Large language models show fragile cognitive reasoning about human emotions

Researchers introduced CoRE, a benchmark testing whether large language models can reason about human emotions through cognitive dimensions rather than just labels. The study found that while LLMs capture systematic relations between cognitive appraisals and emotions, they show misalignment with human judgments and instability across different contexts.

• →LLMs currently struggle with cognitively meaningful reasoning about human emotions despite being trained on emotion-related tasks.
• →The CoRE benchmark tests LLMs' ability to understand emotions through underlying cognitive appraisal theory rather than surface-level recognition.
• →LLMs show systematic understanding of emotion-cognition relationships but misalign with human judgment patterns.
• →Current emotion AI models demonstrate fragility and inconsistency when contexts change.
• →The research highlights gaps in affective computing that could impact AI's ability to meaningfully engage with human emotions.