Deepfake geography: the risks of fake maps

Artificial intelligence specialists used this technology to create fake satellite images; they warn that the results can be harmful but also offer benefits. Researchers from the University of Washington, in the United States, used artificial intelligence (AI) techniques to create fake satellite images and, in this way, advance systems capable of detecting adulterated maps. This method coincides with the procedures of deepfakes, terminology that refers to a lie or deep falsification; in practice, images and/or videos altered with AI software that are usually used to misinform or deceive (although their objectives are not solely harmful).

It should be noted that deepfakes have a much deeper realism than a simple digitally retouched image, for example with Photoshop software.

According to the specialists, with the aforementioned technology they managed to add buildings, vegetation and structures to different maps. For example, they applied landscapes of Tacoma and buildings of Beijing to a map of Seattle, and used a similar technique by exchanging features of the Spanish cities of Cadiz, Madrid and Lugo.

With the more sophisticated AI technologies that we now have, “fake geography” appears to be a growing problem, warned the researchers who, we emphasize, designed falsifications to better understand these maps and design a system capable of identifying adulterations. As they indicate in a statement, the intention is to “advance in a system of verification of geographic data.”

Falsified geography: dangers and opportunities

“It is not just about ‘photoshopping’. It makes the data look incredibly realistic,” said Bo Zhao, assistant professor of geography at the University of Washington and lead author of the study that was published in the journal Cartography and Geographic Information Science. “The techniques are already there. We are just trying to expose the possibility of using the same techniques and the need to develop a resistance strategy,” he noted.

Experts said that map falsification has been used as a strategy since ancient times, in addition to unintentionally transcribed inaccuracies. In addition, sometimes this non-real data is used to avoid copyright infringements. According to them, this move is known as “paper towns”, in reference to cities, mountains, rivers and other false features that are placed on a map discreetly to avoid copyright sanctions.

In the 21st century, digital tools have injected new challenges into this field. “Location impersonation requires much greater sophistication and carries risks,” say experts from the University of Washington. Along these lines, the US National Geospatial Intelligence Agency pointed out that satellite images manipulated with AI can be a serious threat to security.

Specifically, the study led by Zhao seeks to create an AI method that learns the characteristics of a satellite image, then generates a deepfake, and finally compares both data and finds the fake data. “This is similar to how popular image filters can do it by mapping the features of a human face onto a cat,” they said, adding that to identify fakes they focused on technical aspects of image processing, such as color histograms and spatial and frequency domains.

Now, do fake maps only bring risks? According to Zhao, these deepfakes can offer certain benefits. For example, fully understanding the expansion of certain areas or climate change. “There may be a location for which there are no images for a certain period of time in the past, or in predicting the future, so creating new images based on existing ones, and clearly identifying them as simulations, could fill in the gaps and help provide perspective,” he said.

“As technology continues to evolve, this study aims to foster a more holistic understanding of geographic information and data so that we can demystify the question of the absolute reliability of satellite imagery or other geospatial data,” the study leader concluded. “We also want to develop more forward-looking thinking in order to take countermeasures such as data verification when necessary,” he said.