A new study claims that humans can’t tell whether a photo is fake or not. A group of researchers from the University of Warwick in England conducted a series of experiments in which they showed participants slightly altered pictures and asked them whether they thought the pics were fake or real.
Thousands of photos are posted online each day. According to estimates, humans take over 1 trillion photos a year, and these pictures are posted on social media, like Facebook, at a rate of 4,000 per second.
As those numbers grow each year, Sophie J. Nightingale, who studies cognitive psychology at the University of Warwick, decided to test how good humans are spotting fake photos from real ones. The researchers found people could only tell photos were altered 60 percent of the time. Their findings were published Monday in the journal Cognitive Research: Principles and Implications.
Humans can’t tell perfectly when a photo is altered or not
The team took different photos from the Internet and modified them by taking something out of the scene, by adding something to it, by changing the shadows on the picture, among some of the alterations.
Afterward, they showed their altered photos to 707 people aged between 14 and 82 years, who served as the population for the new study.
The researchers presented each participant with ten pictures and asked whether they believed the images were altered in some way. They showed participants five altered photos and five real ones. If participants answered the photo was a fake, they were then required to point at the region of the picture that had been altered.
Each volunteer looked each photo for nearly 44 seconds, on average. The researchers noted when participants thought they were seeing a fake photo, it took them an average of 10.5 seconds to identify the areas where they believed the photo had been altered.
The volunteers classified pictures as either real or fake 66 percent of the time, on average. Overall. They were better identifying real photos -72 percent correctly identified these- than fake ones, as only 60 percent correctly identified altered pics.
“It’s a bit worrying,” said Nightingale, according to The Washington Post. “Photos are incredibly powerful. They influence how we see the world. They can even influence our memory of things. If we can’t tell the fakes ones from the real ones, the fakes are going to be powerful, too.”
Volunteers spotted altered photos when manipulations were ‘physically implausible’
The researchers said while the performance could be considered acceptable, it was still “far from perfect.” For instance, people who did correctly identify an alteration only could determine the exact area of the change 45 percent of the time.
“Although subjects’ ability to detect manipulated images was above chance, it was still far from perfect,” wrote the researchers. “Furthermore, even when subjects correctly indicated that a photo had been manipulated, they could not necessarily locate the manipulation.”
According to the researchers, participants were better at spotting that something looked odd in the photos when the alterations were “physically implausible,” like when an object seemed to cast a shadow in the wrong direction. However, volunteers struggled to identify those things that made the picture seem odd.
After the first round of volunteers had finished the experiment, the team called another 659 volunteers. But in that experiment, instead of using pictures that were already on the Internet and compressed into the JPEG format, they took and modified their own pictures and kept them in high-resolution PNG format.
People can spot fake photos but can’t spot where it has been manipulated
For the second experiment, the researchers also told volunteers to point where a photo had been altered, even if they believed the photo was real. In this experiment, volunteers spent an average of about 58 seconds deciding whether a photo had been tampered with, and about 11 seconds deciding where it had been altered. While they spent more time on average, the researchers said they perform slightly better than the first group.
They classified the photos correctly 62 percent of the time (lower than the first group), but they proved to be better to identify altered photos (65 percent) than the originals (58 percent).
The results turned out to be slightly different in both control groups, but the researchers did find a constant between them. Both groups were more likely to notice an image was fake if it had been altered the most. The authors noted it was a surprising discovery, as volunteers didn’t see the originals of the altered photos, just the fakes, but they were still able to tell it wasn’t real.
“People’s ability to detect manipulated photos of real-world scenes is extremely limited,” said the researchers. “Considering the prevalence of manipulated images in the media, on social networking sites, and in other domains, our findings warrant concern about the extent to which people may be frequently fooled in their daily lives.”
The researchers added that more research is needed to find potential ways to improve people’s ability to identify real photos from fake ones.
Source: Los Angeles Times