The Faces of Generative AI: Predictors of FACES

Area of Science:

• Psychology
• Artificial Intelligence
• Medical Imaging

Background:

• The Facial Appearance as Core Expression Scales (FACES) was developed to evaluate maxillofacial surgery patients' self-perception.
• This study utilized FACES to assess participant ratings of their own faces and those generated by DALL·E, a generative AI.

Purpose of the Study:

• To validate the FACES scale's ability to differentiate between facial images generated by AI using positive versus negative descriptive prompts.
• To explore the relationship between self-esteem and facial self-perception.

Main Methods:

• 16 photorealistic faces were generated using DALL·E, varying in age, gender, ethnicity, and descriptive prompts (positive/negative).
• 333 participants rated these AI-generated faces and their own faces using the FACES scale.
• Rosenberg Self-Esteem and State Self Esteem Scale scores were collected.

Main Results:

• FACES successfully distinguished between faces generated with positive and negative instructions for all tested image pairs.
• Participant self-esteem scores correlated with their ratings of their own faces.
• DALL·E could not accurately generate realistic maxillofacial anomalies.

Conclusions:

• The FACES scale is a viable tool for differentiating AI-generated facial aesthetics based on prompt sentiment.
• Self-esteem is a significant factor in how individuals perceive their own facial appearance.
• Current generative AI limitations exist in depicting specific medical conditions like maxillofacial anomalies.