- Статьи
- Economy
- The sign above: website owners have been freed from control over the labeling of AI content
The sign above: website owners have been freed from control over the labeling of AI content
Online resources will not control the labeling of content created using neural networks. This obligation was contained in the draft law on the use of artificial intelligence, but disappeared from the new version of the document (Izvestia has reviewed it). The previous one stated that online platforms should check for labeling, and in case of its absence, delete such content. But there are no tools to massively and accurately identify the use of AI, experts said. Removing the verification requirement will reduce the burden on businesses, according to the government.
New procedure for monitoring AI content
The provision that online resources will be required to check the labeling of AI-generated content posted on them has disappeared from the new version of the draft law on the use of artificial intelligence. According to the previous version of the document, sites (for example, social networks) where copyrighted content is published had to check whether photos, texts, and videos of users created using neural networks were labeled, and if they were not labeled, delete them.
This provision has attracted criticism from the business community. The Chamber of Commerce and Industry (CCI), a number of industry communities and large companies in their responses to the bill drew attention to the fact that it is technically impossible to fulfill this requirement. In addition, there were concerns that the business would suffer losses, in particular, in the event of fines for non—compliance with the law.
"The measure was softened following a discussion of the bill with the business community," Deputy Prime Minister Dmitry Grigorenko's office told Izvestia. — According to the current version of the document, the responsibility for machine-readable labeling of audiovisual content generated using AI lies with the owners of AI services (neural networks. — Izvestia), and specific cases of mandatory labeling will be determined by the government.
The Ministry of Finance adheres to the approach that the law on the use of AI should be of a framework nature, the agency told Izvestia. And the specifics of regulation should be fixed at the level of industry legislation, taking into account the specifics of the relevant areas: transport, medicine, education, and so on.
In fact, the author of a particular product or work is now responsible for the presence or absence of AI labeling, the ANO Digital Economy said.
— Labeling it is in the interests of the authors. This is a matter of ethics and reputation: honest information about which content is unique and created by humans, and which is generated by AI, increases users' trust in it, makes the audience more loyal, and contributes to its growth.
The labeling verification requirement was originally formulated in such a way that a social network or marketplace was required to determine whether each video, photo, or text was generated by artificial intelligence or not. But it is impossible to do this: nowhere in the world are there tools capable of determining whether AI has been used in the creation of content en masse, according to the Digital Economy. And manual moderation would require enormous expenses and labor.
The purpose of the changes in the bill is to reduce the burden on business, a source close to the government told Izvestia.
Do I need to label all AI content?
AI is one of the means of creating content, a tool, said Irina Levova, Director of Strategic Projects at the Internet Research Institute. Now, for example, no one labels music played on a synthesizer rather than a violin, she argues.
— It is pointless to spend a lot of money on defining a way to create content that is not necessarily bad or good in itself. The bill still requires platforms to provide users with the opportunity to report that AI was used in its creation. This model encourages a normal responsible user approach," Irina Levova believes.
Various scenarios for the use of AI technologies are used on intermediary digital platforms, from product card design and analytics to recommendation systems and anti-fraud tools, noted in RWB (Wildberries & Russ). The requirement to constantly inform people about each such case can significantly complicate the user experience and reduce satisfaction from interacting with services in the absence of obvious practical value for the consumer, they believe.
The rapid development of artificial intelligence technologies has led to their solid integration into both everyday life and business processes of companies, the RWB emphasized.
The company does not see the need to impose excessive requirements, including the mandatory use of trusted AI models at non-public sector facilities and significant critical information infrastructure facilities. Such measures may create unreasonable barriers to the functioning of already implemented solutions and, in general, slow down the development of AI technologies, RWB believes. A similar approach is debatable in terms of mandatory labeling of all content created using AI, they added.
For content authors, the change in the draft law means reducing legal and organizational risks: they will not have to fear automatic deletion of materials just because there is no mention of using AI, said Leonid Drobyshevich, NTI technology expert.
Reliable technologies for mass recognition of AI content do not exist now: detectors regularly make mistakes, falsely trigger and work poorly on recycled materials, short texts, mixed or edited content, he listed. As a result, mandatory labeling without working verification mechanisms would create a formal obligation rather than a realistic requirement.
— Besides, it would inevitably give rise to a large number of controversial situations. For example, is it considered AI content for a text that a person wrote himself, but corrected with the help of a neural network? Or a video where AI was used only for noise reduction and editing? Without clear technological criteria, platforms would either be forced to moderate with a margin, removing questionable materials, or massively ignore violations. Both scenarios create risks, for example, excessive censorship and user dissatisfaction," Leonid Drobyshevich pointed out.
According to the expert, the new version of the bill looks like an attempt to adapt regulation to the real state of technology and the market.
Currently, the draft law on the use of artificial intelligence is at the stage of interdepartmental coordination, a source familiar with the situation told Izvestia. According to him, it is planned that the document will be submitted to the State Duma before the end of the current spring session, that is, until mid-July.
Переведено сервисом «Яндекс Переводчик»
