Do not believe your eyes: how to recognize a deepfake

Deepfake technologies are improving so rapidly that for several years now they have been the object of close attention not only of large companies, but also government agencies responsible for digital security. Can an ordinary person now distinguish a fake from reality and how to do it, Moskovskaya Pravda learned from experts.

We have more or less learned to recognize fake text messages or the use of digital technologies when changing photos, although this also requires certain knowledge and skills. But what to do with videos that seem to be quite plausible, but their content raises some doubts? How to find out if this is a real recording or a video created using artificial intelligence? Questions are not idle these days.

The technology for creating deepfakes was previously used in the overwhelming majority of cases for entertainment purposes and caused nothing but a smile and admiration for the development of technology. Then deepfakes began to be more actively used in the creation of films and TV shows – despite its laboriousness, it fully justifies itself. One of the clearest examples of a deepfake in Russian cinema is the appearance in the 2022 sequel of the Saboteur series by Vladislav Galkin, who starred in the first parts of the film, but passed away in 2010. Deepfakes also help filmmakers “age” or “rejuvenate” actors, dub and translate films.

Meanwhile, as is usually the case with inventions, deepfakes have become used and not at all for good purposes. Thus, for the purpose of illegal profit, scammers began to mislead people by creating and distributing videos on social networks in which famous people advertised certain services, held lotteries, and even simply distributed money to everyone. Do I need to say that the “stars” in the commercials were created using deepfake technology? But until this became known, the scammers managed to “warm up their hands” well. So, according to experts, the losses of the world economy from such technologies have already amounted to about 400 billion rubles.

Moreover, the improvement of technology now allows the use of deepfakes to form public opinion about global political and economic events. For example, recently Russian President Vladimir Putin was shown a deepfake by German Chancellor Olaf Scholz at the Applied AI Technologies exhibition. Of course, according to the “statements” of the fake Scholz, it was clear that this was not a real video, but in general everything looked quite convincing.

On the development and improvement of programs that can recognize deepfakes, they are now working all over the world, including in Russia. One of them, by the way, was shown to the Russian president at the AI ​​technology exhibition. But the possibility of mass use of such programs is still far away. So is it possible to recognize a deepfake without technical devices?

It used to be easy enough to do this. As one of the methods, it was proposed to track how often the person depicted in the video blinks. On average, a person blinks once every 5 seconds. And if the character in the video blinks much less often, it is likely that this is a deepfake. However, now technologies allow us to improve the facial expressions of a deepfake, including normalizing the frequency of its blinking. But, according to Dmitry Kovalev, head of the information security department at Sissoft, careful viewing of the video allows you to “calculate” the deepfake even now. True, special attention should be paid in this case not to the person, but to the background on which he is shot.

– In some situations, without special technologies, it is really difficult for the viewer to find out what is in front of him: a real video or a fake. But more often than not, a deepfake can be recognized. It is given out by a blurry low-quality picture with poor light: the details in the frame are unevenly lit, the shadows and color temperature differ in different parts of the video, the frame trembles, and the background does not always keep up with human movements,” Dmitry Kovalev lists the main distinguishing features of deepfakes.

Sergey Gataullin, an expert in the field of digital technologies and mass communications, agrees that it is difficult to identify high-quality video content as a deepfake without the use of special software, but it is possible:

“Given that deepfakes are inserted into the video frame by frame, when viewing it, even with the naked eye, fluctuations and temporal distortions are noticeable, which are not taken into account by many deepfake detection algorithms that use information from one frame to find a fake out of context with neighboring frames.

For those who want to recognize a fake for sure, Sergey Gataullin suggests using special Internet services that use deepfake recognition technologies and cope with this task with sufficient accuracy.

As for the spread of deepfakes on television, according to the expert, this is more of a legal issue than a technological one. For example, the law can oblige TV companies to mark deepfakes with a special sign.

“However, this is not the only moral and legal question that needs to be answered in the development of the digital economy. The Metaverse is rather a reality already today, and video content is just one of its components,” Sergei Gataullin believes.

In general, experts advise to be critical of video content that is designed for a wide resonance or aimed at evoking certain emotions in viewers or encouraging them to take certain actions. As in the case of other types of fakes, it is worth checking information from other sources, preferably those that you trust. Then the likelihood of becoming a victim of fraud will be much less.

Alena Bodrienko.

Photo from the site itcrumbs.ru