🧠 Can AI Detect Emotion?

Sign Up | Home Page | Meet The Team

In today’s Sunday Special:

Read Time: 7 minutes

📜THE PRELUDE

You adjust your posture: chin centered, shoulders squared, and hands hidden. It’s time for your virtual job interview, but there’s no interviewer.

A set of instructions fills your laptop’s screen:

You speak carefully, conscious of your cadence, striving to project confidence without cockiness and balance authority with approachability. There’s no interviewer to offer an encouraging nod, supportive glance, or reassuring smile. Just a blinking red timer, counting down with an unyielding rhythm.

Your response is recorded and instantly processed by AI-powered interview analysis software to scan sentiment, assess authenticity, and evaluate engagement. You think your answer was engaging, impactful, and memorable. Yet, you wonder: Was my tone too flat? Was my body too stiff? Did I smile enough?

Job interview performance has always had a subjective component. Now, it’s automatically measured. So, how does AI quantify the emotion conveyed through your tone, pace, and pitch?

😱WHAT’S EMOTION?

In 1972, American psychologist Paul Ekman developed BET, proposing six universal emotions: fear, anger, disgust, surprise, sadness, contempt, and happiness. He believed emotions are automatic appraisals influenced by our evolutionary past and personal development. In simple terms, emotions are instinctive impulses triggered by a mix of biological hardwiring and life experiences. So, what inspired his theory?

In 1872, English biologist Charles Darwin published “The Expression of the Emotions in Man and Animals,” in which he argued that emotional expressions are evolutionary adaptations. In other words, if emotions evolved for our survival (e.g., “fear to avoid danger!”), then the facial expressions associated with those emotions are also biologically hardwired and should be widely shared across cultures and societies.

To test this theory, Ekman traveled to Papua New Guinea to photograph the Fore people of the Eastern Highlands, a relatively isolated indigenous society with minimal exposure to Western culture. He shared short stories that correlated with specific emotions. For example, the loss of a loved one for sadness. He photographed the Fore people’s facial expressions as they reacted to these short stories.

The photographed facial expressions were shown to 316 American college students, who identified the Fore people’s emotions with remarkable consistency:

Ekman helped settle a long-held belief that emotional facial expressions are learned behaviors that vary from culture to culture. He demonstrated that while “display rules” (i.e., when it’s appropriate to show emotion) are culturally learned, the “motor programs” (i.e., the muscle movements of the face) are biologically hardwired.

Ekman’s findings led to the discovery of micro-expressions, which suggest fleeting facial movements reveal concealed emotions. Today, it’s widely taught at law enforcement agencies like the FBI to detect deception when performing domestic intelligence. In fact, AI emotion recognition heavily relies on BET because it assumes stable, cross-cultural mappings between facial muscle patterns and discrete emotional states.

🔍CAN AI “DETECT” EMOTIONS?

Swedish psychologists at the University of Bern (UoB) recently discovered that frontier AI models are better at recognizing and responding to emotional cues than humans by identifying emotion-related behavioral patterns through observable expressions, such as facial cues or vocal bursts.

Being able to communicate your emotions is crucial for developing and maintaining social bonds. Individuals with the ability to recognize, understand, express, and respond to emotions achieve better outcomes across life domains. For example, individuals with higher Emotional Intelligence (EI) are perceived as warmer and more competent during workplace conflicts. As frontier AI models become increasingly integrated into our everyday lives, it’s essential that they’re not only functionally efficient but also emotionally intelligent.

They examined the ability of six frontier AI models to recognize and respond to emotional cues across five standard EI Tests. Surprisingly, the six frontier AI models outperformed humans across all five standard EI Tests. OpenAI’s “OpenAI o1,” Google’s “Gemini 1.5 Flash,” and Anthropic’s “Claude 3.5 Haiku” scored an average of about 81% accuracy. In comparison, humans scored an average of about 56% accuracy.

So, what’s going on here? In a nutshell, frontier AI models are trained on vast amounts of high-quality datasets filled with books, articles, movies, and podcasts rich in emotional language and emotional reasoning. While frontier AI models can’t “feel,” they can mimic emotional understanding by recognizing the data-driven patterns that underpin emotional scenarios, enabling them to predict appropriate emotional reactions better than humans on average.

👀HOW MACHINES SEE FEELINGS

Noldus’s AI-enabled facial recognition software gives a complete, contact-free view of facial expressions for behavioral research. The latest version, FaceReader 10, accurately analyzes and assesses emotions about 88% of the time. Here’s how it performs across the six universal emotions:

🔑KEY TAKEAWAY

“Emotion AI,” also known as affective computing, dates back to at least 1995. We’ve always aimed to achieve more natural interactions between humans and machines. Today, our emotions aren’t just “felt”; they’re quantifiable signals waiting to be interpreted by machines. We’re witnessing the boundary between human intuition and artificial perception blur.

📒FINAL NOTE

{{rp_personalized_text}}

Share your unique referral link: {{rp_refer_url}}