LO
CapabilityCognitive automation2022-02-14
People identified AI-synthesized faces with 48.2% accuracy, close to chance, and rated them more trustworthy than real faces, a study in PNAS found
Actor and modeloccupation page →
Event date / reported
2022-02-14
Evidence stage
CapabilityA demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.
Tasks this bears on
Fashion and commercial modelling
Posing for campaigns, catalogues, e-commerce and advertisements.
Automating✓ Evidence-backed
Where this applies
Experiments in which participants judged whether faces were real or generated by a synthesis engine. The authors report average accuracy of 48.2%, close to chance performance of 50%, rising to about 59% with training and feedback, and that synthetic faces were rated more trustworthy than real ones. It concerns still face images, not moving performance or full-body modelling, and says nothing about employment.
What this means
A generated face in a still image already passes for a real person — which is what catalogue and e-commerce images need.
What it does not yet show
Still faces from a 2021-era model; moving performance and acting are not tested.
What you can check
Open the PNAS paper "AI-synthesized faces are indistinguishable from real faces and more trustworthy" and find "48.2%".
Does it change the assessment?
No — and this stage does not move it either. A "Capability" record is real evidence, but it does not upgrade a task judgement on its own. The 1 linked judgement above stand where they were.
Source
Nightingale and Farid — AI-synthesized faces are indistinguishable from real faces and more trustworthy, Proceedings of the National Academy of Sciences 119(8) (published online 14 February 2022; full text via Europe PMC, PMC8872790) · verified 2026-09-30 · Claude (VOLO agent) · interpreted 2026-09-30 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.