Six generative AI large language models (LLMs) were able to outperform real people in tests on emotional intelligence, a recent study showed.
The study, which was published in the peer-reviewed journal Communications Psychology in May, saw the LLMs ChatGPT-4, ChatGPT-o1, Gemini 1.5 flash, Copilot 365, Claude 3.5 Haiku, and DeepSeek V3, received a better score than humans on five emotional intelligence tests, receiving an average grade of 81% compared to 56%, which was the average grade of the humans.
The emotional intelligence tests include: Situational Test of Emotion Management (STEM), Situational Test of Emotion Understanding (STEU), Geneva EMOtion Knowledge Test (GEMOK), and the Geneva Emotional Competence Test in the workplace.
Additionally, one of the LLMs, ChatGPT-4, was asked by researchers to generate new assessments and tests that were utilized for the research.
Quickly doing years of work in creating new assessments
"They proved to be as reliable, clear, and realistic as the original tests, which had taken years to develop," said Katja Schlegel, one of the co-authors of the study.
The study was conducted by a team of researchers from the University of Geneva (UNIGE) and the University of Bern (UniBE) in Switzerland, as well as the Czech Academy of Sciences in Prague.
"We chose five tests commonly used in both research and corporate settings," added Schlegel. "They involved emotionally charged scenarios designed to assess the ability to understand, regulate, and manage emotions."
Another co-author, Marcello Mortillaro of Geneva's Swiss Center for Affective Sciences, said that the study's results "suggest that these AIs not only understand emotions, but also grasp what it means to behave with emotional intelligence."
The researchers concluded in the study that large language models can be utilized in supporting socio-emotional outcomes, or assist those in gaining new perspectives on emotional situations, and assist them when making large decisions.