The Liar's Dividend: What Fake Images Actually Broke

In March 2023, an image of Pope Francis in a brilliant white puffer jacket went viral. It was generated with Midjourney, it fooled an enormous number of people, and it was harmless — the internet had a laugh and learned a small lesson. Two months later, a synthetic image of an explosion near the Pentagon circulated on social media, was amplified by blue-checked accounts posing as news outlets, and briefly moved U.S. financial markets before officials confirmed nothing had happened. Same technology, same laundering pattern — an unattributed image presented as breaking news — but this time with a measurable real-world cost. Those two images, weeks apart, mark the moment synthetic media stopped being a curiosity.

The obvious harm is deception: people believing things that did not happen. Election-season fabrications of politicians in compromising scenes, fake "evidence" injected into court disputes and insurance claims, fabricated product photos and reviews, and — by volume the most damaging category — non-consensual synthetic imagery, which overwhelmingly targets women and causes harm that persists regardless of later debunking. Generative tools collapsed the cost of producing convincing lies from a VFX studio's budget to a free afternoon, and volume matters: fact-checkers debunk at human speed while generation runs at machine speed.

But the deeper harm runs in the opposite direction, and it has a name: the liar's dividend. When the public knows that any image can be fabricated, anyone caught on camera gains a new defense — "that's AI." Authentic footage of real misconduct becomes deniable. The flood of fakes does not just insert false things into the record; it corrodes the evidentiary value of everything real. A public that can be convinced of anything eventually believes nothing, and universal disbelief serves the powerful far better than universal gullibility ever could. That asymmetry is why I find "never trust your eyes anymore" — a common, understandable reaction — to be exactly the wrong lesson. Reflexive disbelief is not media literacy; it is the liar's dividend paid in full.

Institutions are responding, unevenly. The EU's AI Act requires AI-generated content to be labeled as such. Several U.S. states have legislated against deepfakes in elections and intimate imagery, platforms attach AI-content labels of varying diligence, and the C2PA provenance standard is slowly giving authentic media a verifiable paper trail. All of this helps and none of it suffices, because labels and laws bind the compliant, and the compliant were never the problem.

Which leaves the uncomfortable conclusion that some of the defense has to live in individuals. Not because every citizen should forensically analyze every image — nobody has time — but because the realistic goal is calibrated skepticism: knowing which claims warrant a pause, recognizing the laundering patterns (single source, no corroboration, emotional urgency, conveniently poor quality), and knowing the basic checks before sharing. Calibration is a skill, and skills are trainable. That belief is the entire reason ImageGuesser exists as a game rather than an essay: reading about detection produces agreement, but attempting detection daily — and being wrong, publicly, against a score — produces ability. Our players' aggregate statistics show measurable improvement with practice, which is the most optimistic data I can offer about this whole subject.

The ethics module goes deeper into the legal landscape and the responsibility that comes with detection skills — including the duty not to weaponize "that's fake" against authentic evidence. Because the endgame of media literacy is not cynicism about images. It is keeping the distinction between real and fake legible at all, for as long as that remains possible — and it remains possible considerably longer if millions of people can each spot the seams.