When we talk about deepfakes, possibly the first thing that comes to mind is a video or audio recording made or modified by artificial intelligence (AI) , but which is characterized by a very specific characteristic: it is originally false content, that the AI makes it deeply credible.
Deepfakes come from a technique invented by researcher Ian Goodfellow in 2014, known as GAN (Generative Adverarial Networks), by means of which two algorithms train each other, in such a way that one tries to make the fakes the same. most reliable possible; and, the other, is “in charge” of trying to detect counterfeits.
In this way, both algorithms improve together over time through their corresponding training, so the greater the number of samples available, the greater the improvement in those samples .
Since 2017, the year in which the deepfake phenomenon was officially born, appearing for the first time on the Reddit website, its number has increased considerably, reaching, in 2019, around 15,000 deepfake videos, when the previous year they reached little less than 8,000 videos. But its use can go much further.
A group of researchers from the University of Washington have detailed that it is possible to generate deepfake satellite images through the use of artificial intelligence . Although they can be very useful to, for example, understand what a certain place was like in the past, or how a certain location could be transformed if some changes were carried out, satellite images created through deepfakes could be used for the opposite ; and misinformation could be one of the most dangerous targets.
Precisely, for the production of these false images, the North American research team used GAN networks, by means of a relatively easy-to-use type of AI, which is commonly used to create false faces.
As the researchers warn and alert, the use of deepfakes in satellite images can end up being very dangerous, since when viewing this type of snapshot it is very common to think that they are real, since, it is assumed, they are photographs of the place, and not they correspond to interpretations carried out by a cartographer. But deepfakes have shown that it doesn’t have to be this way.
And they go further: they fear that if an intruder manages to access satellite images of military equipment, it is possible that they could add false structures in order to trap the enemy. They could even hide important buildings or structures from the enemy, preventing their destruction, by making people believe that they do not exist when in fact they do.
Now, through reverse engineering, they hope to create tools that can easily detect when a satellite image has been modified using AI .
