AI and Large Language Models Show Promising Skills in Emotional Intelligence Testing

Recent research indicates that large language models (LLMs), which are the artificial intelligence systems powering conversational tools like ChatGPT, are highly capable of both creating and solving emotional intelligence (EI) tests. These tests are designed to evaluate a person's ability to recognize, understand, and manage emotions—skills that are vital for social interactions throughout life.

A study conducted by researchers at the University of Bern and the University of Geneva examined the performance of multiple LLMs, including ChatGPT-4, Gemini 1.5 flash, Claude 3.5, Haiku, and DeepSeek V3, on five widely used EI assessments originally developed for humans. The models were tasked with interpreting short emotional scenarios and selecting the most appropriate responses, which measure emotional recognition, reasoning, and regulation.

Remarkably, the LLMs achieved an average accuracy of 81% on these tests, significantly surpassing the human average of 56%. Furthermore, ChatGPT-4 not only successfully solved the tests but also generated new EI test items with clarity and realism comparable to the original ones, indicating a sophisticated understanding of emotional concepts.

The study further explored the models' ability to create new psychological evaluations. Over 460 human participants then rated both the original and AI-generated tests based on objectives like difficulty, clarity, and realism. Results showed high alignment between AI-produced tests and human standards, emphasizing the models' capacity for deep emotional reasoning.

These findings open promising avenues for developing automated tools in psychological assessments, training materials, and social simulation scenarios. Such applications could streamline the creation of emotional intelligence resources and enhance the capabilities of social agents like mental health chatbots and educational tutors, particularly in emotionally sensitive interactions.

Looking ahead, researchers aim to test the models in more complex, real-world emotional conversations and assess their cultural sensitivity, as current models primarily reflect Western-centric data. Overall, this study underscores the growing potential of AI to understand and replicate human emotional skills, offering new tools to support mental health, social skills training, and human-AI interaction.