DALL-E-2 AI-Generated Images of Infants and Toddlers are Distinguishable from Real Photos

Images generated by artificial intelligence (AI) are becoming more realistic, with AI-generated images of adult faces indistinguishable from human faces. Can the same be said for AI-generated images of young children? We compared 76 participants’ discrimination of 72 images representing three types of stimuli: real photos, randomly selected AI images, and AI images curated with human oversight. Real photos were identified as real more frequently (78.6%) than Curated AI (40.3%) and Random AI (29.1%). However, Curated AI stimuli were judged as real more often than Random AI stimuli. Performance variability between images based on age, gender, and race suggests that AI better portrays children with specific characteristics. Images of toddlers and girls were judged as real more than infants and boys. Additional refinement is needed in training and testing text-to-image algorithms to realistically portray children and promote the ethical use of AI in developmental research.