Physix FrontierHuman Eyes Aren't Blind, Just Untrained on Where to Look
People aren't unable to spot AI faces; no one has taught them where to look. I recently came across news about a paper on "training humans to identify AI-generated faces," stating that just a few minutes of focused practice can make people better at distinguishing fake faces. That was my immediate reaction.
I resonate strongly with this viewpoint. When I first joined the lab, looking at generative model images on arXiv (the preprint server), I always felt "they all look real." Later, I discovered that if you focus on hair edges, behind the ears, teeth, and background blur, the flaws become apparent. Reading papers is the same—it's not about memorizing terminology, but having a reusable observation path.
The evidence isn't complex either. Related research mentions that people who undergo brief training become significantly more stable when judging real vs. generated faces; traditional detection also looks at facial edges, blink rhythm, voice anomalies, and other details. Put simply, AI faces don't "fool everyone," but rather fool those who haven't been alerted.
When I tried it myself, the biggest gain wasn't "I became stronger," but realizing that human judgment is often stuck in default settings. If you ask someone to look at a face, they look at the overall vibe; if you ask them to find fakes, they don't know where to start. Give them specific anchors, and things change immediately.
Instead of believing "human eyes can naturally distinguish truth from falsehood," we should admit that human eyes need calibration.
So the takeaway from this study for me is simple: preventing AI fake faces can't rely solely on stacking detection models. Writing down "how to look" into steps that ordinary people can practice might be more urgent than releasing another algorithm version. Going forward, even paper novices can learn something here: don't just ask if the model is accurate; first ask if humans have been taught properly.
📌 This article is compiled from Hacker News. Original source: https://www.southampton.ac.uk/news/2026/09/researchers-help-people-detect-ai-faces.page
Copyright belongs to the original authors. This is a compilation and independent analysis based on public reports.
