Using Large Language Models (LLMs) to Generate Emotional Intelligence Test Items
This project investigates whether Large Language Models (LLMs) can be used to generate valid performance-based emotional intelligence test items. In a series of studies, we compared AI-generated test items with established emotional intelligence measures in terms of clarity, realism, difficulty, reliability, and validity. The results, published in Communications Psychology, showed that items generated by ChatGPT-4 performed similarly to human-developed items across these criteria. These findings suggest that LLMs may offer an efficient way to support the development of new emotional intelligence assessments and training materials, while also providing insights into how well AI systems understand and reason about emotions.
Read more in our Springer Nature Research Communities Blog post
Collaborators: Nils Sommer, Marcello Mortillaro