Understanding AI Emotional Intelligence

Welcome to our collaborative effort to develop a comprehensive framework for evaluating emotional intelligence in AI systems.

Our work is guided by the four SERA axes—Sense, Explain, Respond, and Adapt—which together describe how emotionally intelligent systems should perceive, interpret, engage with, and adjust to people:

• Sense emotional indicators and cues
• Explain mental states and emotional reasoning
• Respond appropriately in emotionally charged contexts
• Adapt to diverse cultural norms and evolving social signals

While the pilot phase centers on the SER axes, the Adapt dimension remains an integral part of the long-term SERA framework we are building.

Our interdisciplinary team collaboratively identifies, measures, and validates these axes across levels of human-AI interaction.

The project maintains independent oversight. Our aim is to produce a gold standard EQ benchmark through unbiased, interdisciplinary collaboration.

We are now in Phase 3 — Pilot Readiness & Reporting Prep, finalizing the text-only, non-branching SER prompts, weights, and rubric materials while back-porting any needed updates to earlier phases. Full pilot execution and reporting happen in Phase 4, and a prospective Phase 5 will build the multimodal, cross-cultural, and longitudinal benchmark framework that extends beyond the pilot.