Follow us on Google News (click on ☆)
LLMs achieved significantly better scores with 82% correct answers compared to 56% for humans.
© Adobe Stock
Can artificial intelligence (AI) suggest appropriate behaviors in emotionally charged situations? A team from the universities of Geneva (UNIGE) and Bern (UniBE) submitted six generative AIs, including ChatGPT, to emotional intelligence tests typically reserved for individuals. Result: these AIs surpass human performance and can generate new tests in record time. These findings open new possibilities for AI applications in education, coaching, and conflict management. They are published in Communications Psychology.
Large language models (LLMs) are artificial intelligence (AI) systems capable of processing, interpreting, and generating human language. Generative AI like ChatGPT is based on this type of model. These models can answer questions and solve complex problems. But can they also suggest emotionally intelligent behaviors?
These results pave the way for AI use in contexts previously thought to be exclusive to humans.
Scenario-based testing
To find out, a team from the Institute of Psychology at UniBE and the Swiss Center for Affective Sciences (CISA) at UNIGE submitted six LLMs (ChatGPT-4, ChatGPT-o1, Gemini 1.5 flash, Copilot 365, Claude 3.5 Haiku and DeepSeek V3) to emotional intelligence tests.
"We selected five tests used in research and professional settings. They consisted of numerous emotionally charged scenarios designed to assess emotional understanding, regulation, and management skills," explains Katja Schlegel, lecturer and principal investigator in personality psychology, differential psychology, and psychological assessment at the Institute of Psychology at UniBE, and lead author of the study.
For example: "Michael's colleague stole his idea and is receiving undeserved praise. What would be Michael's most effective reaction?"
a) Argue with the colleague in question.
b) Discuss the situation with his supervisor.
c) Resent his colleague silently.
d) Steal an idea from him in return.
Here, reaction b) was considered the most appropriate.
The same five tests were also given to human participants. "Ultimately, the LLMs achieved significantly better scores with 82% correct answers compared to 56% for humans," reveals Marcello Mortillaro, scientific assistant at CISA at UNIGE, involved in this work. "This demonstrates that these AIs possess knowledge about emotions and what constitutes emotionally intelligent behavior."
New tests in record time
In a second phase, researchers asked ChatGPT-4 to create new emotional intelligence tests with original scenarios. These automatically generated tests were then given to over 400 participants.
"They proved to be as reliable, clear, and realistic as the original tests, which had taken years to develop," notes Katja Schlegel.
"LLMs are therefore not only capable of identifying the best answer among given options but also of generating new scenarios adapted to a desired context. This reinforces the idea that LLMs like ChatGPT are capable of emotional reasoning," adds Marcello Mortillaro.
These results open the door to AI applications in areas previously considered human-exclusive, such as education, coaching, or conflict management, provided it is supervised by experts.