Dataset Scale and Societal Consistency Mediate Facial Impression Bias in Vision-Language AI

Open in new window