The geospatial fraud was as rampant as it was frivolous, but a bigger danger was lurking.
Bo Zhao, a geographer focused on geographic information systems (GIS), was working on his PhD at the Ohio State University in the early 2010s when he began noticing that people on Twitter were using geotags in their tweets to lie about their locations. Some, it seemed, were trying to fool people into thinking they were witness to global news events, but others appeared to be lying just for the ease of it.
Then in 2016, when the augmented reality smartphone gamePokémon Gobecame a global phenomenon, Zhao saw that people were using virtual private network (VPN) connections to fake their geolocation in the game in order to access the rare Pokémon targets pegged to real locations around the world. Essentially cheating, the trick also had an element of evening out the playing map. “The company actually distributes the Pokémon very unevenly. Some places have more, some places less,” says Zhao,now an assistant professor华盛顿大学地理系。“这实际上是避开这种失衡的一种方式。”
That led him to wonder what other more important geospatial data was being spoofed.
Zhao and colleagues from Oregon State University and Binghamton University began to look into satellite imagery, a major source of geospatial data used in applications ranging from climate observation to global shipping. In a recent paper, they explore the potential—and, as they show, the very real threat—of people using artificial intelligence to create convincing but fabricated satellite imagery. Like AI systems that have been created to产生现实的面孔或者malicious pornographers who’ve used cruder systems to makefake explicit videos赵和他的同事使用名人的相似性，表明也可以制作深击卫星图像。
Their paper, published recently in the journal制图和地理信息科学，从西雅图和北京使用的地图和卫星图像创建了华盛顿塔科马市内一个社区的真实但假的卫星图像。在图像中，塔科马道路出现在其准确的位置，但低层西雅图或高层北京的建筑物已被交换。从表面上看，这些图像似乎代表了真实的地方。
But Zhao and his colleagues aren’t just opening a Pandora’s box of potential geospatial deepfakery. They’ve also begun to develop an approach to help identify when satellite images are actually AI creations. Zhao says it could take the form of an application programming interface that geographers can use in conjunction with geographic information systems likeArcGIS。
Faked satellite images aren’t necessarily problematic. The approach could also be used for good, Zhao says. For example, an AI-generated image could be used to estimate and fill in missing data from long-term climate change observations, helping to provide information that can inform future projections. In an urban development context, an AI-generated satellite image could envision how a city or region would change over time if development were left to sprawl uncontrolled.
赵说，无论是好，坏还是欺骗它们Pokémon Go，伪造的地理空间数据应该是依靠这些数据的任何人的警钟。一些计算机科学家之间的推动make artificial intelligence more accessible对于公众来说，可能会导致人们像伪造名人视频一样轻松伪造地图数据。
赵说：“制图中有一个说法：地图不是领土。”“这主要是由Mapmaker提出的主观论点。因此，这实际上取决于谁做到了，以及出于什么特定目的。而且，随着使用AI的标准确实很低，我们知道事物背后的真正意图将是一个挑战。”
Zhao呼吁他在计算机科学和地理世界中的同事对他们所依赖的数据更加谨慎，并开发一组伦理，通过这些道德规范创建和使用地理空间数据。他说：“我试图鼓励人们不要在没有质疑或批评的情况下使用地理空间数据。”“人们需要意识到，地理空间数据可能被滥用或可能出于恶意目的而产生。”