AI Now Outperforms Humans in Emotional Intelligence

We often praise artificial intelligence for its impressive coding and math skills, but how does it handle something as profoundly human as emotions? A recent study from the University of Geneva and the University of Bern delivered a startling answer: popular AI systems like ChatGPT may have a better grasp of emotional intelligence than the average person.

Researchers set out to see if machines could recognize and reason about emotions in the same way humans do. The results were not just a simple yes—they were a resounding one. Across five distinct tests measuring emotional understanding and regulation, the six AI models studied answered an average of 81% of questions correctly. In contrast, the average human participant only managed a 56% correct response rate. These findings challenge the deep-rooted assumption that empathy and emotional awareness are exclusive to humanity.

How Did Researchers Test AI's Emotional IQ

To measure the AI's capabilities, the research team used well-established assessments that psychologists use to evaluate "ability emotional intelligence." Unlike personality quizzes, these tests have objectively right and wrong answers. The questions presented subjects with a specific situation and asked them to identify the most likely emotion a person would feel or to choose the best course of action to help someone manage their feelings.

The models tested included ChatGPT-4, Gemini 1.5 Flash, and Claude 3.5 Haiku, among others. Each system took the tests ten times to establish a reliable average score, which was then compared against the scores of human participants from previous validation studies. The conclusion was clear: every AI model outperformed the human average on every single test. The models also showed a high degree of agreement with each other, suggesting they were arriving at similar emotional judgments despite not being specifically trained on these emotional evaluations.

"LLMs can not only identify the best option among many available ones, but also create new scenarios that suit the desired context," noted Katja Schlegel, a lecturer at the University of Bern's Institute of Psychology and the study's lead author.

The Psychological Framework of the Study

The tests were designed to cover different facets of emotional intelligence. The Situational Test of Emotion Understanding (STEU) and the Geneva Emotion Knowledge Test - Blends (GEMOK-Blends) focused on recognizing emotional states. Meanwhile, the Situational Test of Emotion Management (STEM) and subtests from the Geneva Emotional Competence Test (GECo) evaluated skills in emotional regulation and management.

Each question was grounded in a realistic scenario. For example, if a colleague steals your idea and gets praised for it, the emotionally intelligent response isn't to confront them angrily but to calmly approach a supervisor to discuss the situation. This demonstrates emotional control, a key aspect of what the tests measured.

"The results showed significantly higher scores for the LLMs - 82%, compared to 56% by human participants," explained Marcello Mortillaro, a senior scientist on the project. "This indicates that these AIs not only comprehend emotions, but also possess an understanding of functioning with emotional intelligence.”

Can AI Not Only Pass the Test but Write It Too

After seeing the AI models excel at taking the tests, the researchers posed an even more ambitious question: could an AI write a valid emotional intelligence test itself? They prompted ChatGPT-4 to generate new test scenarios, complete with multiple-choice answers and a designated correct response. This AI-generated test, along with the original human-made version, was then given to 467 human participants.

The outcome was once again unexpected. The test created by ChatGPT-4 was just as effective and challenging as the one designed by human experts. Participants scored similarly on both versions, and the AI's questions were rated as equally clear, realistic, and varied. Statistically, the two tests had an "equivalent difficulty," proving that the AI understood the very logic behind how psychologists measure emotional reasoning.

Acknowledging the AI's Limitations

The researchers did look for potential flaws in the AI-written tests. They found that human-written questions were sometimes rated as slightly clearer and that the scenarios created by the AI were a bit less diverse. However, these differences were minor and did not change the overall conclusion.

Crucially, the AI was not just plagiarizing. An analysis showed that 88% of the items it generated were completely original. The AI-created tests also showed similar correlations to vocabulary and other emotional intelligence measures, just like the originals, confirming that the AI was testing the correct abilities.

The Future of Emotionally Aware Technology

For years, the field of affective computing has tried to train machines to recognize emotions from faces, voices, or words, but these systems often lacked genuine emotional understanding. Modern large language models appear to have overcome this hurdle, having implicitly learned emotional reasoning patterns from the vast amounts of human text they were trained on.

While researchers are careful to state that these models don't "feel" emotions, they can clearly understand them and respond appropriately. This means a tutoring bot could recognize a student's frustration or a healthcare assistant could offer a comforting response, providing real help without having its own human-like emotional experiences. This opens new doors for emotionally aware technology in therapy, coaching, and education, paving the way for a future where AI can offer more empathetic and effective support.