Experts explained how to label content created by neural networks

It is possible to label content created by neural networks, according to experts interviewed by Vedomosti.

On the morning of October 26, the head of the State Duma Committee on Information Policy, Alexander Khinshtein, said that legislators were discussing the structure of the “digital code,” according to which it is proposed to include in the law the labeling of content created by artificial intelligence (AI) technology.

According to Alexandra Murzina, head of the advanced technologies department at Positive Technologies, identifying such content will be quite difficult – existing programs for checking the origin of texts and photographic images have not yet become successful. Creators of networks for generating Stable Diffusion images, for example, began adding invisible watermarks to their images. But it is still impossible to say for sure how well this technology works.

The easiest way is to oblige AI services to affix markings, and then for government agencies to monitor compliance with these requirements, says Stanislav Goncharov, head of the Digital Risk Protection department at FACCT. However, none of the existing solutions provides 100% reliability.

“I think that the solution to the issue may lie in the plane of a legislative ban on the use of materials created by AI without mentioning this fact, with some kind of sanction if such use is detected,” says Yaroslav Shitsle, head of the It&Ip Dispute Resolution department at the law firm Rustam Kurmaev and Partners.

In his opinion, existing recognition software is becoming increasingly useless with each new stage of AI development. Media lawyer Svetlana Kuzevanova believes that no one can oblige a person to mark his authorship or the fact that he used a program to create content, and imposing a sanction on this would be “the most strange initiative.”