Generative AI emotion recognition from bodily gestures and vocal tone reveals modality-specific performance and positivity bias
David Piterman, Hadar Dery, Kfir Bar, Gunther Meinlschmidt, Alon Geller, Dorit Hadar Shoval, Zohar Elyoseph, Elad Refoua · 2026
Read Generative AI emotion recognition from bodily gestures and vocal tone reveals modality-specific performance and positivity bias by David Piterman, Hadar Dery, Kfir Bar, Gunther Meinlschmidt, Alon Geller, Dorit Hadar Shoval, Zohar Elyoseph, Elad Refoua (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13542163.txt.
Description
David Piterman and seven co-authors test whether generative AI can read feelings from the body and the voice, in this 2026 study. Mental health tools are eyeing these models, but a chatbot that misreads distress is a hazard. The team pitted three Gemini models against validated human benchmarks on a standard set of acted gestures and vocal tones expressing basic emotions.
On vocal tone, the advanced models scored near human accuracy. On bodily gesture, they fell significantly behind people. A skew ran through both channels: positive emotions were recognized more accurately than negative ones, significantly so for gestures in every model and for voice in the advanced ones. The resulting profile is not humanlike at all: strong on cheer, weak on pain. The authors warn against rushing such systems into sensitive care roles.
Catalog record
| CREATOR | David Piterman, Hadar Dery, Kfir Bar, Gunther Meinlschmidt, Alon Geller, Dorit Hadar Shoval, Zohar Elyoseph, Elad Refoua |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |