The faces staring out from your screen may never have existed. AI image generators have grown so good at inventing human faces that people struggle to distinguish them from photographs — and may even rate the synthetic ones as more trustworthy, a phenomenon researchers call "hyperrealism," Scientific American reports, drawing on work by psychologists at the Australian National University.

Why the fakes fool us

The trick lies in how the technology works — and in how our brains work. AI generators learn from millions of images and tend to drift toward the statistical average, producing faces that are unusually symmetrical, well-proportioned and emotionally neutral. Those are precisely the qualities people instinctively find appealing and trustworthy: we are wired to favor "average" faces. The result is a face that can feel more real, and more reassuring, than a genuine one with all its quirks.

What to look for

Researchers have identified some recurring tells, though none is foolproof:

  • Too much symmetry. Real faces are subtly lopsided; AI faces are often suspiciously balanced.
  • A blank quality. Synthetic faces can lack the small signs of mood, fatigue or character that animate real ones.
  • Forgettable, generic features. The memorable oddities — a crooked nose, an unusual eye, a scar — are frequently absent.
  • Glitches at the edges. Older systems left obvious errors (impossibly smooth skin, mismatched ears, oddly placed eyes), and even now problems crop up with hair around glasses, jewelry, hands, and reflections.
  • Lighting that doesn't add up. Shadows and highlights on the face may not match the light in the background.

The catch

Here is the hard part: these cues are a moving target. As the technology improves — and it improves constantly — the visible flaws disappear, leaving anyone hunting for "wonky ears" chasing a problem that the next software update has already fixed. Researchers describe it as an arms race that the human eye is losing. Tellingly, one study found that training people to read broad patterns, rather than to spot single glitches, nearly doubled their accuracy — but even that edge erodes over time.

The real defense

Because the eye alone is becoming an unreliable judge, the more durable fixes lie elsewhere: digital watermarking and "content provenance" systems that label an image's origin at the source, so a viewer doesn't have to play detective. Major technology companies are rolling out such content-credential schemes. Until they are universal, the best protection is old-fashioned skepticism — be wary of faces attached to high-stakes claims, cross-check against trusted sources, and run a reverse image search when something feels off. The unsettling bottom line from the science is simple: seeing is no longer believing.