Evaluating the performance of general purpose large language models in identifying human facial emotions
We evaluated the ability of three leading LLMs (GPT-4o, Gemini 2.0 Experimental, and Claude 3.5 Sonnet) to recognize human facial expression using the NimStim dataset. GPT and Gemini matched or exceeded human performance, especially for calm/neutral and surprise. All models showed strong agreement with ground truth, though fear was often misclassified. Findings underscore the growing socioemotional competence of LLMs and their potential for healthcare applications.
Authors
- John Blake Torous (ORCID: https://orcid.org/0000-0002-5362-7937)
- Ari Winbush (ORCID: https://orcid.org/0000-0003-4183-731X)
- Nicholas B. Allen (ORCID: https://orcid.org/0000-0002-1086-6639)
- Benjamin W. Nelson (ORCID: https://orcid.org/0000-0002-5474-7674)
- Matthew Flathers (ORCID: https://orcid.org/0009-0004-3351-0688)
- Steven Siddals (ORCID: https://orcid.org/0009-0007-0514-8978)
Institutions
- Beth Israel Deaconess Medical Center (US)
- Harvard University (US)
- University of Oregon (US)
- University of La Verne (US)
Publication Details
- Journal
- npj Digital Medicine
- Published
- 2025-10-16
- DOI
- https://doi.org/10.1038/s41746-025-01985-5
- Citations
- 3
- Primary Topic
- Face recognition and analysis
- Type
- article
- Field-Weighted Citation Impact
- 3.83