When thinking of deep fakes, we tend to imagine people generated by AI. This can be as playful as deepfake Tom Cruise, or as malicious as porn with disagreements. Us Do not Imagine deep fake geography. This is an AI-generated urban landscape and countryside image. But that’s exactly what some researchers worry about.
In particular, geographers are concerned about the proliferation of fake satellite images generated by artificial intelligence. Such a picture can be misleading in many ways. It can be used to create tricks for wildfires or floods, or to offend articles based on real satellite images. (Think of a report on China’s Uighur camps that have earned trust from satellite evidence. As geographic deep fakes are widespread, the Chinese government may also argue that these images are fake.) Deep fakes are used by geopolitical enemies, so deep fakes can be used. Fake geography can also be a national security issue. Satellite images that mislead the enemy.
The U.S. military has warned of this very prospect in 2019. Todd Myers, an analyst at the National Geospatial-Intelligence Agency, envisioned a scenario where military planning software was tricked by fake data showing a bridge in the wrong position. “So, from a tactical point of view or mission planning, you train your army to take a specific path towards the bridge, but it’s not there. Then a big surprise awaits you,” Myers said.
The first step towards solving these problems is to let people know that you have a problem in the first place, says Bo Zhao, assistant professor of geography at the University of Washington. Zhao and his colleagues recently published a paper on the subject of “Deep Fake Geography,” which includes self-experiments of creating and detecting this image.
Goal, Zhao The Verge “To understand the absolute reliability features of satellite images and to raise public awareness of the potential impact of fake geography,” via email. He says deep fakes are widely discussed in other fields, but his thesis will be the first to be addressed in geography.
“A lot of GIS [geographic information system] Practitioners have celebrated the technological advantages of deep learning and other types of AI for solving geographic problems, and few have publicly recognized or criticized the potential threat of deepfakes to geography or beyond.” I wrote.
Far from presenting deep fakes as new challenges, Zhao and his colleagues find this technique in a long history of fake geography that goes back thousands of years. Humans, they say, have been lying on the map for almost as long as the map exists. They speak from mythological geography devised by ancient civilizations such as Babylonia to modern propaganda maps distributed “to shake the morale of the enemy” during the war.
A particularly interesting example comes from the so-called’Paper Village’ and’Trap Street’. These are fake settlements and roads inserted into maps by cartographers to catch rivals stealing their work. If you have anyone making a map that includes Fakesville, Ohio, you can know and prove that they are copying your cartography.
New technology creates new challenges, but Zhao of pseudo-geography says, “This is a phenomenon from centuries ago. “The deep fake satellite imagery is so real that it’s partly novel. Untrained eyes will easily think they are real.”
It is definitely easier to create fake satellite images than fake human videos. The lower resolution is convincing and the satellite imagery is inherently reliable as the medium. This could be due to what we know about the cost and source of this photo, Zhao says. “Most satellite images are produced by experts or governments, so the public generally wants to believe that this is real.”
As part of the study, Zhao and his colleagues created software to generate deep fake satellite images using the same basic AI methods (a technique known as generative adversarial networks or GANs) used by well-known programs such as ThisPersonDoesNotExist.com. Then I created detection software that could detect fakes based on characteristics such as texture, contrast, and color. However, as experts have warned over the years of people’s deepfakes, all detection tools need constant updates to keep up with the improvements in deepfake generation.
But the most important thing for Zhao is to raise awareness so that geographers are not vigilant. He and his colleagues wrote: “If we continue to not recognize that we are not ready for deepfakes, we risk getting into a’fake geography’ dystopia.”