Artificial intelligence has crossed another threshold in its evolution toward more human-like capabilities. On May 22, 2025, researchers from the University of Geneva (UNIGE) and the University of Bern (UniBE) published findings in Communications Psychology showing that AI systems can not only understand emotions but may actually be better at suggesting appropriate emotional responses than humans.
The research team, led by Katja Schlegel from UniBE's Institute of Psychology, evaluated six advanced large language models (LLMs) including ChatGPT-4, ChatGPT-o1, Gemini 1.5 Flash, Copilot 365, Claude 3.5 Haiku, and DeepSeek V3. These AI systems were subjected to five emotional intelligence tests commonly used in academic and corporate settings.
The tests presented emotionally charged scenarios designed to assess the ability to understand, regulate, and manage emotions. In one example, participants were asked to determine the most effective reaction for someone whose colleague had stolen their idea and was being unfairly congratulated. When the same tests were administered to human participants, the results were striking: the AI systems achieved an average score of 82% correct answers, significantly outperforming humans who averaged only 56%.
"This suggests that these AIs not only understand emotions, but also grasp what it means to behave with emotional intelligence," explained Marcello Mortillaro, senior scientist at UNIGE's Swiss Center for Affective Sciences, who participated in the research.
Perhaps even more remarkable was the second phase of the study, where ChatGPT-4 was tasked with creating entirely new emotional intelligence tests with original scenarios. These AI-generated tests were administered to over 400 human participants and proved to be as reliable, clear, and realistic as the original tests that had taken experts years to develop.
The findings open promising avenues for AI applications in fields previously considered exclusively human domains, such as education, coaching, and conflict management. However, researchers emphasize that such applications would require proper expert supervision. As AI continues to demonstrate increasingly sophisticated emotional capabilities, the boundary between human and machine intelligence grows ever more nuanced.