Recommendation systems: development of legal approaches in Russia and China
Recommendation systems as a sub-technology of artificial intelligence are used by digital platforms outside the framework of states. Legislators around the world are concerned about creating rules to ensure transparency of algorithms and recommendations based on them to protect the rights of users in the digital environment. The transparency of algorithms is becoming a kind of form of government control over content moderators. A cautious approach to the regulation of recommendation systems seems to be very productive. The rules are being created gradually, in conjunction with the norms of public legislation in the field of ensuring citizens' rights to personal data protection. A comparison of the approaches of Russian and Chinese jurists will allow us to find solutions suitable for eliminating uncertainty in the field of application of recommendation systems.
Keywords: artificial intelligence, recommendation system, recommendation algorithm, personal data, protection of the rights of the subject of personal data.
Introduction
At the end of 2021, the Committee on Information Policy, Information Technologies and Communications announced the introduction of a conventionally designated "draft law on the regulation of recommendation services" to the State Duma of the Federal Assembly of the Russian Federation (hereinafter - the State Duma of the Federal Assembly of the Russian Federation) in the spring session <1>. The problems of using recommendation systems are discussed all over the world. In the People's Republic of China (hereinafter referred to as the PRC) on November 16, 2021. At the 20th meeting, the Chinese Cyberspace Administration (CAC), the Ministry of Industry and Informatization of the People's Republic of China, the Ministry of Public Security of the People's Republic of China and the General State Administration for Market Supervision and Regulation reviewed and approved the Regulation on the Management of Algorithmic Recommendations of Information Services on the Internet <2>, which comes into force on March 1, 2022.
--------------------------------
<1>
<2> .
Recommendation systems: basic properties and characteristics
Recommendation systems are essential components of the architecture of online platforms, ensuring the delivery of money-making content in the form of advertising to users. Without algorithmic moderation, the organization of the existing mass of content would simply be impossible. The definition of recommendation systems (RSs) has gained great popularity in the business environment as software tools and methods that provide product offers that can be useful to the user <3>. The suggestions relate to various decision-making processes, for example, which products to buy, which music to listen to, or which news to read on the Internet <4>.
--------------------------------
<3> Mahmood T., Ricci F. Improving Recommender Systems with Adaptive Conversational Strategies // Proceedings of the 20th ACM conference on Hypertext and hypermedia (Torino, Italy, 29 June 2009 - 1 July 2009) / C. Cattuto, G. Ruffo, F. Menczer. New York: Association for Computing Machinery, 2009. P. 73 - 82.
<4> Ricci F., Rokach L., Shapira B. Introduction to Recommender Systems Handbook // Recommender Systems Handbook / eds. by F. Ricci, L. Rokach, B. Shapira, P.B. Kantor. Boston, MA: Springer, 2011. P. 1 - 35.
However, recommendation systems as a subtechnology of artificial intelligence are most fully defined in computer science in the most general form as a class of solutions that ensure the execution of the process without human participation, support in choosing solutions, as well as the prediction of objects that will be of interest to the user ("Roadmap for the development of end-to-end digital technology "Neurotechnology and Artificial Intelligence", Russia, 2019 <5>).
--------------------------------
<5>
The use of recommendation systems by digital platforms significantly affects the ways information is disseminated online today. The technology of filtering user information becomes an important element of the recommendation system and allows you to achieve the most accurate personalization of offers to a specific user.
The essence of the recommendation system as a phenomenon lies in the fact that recommendations are formed separately for each user based on past Internet activity. In addition, the behavior of other users of the system and the mechanisms of social interaction of communities on the Network are taken into account. Recommendation systems collect information about user interaction, which allows you to use implicit user reviews instead of explicit ratings as an expression of user preferences <6>.
--------------------------------
<6> et al. Exploiting the User Social Context to Address Neighborhood Bias in Collaborative Filtering Music Recommender Systems // Information. 2020. Vol. 11. Is. 9. P. 439.
There are two main approaches to building recommendations <7>:
1) based on collaborative filtering, which uses information about user behavior in the past, for example, a list of purchases or evaluations of objects made earlier on the online store site by users from the same interest group <8>;
2) based on content filtering (English Content-based information filtering), based on a holistic approach to profiling user data. The data of each profile is analyzed, including personal information of users: social status, age, place of residence, occupation, as well as characteristics expressing the user's interest in the object; profiles of objects of interest include characteristics of interest to the user <9>.
--------------------------------
<7> Berry M.W. Large-Scale Sparse Singular Value Computations // The International Journal of Supercomputing Applications. 1992. Vol. 6. Is. 1. P. 13 - 49.
<8> Goldberg D., Nichols D., Oki B.M. et al. Using Collaborative Filtering to Weave an Information Tapestry // Communications of the ACM. 1992. Vol. 35. Is. 12. P. 61 - 70.
<9> Joaquin D., Naohiro I., Tomoki U. Content-Based Collaborative Information Filtering: Actively Learning to Classify and Recommend Documents // International Workshop on Cooperative Information Agents. Berlin; Heidelberg: Springer, 1998. P. 206 - 215.
