Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

AI Faces the Criteria for Being Real; A Behavioral and Pupillary Reactivity Study.

Poster Session E - Monday, March 9, 2026, 2:30 – 4:30 pm PDT, Fairview/Kitsilano Ballroom

Jessica Samir¹ (), Madison Dawes², Kayla Challacombe³, Karlie Souder⁴, Carole Scherling⁵; ¹Belmont University, Nashville, TN

Technological advances in artificial intelligence (AI) are challenging our perceptions of real vs. artificial, extending to facial affect. Generative AI is increasingly capable of creating images based on emotion prompts (Sivasathiya, 2024); one study found that AI facial images passed the uncanny valley test and were deemed to be truly human (Reuten, 2018). However, little systematic work has validated the emotional quality of AI-generated images. Sixty undergraduates (mostly female) completed a happy and sad face-viewing paradigm, exposed to 12 real faces (FACES) and 12 AI-generated faces (the latter piloted in 14 undergraduates). Each face was randomly presented in full-screen color and outcomes measures included pupillary reactivity (IMotions) and judgements of valence and approachability (sliding 100-point Likert). After eliminating the first 2 trials as practice, preliminary 2x2 ANOVAs did not reveal pupillary differences between the real and AI faces for happy and sad affect (p>0.05). Similarly, no behavioral differences were revealed between real and AI faces (p>0.05), with all happy faces being recorded as more positively-valanced and approachable compared to sad (p<0.05). Overall, real and AI-generated faces showed expected judgement ratings, aligning with both the valence and approach-withdrawal hypotheses. The lack of physiological differences in cognitive engagement (pupil reactivity) between the 2 face types indicates similar affective analyses. Overall, current AI can generate facial images that accurately depict emotional qualities, at least for happy and sad faces. Future work must investigate other basic and learned emotions. AI-generated images may accurately depict basic emotions but fall short for self-referential and/or mixed emotions.

Topic Area: EMOTION & SOCIAL: Person perception