Public Law Aspects of Technical Regulation of Artificial Intelligence in Russia and the World.

Atabekov Atabek Rustamovich PhD in Economics Associate Professor of the Department of Administrative and Financial Law, Legal Institute, Peoples' Friendship University of Russia 117198, Russia, Moscow, Miklukho-Maklaya str., 6 atabekoff1@mail.ru Other publications by this author

The development of the technological potential of artificial intelligence is one of the significant tasks for both the Russian Federation ^[1-2] and the international community ^[3].

At the same time, it is impossible not to note the emerging issues of positioning artificial intelligence (AI) in the context of technical regulation and the position of both national institutions in the field of standardization and public authorities in this area.

Basic and fundamental work in the field of developing unified approaches to standardization of AI is carried out at the level of international standardization bodies - ISO/IEC (TC1/SC42) and their various joint subcommittees. Within the framework of the functioning of this TC, a wide list of standards in the field of AI has already been published ^[4-6].

At the moment, the use of AI for the purposes of public authorities is recommended by ISO to be guided by ISO/IEC 38507:2022, which prescribes the basic principles, as well as the potential negative consequences of the use of AI:

1. The risk of monopolization of the market due to the industry specialization of AI.

2. Bias in judgments when processing AI data without thorough verification.

3. The risk of loss of human control, taking into account the speed of AI learning.

4. Discrimination and potential harm to the fundamental rights of employees who can be replaced by AI.

5. Reputational costs in the absence of AI control.

Specialized authorities are being created at the level of various states, relevant strategic documents are being developed ^[7].

Thus, the US National Strategic Plan defines the need to increase the availability of AI test benches and subsequent work to increase indicators related to the development of standards and the involvement of the professional community ^[8]. In the following years, the American National Institute was represented by the Secretary in the joint Technical Committee 1/sub-committee 42 formed by ISO and IEC ^[9]. In addition, understanding the military potential of AI, the US National Security Commission on Artificial Intelligence (NSCAI), focused the attention of national authorities involved in the development of standards, the need to systematically improve the reliability of the results and functioning AI models ^[10].

In China, the relevant work is organized and coordinated by the Ministry of Industry and Information Technology, which published a White Paper on AI Standardization in 2018 ^[11]. At the same time, China's initiatives to bring together approaches in the field of standardization ^[12-13], for cloud computing, industrial software does not cancel the principles of sovereign regulation of AI. Within the framework of the national standardization strategy called "Chinese Standards 2035", it is recorded that national standards will differ from international standards, in order to develop domestic industry ^[14].

At the same time, the activities of bodies in the field of standardization are distributed among 5 organizations ^[15]:

1. The National Technical Committee for Standardization of Information Technologies (SAC/TC 28) is the basic body that deals with terminology in the field of AI and related fields related to AI and human interaction.

2. China National Technical Committee for Standardization of Automation and Integration Systems specializes in the standardization of industrial robots.

3. The National Technical Committee for Standardization of Audio, Video and Multimedia specializes in standards for smart homes.

4. The National Technical Committee for Standardization of Information Security specializes in security standards for smart cities and industrial production.

5. The National Technical Committee on Intelligent Transport Systems specializes in standards in the field of intelligent transport.

The White Paper on Standardization of China also highlights fundamental problems in the field of standardization and certification of AI:

1. Heterogeneity of AI technical development at the level of different countries gives rise to different interpretations of AI at the national and supranational levels.

2. Multicomponence and cross-border universality of technology can generate an excessive or duplicative contour of standards.

3. Standardization as a way of lobbying for the interests of international corporations and associations.

4. An unlimited number of users of products, which leads to difficulties in coordinating the standardization process.

5. Security and ethics issues are the most difficult category in the field of standardization, which slow down the work on the main block of AI standardization.

In the European Union, the fundamental standardization bodies are the European Committee for Standardization and the European Committee for Electrotechnical Standardization (CENELEC) and the European Telecommunications Standards Institute (ETSI). In 2019, CENELEC created an AI Task Force ^[16], after which a roadmap for AI standardization was published in 2020 ^[17]. ETSI has also created various focus groups on artificial intelligence ^[18].

It is also impossible not to mention the initiative of the EU legislative initiative regarding artificial intelligence, which has a risk-oriented approach and lays its fundamental foundations, including for the purposes of standardization and regulation by the authorities (Regulation) ^[19].

