Evaluates AI's performance in emotion recognition, finding collective human intelligence exceeds AI, suggesting synergy between human and machine intelligence.
Key Points
The aim is to compare GPT-5 mini's performance on emotion recognition tasks against that of human participants and to explore collective intelligence effects.
Compared GPT-5 mini with human responses on the RMET and MRMET tasks.
Aggregated independent human responses using bootstrap-resampled plurality voting.
Investigated the performance of combined human and AI responses.
GPT-5 mini outperforms individual human participants on both RMET and MRMET.
When aggregated, human responses significantly surpass those of GPT-5 mini.
An augmented intelligence approach combining human and AI responses outperformed both independently.