Even top face-spotters get tricked by AI images, study shows
A new study found that even "super-recognizers"—people who are amazing at remembering faces—struggle to tell real faces from AI-generated ones.
Researchers tested participants using images made by generative adversarial networks (GANs) and discovered that these super-skilled folks were fooled almost as often as everyone else.
Super-recognizers vs. AI: Not as foolproof as you'd think
Super-recognizers only spotted fake faces with 41% accuracy, while regular participants scored even lower—all below random guessing.
The catch? AI faces look super realistic because their features are so perfectly balanced.
Quick training helps, but it's not magic
After a five-minute crash course on common AI mistakes—like weird teeth or funky hairlines—super-recognizers improved to 64% accuracy, and typical folks reached 51%.
The boost came from being more alert to subtle flaws, not just guessing differently.
Why this matters for all of us
If even the best face-detectors can be tricked by AI, spotting fake content online is harder than we thought.
Plus, taking extra time to check means there's a trade-off between speed and getting it right—a heads-up for anyone scrolling through social feeds or checking profile pics.