This regulation defines a horizontal approach to AI (unlike other EU legislation on product safety), which indicates a comprehensive consideration of the issue of AI regulation, as well as minimizing risks at the level of cross-border movement of voluntary national and international standards.

At the same time, the AI products themselves are graded by the Regulations depending on the degree of risk to society, with their subsequent restrictions and obligations (Articles 5, 8, 52 and 69 of the Regulations) in the field of ensuring the safety and reliability of the technology.

At the same time, Decision No. 768/2008/EC of the European Parliament and of the Council of July 9, 2008 on the general framework for the sale of products ^[20], contains provisions that the specification of local standards in the field of AI is the target task of standardization institutes (CENELEC, ETSI, etc.). At the same time, the conformity assessment of products is a duty only the manufacturer, which is also reflected in relation to AI (Articles 16,17, 51, 60 86 of the Regulations). In addition, article 71 of the Regulations fixes the amount of a negotiable administrative fine in the amount of 2%, 4% or 6% of the violator's annual turnover.

The risk-oriented approach proposed by the EC in relation to AI has an interesting approach in the field of standardization, however, it retains issues in the field of AI verification by the manufacturer of products, and not by third parties, and these institutions are not public entities, but NGOs.

In Russia, the standardization of activities is entrusted to Rosstandart, which has prepared, together with the Ministry of Economic Development, a standardization program for the priority area "Artificial Intelligence" for the period 2021-2024 ^[21]. The program contains more than 70 industry metrological standards.

At the same time, an interesting phenomenon is both the discrepancy at the level of GOST R 43.0.8-2017 and GOST R 59277-2020 documents regarding the definition of AI, and in terms of the practical implementation of standardization in the field of AI according to the above program. According to this program, 44 intersectoral standards in the field of AI are planned by 2024 (11 in 2022, 21 in 2023, 12 in 2024), 58 industry metrological standards (21 in 2022, 19 in 2023, 18 in 2024), 91 accompanied international document (standard) on AI (20 in 2021, 22 in 2022, 24 in 2023, 25 in 2024). The industry specialization of these documents covers almost all spheres of human activity.

Rosstandart, guided by Article 9 of Federal Law No. 162-FZ of 29.06.2015 "On Standardization in the Russian Federation" (the Law) implements state policy in the field of standardization, determines the procedure for achieving a unified position of different parties in the development of national standards, the procedure for their examination, organizes the development of standards, etc. At the same time, according to paragraph 25 of Article 9 of the Law, it is possible to make a decision on the creation and liquidation of technical committees for standardization, which implies the creation on a voluntary basis of interested legal entities, as well as authorities participate in the development, examination of documents in the field of standardization in the areas assigned by it (Article 11 of the Law).

In the field of AI, TC 164 "Artificial Intelligence" acts as such ^[22]. At the same time, it should be noted that both scientific organizations (GOST R 59898-2021, GOST R 58776-2019, etc.) and commercial companies with a high degree of technical involvement (GOST R 59391-2021), as well as companies whose competence raises potential questions (GOST R 70249-2022, GOST R 59385-2021).

According to the author, it should be noted that the basic general problem of standardization issues is that, directly or indirectly, the field of technical regulation of AI is influenced by non-governmental organizations with the participation of interested companies that lay down their competitive advantages and subsequently weaken state control, including in relation to such technology as AI, which has a whole the spectrum of unpredictable consequences for society ^[23-27].

The second problem is the potential duplication and divergence of positions regarding the regulation of AI, taking into account its multicomponent base, self-learning system and the possibility of its deployment on any data sample (military, medical, banking, government, etc.).

The third fundamental problem is the issue of competencies in the field of developing AI standards for the needs of the state segment, which remains in a vulnerable position when issuing a standard by private law organizations.

The specified contour of problems, according to the author, can be solved by introducing a number of initiatives at the level of authorities, namely by forming administrative barriers, in order to ensure conditions for the development of standards for the needs of public bodies exclusively by state scientific institutions, as well as the formation of a legislative initiative for additional verification of the standards proposed by TC 164 by the relevant authority responsible for the development and AI regulation, both in order to minimize cross-border risks of excessive AI regulation and duplication of regulatory provisions presented in the standards.