Too beautiful to be fake: Attractive faces are less likely to be judged as artificially generated

Dominique Makowski; An Shu Te; Ana Neves; Stephanie Kirk; Ngoi Zi Liang; Panagiotis Mavros; S H Annabel Chen

doi:10.1016/j.actpsy.2024.104670

Too beautiful to be fake: Attractive faces are less likely to be judged as artificially generated

Acta Psychol (Amst). 2024 Dec 19:252:104670. doi: 10.1016/j.actpsy.2024.104670. Online ahead of print.

Authors

Dominique Makowski¹, An Shu Te², Ana Neves³, Stephanie Kirk², Ngoi Zi Liang², Panagiotis Mavros⁴, S H Annabel Chen⁵

Affiliations

¹ School of Psychology, University of Sussex, United Kingdom of Great Britain and Northern Ireland; Sussex Centre for Consciousness Science, University of Sussex, United Kingdom of Great Britain and Northern Ireland. Electronic address: D.Makowski@sussex.ac.uk.
² School of Social Sciences, Nanyang Technological University, Singapore.
³ School of Psychology, University of Sussex, United Kingdom of Great Britain and Northern Ireland.
⁴ Department of Economics and Social Sciences, Télécom Paris, France.
⁵ School of Social Sciences, Nanyang Technological University, Singapore; LKC Medicine, Nanyang Technological University, Singapore; National Institute of Education, Singapore; Centre for Research and Development in Learning, Nanyang Technological University, Singapore.

PMID: 39705943
DOI: 10.1016/j.actpsy.2024.104670

Abstract

Technological advances render the distinction between artificial (e.g., computer-generated faces) and real stimuli increasingly difficult, yet the factors driving our beliefs regarding the nature of ambiguous stimuli remain largely unknown. In this study, 150 participants rated 109 pictures of faces on 4 characteristics (attractiveness, beauty, trustworthiness, familiarity). The stimuli were then presented again with the new information that some of them were AI-generated, and participants had to rate each image according to whether they believed them to be real or fake. Despite all images being pictures of real faces from the same database, most participants did indeed rate a large portion of them as 'fake' (often with high confidence), with strong intra- and inter-individual variability. Our results suggest a gender-dependent role of attractiveness on reality judgements, with faces rated as more attractive being classified as more real. We also report links between reality beliefs tendencies and dispositional traits such as narcissism and paranoid ideation.

Keywords: AI-generated images; Attractiveness; Fake news; Fiction; Sense of reality.