ANU Researchers Train Humans to Spot AI Faces With Near-Perfect Accuracy in 45-Person Study
Updated
Updated · Gizmodo · Jun 29
ANU Researchers Train Humans to Spot AI Faces With Near-Perfect Accuracy in 45-Person Study
3 articles · Updated · Gizmodo · Jun 29
Summary
A 45-participant ANU study reported that training people on broad facial cues nearly doubled their accuracy in identifying AI-generated faces, with the best performers reaching near-perfect results.
Six training blocks of 96 tasks taught participants to judge global impressions—symmetry, proportionality, attractiveness, expressiveness, distinctiveness and memorability—rather than hunt for obvious glitches.
The method builds on a finding that AI faces tend to look more average: more symmetrical, well-proportioned and attractive, but less distinctive, memorable and expressive than real human faces.
That approach aims to fill gaps left by commercial AI detectors, which can produce false positives and opaque judgments, as deepfakes have grown about 900% annually from 2023 to 2025.
Researchers said the online training could be deployed more widely, though they cautioned the results cover still-image face generators only, not audio or video deepfakes.
If our brains are trained to spot AI fakes, will our perception of human beauty and trust fundamentally change?
As deepfake video scams surge, can human intuition be trained fast enough to become our best defense?
Human Training Breakthrough: How People Can Now Accurately Detect AI-Generated Faces (2026 Report)
Overview
As AI image generators rapidly advance, creating faces that look increasingly real, it has become harder for people to tell the difference between genuine and AI-generated faces. Recent research from the Australian National University reveals a breakthrough: humans can be trained to accurately spot these synthetic faces. This is possible because AI, while aiming for realism, often produces faces that are more symmetrical and average than real human faces, missing subtle imperfections. Training people to notice these differences offers a crucial way to defend against the growing challenge of distinguishing real from fake images in our digital world.