LLMs & Emotional Intelligence: New Study Findings
- Large language models (LLMs) are showing surprising aptitude in understanding and responding to emotions, according to new research.
- The ability to perceive, manage, and understand emotions, known as emotional intelligence, is crucial for social interactions.
- Katja Schlegel, the study's first author, noted the natural progression from developing EI tests for humans to evaluating LLMs.
large Language Models (LLMs) are demonstrating remarkable proficiency in emotional intelligence (EI), according to new research.Discover how cutting-edge AI,like ChatGPT-4,not only aces EI tests but also creates them with surprising accuracy. The study’s findings reveal LLMs achieving an notable 81% average accuracy on EI tests, outperforming human averages. These developments underscore a important leap in AI’s capacity for emotional reasoning and open up new avenues for developing advanced training materials and social agents. By examining how AI can mimic human emotional intelligence, we can enhance AI interfaces while gaining deeper understanding of human psychology. News Directory 3 brings you the latest updates on how Large Language Models are changing the world, specifically around the request of secondary_keyword. Discover what’s next as researchers examine how llms perform in real-world emotional exchanges.
Large Language Models Excel at Emotional Intelligence Tests, study Finds
Updated June 4, 2025
Large language models (LLMs) are showing surprising aptitude in understanding and responding to emotions, according to new research. A study reveals that these AI systems,including models like ChatGPT,can not only solve emotional intelligence (EI) tests but also create them with a level of quality comparable to human-generated tests. This suggests that LLMs possess a significant capacity for emotional reasoning.
The ability to perceive, manage, and understand emotions, known as emotional intelligence, is crucial for social interactions. Psychologists have long used EI tests to assess these skills in individuals across various settings.Now, researchers from the University of Bern and the University of Geneva have explored how well LLMs perform on these tests.
Katja Schlegel, the study’s first author, noted the natural progression from developing EI tests for humans to evaluating LLMs. The research, published in Communications Psychology, delves into whether AI can genuinely understand and respond to emotions, a key aspect of empathy.
the study involved having six LLMs—including ChatGPT-4, Gemini 1.5 Flash, and Claude 3.5 Haiku—complete five established EI tests. These tests presented emotional scenarios, requiring the models to identify appropriate responses.The LLMs’ scores were then compared to average human scores from previous studies.
In a second phase, the researchers tasked ChatGPT-4 with creating entirely new EI tests, complete with scenarios, questions, and answer keys. These AI-generated tests, along with the original versions, were then given to over 460 human participants to assess their difficulty, clarity, and realism.
The results showed that LLMs achieved an average accuracy of 81% on the EI tests, surpassing the average human accuracy of 56%. Moreover, human participants rated the AI-generated test items as similarly clear and realistic as the original ones.
Even more impressively, ChatGPT-4 was able to generate entirely new EI test items that were rated by human participants as similarly clear and realistic as the original items and showed comparable psychometric quality,” said Schlegel.
These findings suggest that LLMs possess a high level of conceptual understanding of emotions. The ability of large language models to both solve and construct EI tests opens new avenues for developing training materials and social agents, such as mental health chatbots and educational tutors.
What’s next
Future research will explore how well LLMs perform in unstructured,real-life emotional conversations and examine the cultural sensitivity of their emotional reasoning,as current models are primarily trained on Western-centric data. This exploration into emotional intelligence and large language models promises to refine AI’s ability to interact with humans in emotionally intelligent ways.