Gluing joints crack at the assembly seams – Newspaper Kommersant No. 244 (7445) dated 12/30/2022

By 2030, Russia will have at least five systems for recognizing video fakes in the media and social networks. By 2025, such systems should recognize at least 70% of fakes. The authorities plan to allocate grants for this. Market participants warn that 70% recognition will not be enough to calculate the generated professional content. Domestic video services are not opposed to implementing such software, but if foreign platforms are required to use it, they are unlikely to comply with the requirements, experts say.

The government plans to finance the development of five programs for recognizing generated content (deep fakes) until 2030, follows from the roadmap for the development of the National System-wide Software, adopted by the government on December 16. It is planned to allocate grants for projects, their development will be carried out by the Innovation Assistance Fund. It follows from the document that by 2025 it is necessary to provide a solution that will be able to recognize at least 70% of fake content. Such systems, as follows from the road map, can be used in the media and social networks. The Innovation Promotion Foundation declined to comment. The Ministry of Digital Development did not answer “Kommersant”.

This year, the authorities have already begun to pay attention to the fight against deepfakes. In particular, Roskomnadzor is interested in the development of NRU ITMO in the field of lie detection from video using artificial intelligence and admits that the system can be used to detect “deepfakes”.

The market has long had the problem of detecting “deepfakes”, they say in RecFaces (developing biometric identification programs based on artificial intelligence). But they do not consider the initiative to be implemented, since the format of project support in the “road map” implies the presence of an “anchor” customer, and in this case there is none yet. RecFaces also noted that there are professional and amateur generated content: “If for amateur content, recognition accuracy of 70% may be enough, then for professional content, with actors and video editors, this is not enough, since additional data needs to be analyzed.” According to RecFaces, the development of software that can distinguish between professionally generated content can cost billions of rubles.

The use of “deepfakes” is not regulated now, but it is beginning to actively manifest itself in advertising and video production in general, a Kommersant source in one of the large IT companies notes: “This has already become the subject of litigation, and in the future their number may grow exponentially” . In his opinion, software for recognizing “deepfakes” will be in demand in the courts to confirm the content fabricated with their help.

According to Spartak Khulkhachiev, legal adviser of the EDB intellectual property practice, the state may try to implement the forced installation of such software on Russian video services, YouTube or TikTok services, but the sites are unlikely to take this unambiguously positively.

Alexey Chikov, RuTube business development director, assured that “in the future, the platform would be interested in using deepfake recognition technology, now this is not a priority.” According to him, while the amount of illegitimate content with “deepfakes” in the video hosting is small, and “the moderation system can handle itself.” Rostelecom also admits the possibility of testing such software.

First of all, an accuracy of 70% means that in about every third case the neural network will be wrong, Evgeny Chereshnev, Executive Director for Strategy and Innovative Development of the MTS Cybersecurity Block, emphasizes: “Globally, this will not help fight fakes. In addition, it would be more correct not to create technology from scratch. In the conditions of the current geopolitical situation, we do not have three years to solve the problem.”

There are similar solutions on the market. In particular, VisionLabs solves the problem of determining the “physical” face substitution with fakes in the form of printed and 3D masks, photographs or videos from phone screens, says Dmitry Markov, CEO of the company. According to him, one of the ways cybercriminals use deepfakes can be to quickly create and distribute compromising videos in good quality: “Due to the large amount of generated content and the very good quality of fakes, social media moderators may not be able to track them, and therefore an automatic algorithm is needed. detection.”

Timofey Kornev, Tatyana Isakova