Can artificial intelligence (AI) truly understand how humans should behave in emotionally intense moments? A recent study from the University of Geneva (UNIGE) and the University of Bern (UniBE) suggests it might. When tested on emotional intelligence (EI) assessments traditionally reserved for people, six major AI systems, including ChatGPT, demonstrated a surprising edge over the human average.
These AI models didn’t just perform well; they outscored humans significantly and even created new test material that rivaled existing tools developed over many years. The findings, now published in Communications Psychology, suggest AI could soon play a role in domains requiring emotional sensitivity—like education, coaching, or conflict resolution.
What should Michael do?
“These aren’t just language models parroting information,” said Dr. Katja Schlegel, lead researcher from UniBE. “They’re demonstrating a contextual understanding of emotional behavior that rivals human reasoning.”
“This ability to both understand and create emotionally nuanced content suggests a level of emotional reasoning previously thought unique to humans,” said Marcello Mortillaro, a senior scientist at UNIGE’s Swiss Center for Affective Sciences.
Still, the implications are profound. “AI can now meaningfully contribute to areas that require not just intelligence, but emotional intelligence,” said Mortillaro.
With the right guardrails, these systems could help train future leaders, mediate conflicts, or even assist educators in handling emotionally complex classroom scenarios—tasks once thought to be beyond the scope of machines.
Image: DIW-Aigen
Read next: Chatbots Hallucinate More With Confident or Short Prompts, Accuracy Drops Up to 20% in Critical Tasks
These AI models didn’t just perform well; they outscored humans significantly and even created new test material that rivaled existing tools developed over many years. The findings, now published in Communications Psychology, suggest AI could soon play a role in domains requiring emotional sensitivity—like education, coaching, or conflict resolution.
Testing AI in Emotionally Complex Scenarios
The researchers drew on five emotional intelligence assessments widely used in psychology and workplace evaluations. These tests ask respondents to judge or act on emotionally tense scenarios. One such prompt involved a worker named Michael, whose colleague had taken credit for his idea:What should Michael do?
- Confront the colleague
- Speak with his manager
- Hold a grudge
- Steal an idea in return
“These aren’t just language models parroting information,” said Dr. Katja Schlegel, lead researcher from UniBE. “They’re demonstrating a contextual understanding of emotional behavior that rivals human reasoning.”
Building New Tools in Hours, Not Years
In the second part of the study, researchers asked ChatGPT-4 to go further—by generating new EI assessments from scratch. It did so within hours. These fresh tests, after being reviewed and administered to over 400 human participants, proved to be just as clear, realistic, and reliable as those crafted through years of psychological research.“This ability to both understand and create emotionally nuanced content suggests a level of emotional reasoning previously thought unique to humans,” said Marcello Mortillaro, a senior scientist at UNIGE’s Swiss Center for Affective Sciences.
A Future Role for AI in Emotion-Centric Fields
While the findings are promising, researchers stress that human oversight remains essential. Even if AI can navigate emotions with apparent fluency, its use in sensitive areas like counseling or leadership development must be carefully guided.Still, the implications are profound. “AI can now meaningfully contribute to areas that require not just intelligence, but emotional intelligence,” said Mortillaro.
With the right guardrails, these systems could help train future leaders, mediate conflicts, or even assist educators in handling emotionally complex classroom scenarios—tasks once thought to be beyond the scope of machines.
Read next: Chatbots Hallucinate More With Confident or Short Prompts, Accuracy Drops Up to 20% in Critical Tasks
