We aren't very good at spotting fake photos
Add Axios as your preferred source to
see more of our stories on Google.
Sophie Nightingale/University of Warwick
Fake photos abound and, unfortunately, we don't seem to be particularly good at spotting them: about one-third of doctored photos went undetected by participants in a new study.
Why it matters: Photos — admitted as evidence in courts, used to bolster positions in campaigns and to diagnose conditions — can shape our memories and beliefs. They're being altered in sophisticated ways that, if undetected, could have profound consequences.
"Pretty much everybody is susceptible to falling for fake images, which was surprising," says lead author Sophie Nightingale from the University of Warwick.
What they did: In two experiments, Nightingale asked more than 1300 people total whether 10 photos were altered or originals. The images of real-world scenes were manipulated in two different ways: physically implausible manipulations like shadows from two sources, and plausible ones like whitening teeth or adding trash cans to the picture.
- On average, people could classify the photos correctly 60% of the time in the first experiment (and 65% in the second) — slightly better than the odds of a 50/50 guess.
- Even when they recognized something was wrong with an image, participants in the first study could correctly identify what had been manipulated only 45% of the time on average.
- Test yourself here.
What's next: The researchers are looking into whether people can be trained to better spot forged images, Nightingale told Axios. For example, when forgers change an image they may leave tell tale signs in the shadows. Computers can look at the position of light and find what has been manipulated. The researchers want to know if people can be taught to do this, too.
Go deeper: Altering images can be laborious (Nightingale notes the study included only 10 manipulated images because of the time and skill required to doctor them) but soon computers will likely be creating fakes — and detecting them.
