How AI is Redefining Emotional Intelligence: Insights from a Groundbreaking Study
Artificial Intelligence (AI) has already transformed industries from finance to healthcare, but can it truly understand human emotions? A recent study by Schlegel, Sommer, and Mortillaro, published in Communications Psychology, explores this question through the lens of Emotional Intelligence (EI)—and the results are nothing short of remarkable.
Emotional Intelligence Meets Artificial Intelligence
Emotional Intelligence, or the ability to recognize, understand, and manage emotions in ourselves and others, is critical in social interactions and personal success. But it's also one of the most elusive traits to replicate in machines. This study tested the capabilities of Large Language Models (LLMs) such as ChatGPT-4, Claude 3.5 Haiku, Copilot 365, Gemini 1.5 Flash, and others, to see how well they can perform on standard EI assessments—and even create them.
Outperforming Humans in EI Tests
The research team evaluated six state-of-the-art LLMs using five widely recognized emotional intelligence tests. These tests assessed the models’ abilities to:
- Understand emotional causes and effects (e.g., Geneva Emotion Knowledge Test)
- Recommend emotion regulation strategies (e.g., Geneva Emotional Competence Test)
- Identify effective emotional responses in workplace or life scenarios
Key finding: The LLMs achieved an average accuracy of 81%, significantly outperforming the human average of 56% reported in original test validations. In simpler terms, the machines demonstrated a better grasp of emotional decision-making than most people.
Can AI Design Its Own EI Tests? Apparently, Yes.
In the second phase, ChatGPT-4 was tasked with generating new items for each emotional intelligence test. These new versions were then given to human participants, alongside the original tests, across five different studies involving 467 people.
The result? ChatGPT’s test versions were statistically equivalent in difficulty to the originals. While there were small differences in clarity, realism, and test reliability, none exceeded a medium effect size (Cohen’s d = ±0.25). Furthermore, the AI-generated and original test scores were significantly correlated (r = 0.46), suggesting similar measurement validity.
Why This Matters
This research holds big implications for both AI development and emotional learning technologies:
- AI Systems with Emotional Intelligence: From customer service chatbots to therapeutic companions, emotionally intelligent AI could revolutionize user experience by better understanding and responding to human emotions.
- Educational Tools: AI can help develop personalized training for emotional regulation, a skill crucial in schools, workplaces, and healthcare settings.
- Test Development: ChatGPT-4’s ability to generate valid test items could drastically reduce the time and cost of creating new assessments in psychology and education.
Challenges and Ethical Considerations
While these findings are exciting, they come with caution. Just because an AI scores highly on EI tests doesn’t mean it feels emotions or understands them in a human way. There’s a fine line between simulating empathy and experiencing empathy—a line that’s still philosophically and scientifically murky.
Final Thoughts: Toward Affective AI
This study suggests that LLMs are not just linguistically capable—they're beginning to show competence in emotional reasoning. As AI continues to evolve, integrating emotional intelligence into machines might not just be possible—it might become essential.
Whether in healthcare, education, or the workplace, emotionally aware AI could be a powerful tool for improving human-AI interaction and fostering more compassionate technology.
Check out the full publication for more information here:
- Katja Schlegel, Nils R. Sommer, Marcello Mortillaro. Large language models are proficient in solving and creating emotional intelligence tests. Communications Psychology, 2025; 3 (1) DOI: 10.1038/s44271-025-00258-x
