate between a wink and a blink, making a point about how AI systems are unable to grasp the social and interpersonal contexts that are key to actually understanding people's emotions beyond simple expressions. "We humans aren't emojis. We frown when we're angry, but also when we're concentrating. We smile when we're happy, but we also fake-smile in uncomfortable situations. We cry at funerals and also at weddings and graduations," the narrator says. "We interpret emotional expression in the context of everything going on around us." While facial recognition tech has been the subject of much controversy for its privacy issues and biases around race and gender, emotional recognition—which suffers from discrimination issues of its own—has yet to receive the same level of attention despite its rapid growth, the researchers behind the tool contend. They are hoping demonstrations like these will help make the practice a bigger part of the conversation around AI and machine learning ethics.

AI News

Researchers Create Tool To Raise Awareness About Emotional Recognition AI

The web app is meant to highlight the flaws in this fast-growing technology

Patrick Kulp

About

Subscriptions

Events

Publications

WORK SMARTER - LEARN, GROW AND BE INSPIRED.

Subscribe today!

Patrick Kulp