Can artificial intelligence (AI) propose appropriate behaviors in an emotional situation? A team from the University of Geneva (Unige) and the University of Bern (UNIBE) included six generative AISs, including Chatgpt, into the Emotional Intelligence (EI) assessment, which is usually designed for humans. Results: These AISs outperformed the average performance of humans and were able to produce new tests in record time. These findings open up new possibilities for AI in education, coaching and conflict management. The study was published in Communication Psychology.
Large Language Models (LLMs) are artificial intelligence (AI) systems that can process, interpret and generate human language. For example, CHATGPT generates AI based on this type of model. LLM can answer questions and solve complex problems. But can they still propose emotionally intelligent behavior?
These results pave the way for AI to be used in contexts believed to be reserved for humanity.
Emotional scene
To find out, a team from Unibe, the Institute of Psychology and Unige Swiss Centre for Emotional Sciences (CISA) conducted emotional intelligence tests on six LLMs (Chatgpt-4, Chatgpt-O1, Gemini 1.5 Flash, Copilot 365, Claude 3.5 Haiku and DeepSeek V3). “We selected five tests commonly used in research and company settings. They involve emotional scenarios designed to assess the ability to understand, regulate and manage emotions,”
For example: A colleague of Michael stole his ideas and congratulated him unfairly. What is Michael’s most effective response?
a) Arguing with colleagues involved
b) Talk about the situation with your superiors
c) Silently hate his colleagues
d) Steal back
Here, option b) is considered the most appropriate.
At the same time, the same five tests were performed on human participants. “Ultimately, LLM scored significantly higher – 82% of the correct answers versus 56% of humans. This shows that these AIs not only understand emotions, but also master what the behavior of emotional intelligence means,” explains Marcello Mortillaro, senior scientist at Unige’s Swiss swiss Center for Swiss (CISA) (CISA).
New tests for record time
In the second phase, scientists asked Chatgpt-4 to create new emotional intelligence tests through new scenarios. These automatically generated tests were then performed by more than 400 participants. “It turns out they are as reliable, clear and realistic as the original tests, which has taken years.” So LLM is able to find the best answers among the various available options, but also generate new scenarios that fit the desired context. This exacerbates the notion that LLM (e.g. Chatgpt) has emotional knowledge and can reason about emotions. ”
These results pave the way AI is used in contexts considered reserved for humanity, such as education, coaching, or conflict management, as long as it is used and supervised by experts.