In the process, recommendation systems collect data about users using a combination of explicit and implicit <10> methods, analyzing both user responses to questionnaires and questions about satisfaction with the service, analysis of direct preferences of the subject, and data on user views of certain content on the Network, user behavior online, as well as tracking the contents of the user's computer.
--------------------------------
<10> Adomavicius G., Tuzhilin A. Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions // IEEE Transactions on Knowledge and Data Engineering. 2005. Vol. 17. Is. 6. P. 734 - 749.
Regulation of the use of recommendation systems under Chinese law
According to the Regulation on the management of algorithmic recommendations of information services on the Internet (hereinafter referred to as the Regulation), the emphasis is on algorithmic recommendations <11>. The "application of algorithmic recommendation technologies" <12> refers to actions when, when providing information to users, such types of algorithmic technologies are used as (1) generation and synthesis (of data); (2) personalized offer; (3) sorting; (4) search and filtering (of data); (5) forecasting and the choice of a solution (paragraph 2 of Article 2 of the Regulation).
--------------------------------
<11> In the Russian legal field, the phrase "recommendation algorithms" is more familiar, although from the point of view of literal translation, this Provision focuses precisely on how rules are created-recommendations that affect the user. In this case, we will stick to the literal translation.
<12> Algorithmic Recommendation Technology.
The rules for recommendation systems in Chinese law are introduced to regulate the use of algorithmic recommendation technology in information services on the Internet, promote basic socialist values, ensure national security and social public interests, protect the legitimate rights and interests of citizens, legal entities and other organizations, as well as promote the healthy and orderly development of information services on the Internet. A rather broad approach has been chosen to define the boundaries of the application of criteria for recommendation algorithms, which in the Chinese act were called "algorithmic recommendation service" <13>. Recommendation algorithms should be created taking into account ethical guidelines (social, business, professional ethics), as well as follow the principles of impartiality, fairness, openness and transparency, scientific rationality, honesty and integrity (Article 4 of the Regulations).
--------------------------------
<13> The Algorithmic recommendation service appears in Article 2 of the Regulation as an abbreviated statement (provision of services on the Internet using the use of recommendation algorithm technologies). That is, it is understood as a "recommendation on algorithms", which is well known to users in China due to the large number of abuses by Chinese companies. And the direct regulation of "algorithms", in the opinion of the authors, on the one hand, is the Chinese legislator's answer to acute social issues, on the other hand, due to the fact that algorithms are the basic technology of the recommendation system, this approach makes it more convenient to open a technical "black box" in the implementation of legal regulation. In addition, from the point of view of the authors, "algorithmic recommendation" in everyday life should be understood as a synonym for "recommendation system".
These rules are addressed to service providers of algorithmic recommendations, the general legal status of which is not separately formulated in the Regulation.
The norms enshrined in the Regulation in force on the territory of the People's Republic of China take precedence over the rules contained in other sources, including laws and administrative regulations (Article 2 of the Regulation). At the same time, some provisions of the Regulation are partially based on other acts, such as the Law of the People's Republic of China "On Cybersecurity" <14>, the Law of the People's Republic of China "On Data Security" <15>, the Law of the People's Republic of China "On the Protection of Personal Information" (PIPL) <16>, which entered into force on November 1, 2021, and Measures to regulate information services on the Internet <17>.
--------------------------------
<14>
<15>
<16> Personal Information Protection Law of the People's Republic of China.
<17> .
The Regulation emphasizes that beyond the direct installations of the platform, operators of recommendation systems should be guided by the principles of self-regulation, industry standards, improve service specifications, provide services in accordance with the law and under public control. Thus, the emphasis is on self-regulation within the framework of ethical and legislative norms, as well as the requirements of Chinese society.
The Regulation contains several sections regulating various areas of creation and application of recommendation systems. From the point of view of technical requirements for information services and by virtue of Article 10 of the Regulation, it is established that service providers of algorithmic recommendations must manage user models and user tags (including giving the user the opportunity to select or delete tags related to their personal characteristics, at their discretion, art. 17), improve the rules for determining geolocation and recommendations of so-called points of interest (POIs) for user visits, and should not record illegal and unwanted information keywords in user points of interest or as user tags.
Also, within the meaning of Article 12 of the Regulation, the service provider of algorithmic recommendations has the obligation to "optimize" the transparency and interpretability of rules, such as search, sorting, selection, presentation and display of content, in order to avoid the adverse impact of undesirable content on users.
--------------------------------
<18> The so-called push technology is literally indicated (English push letters. "pushing through").
The use of the term "optimize transparency and interpretability of rules" for recommendation algorithms is not chosen by chance in this context. We are talking about two responsibilities of the service provider of recommendation algorithms - to optimize the interpretability of algorithms and to optimize the transparency of algorithms. The complex optimization of these two responsibilities reflects a complex balance of values of various factors: on the one hand, it is (1) respect for the right of users to information and (2) related rights and legitimate interests of consumers or legitimate interests in the administrative process; on the other hand, these are values such as technological innovation, business viability, protection intellectual property and even national competitiveness. The requirements for promoting openness and transparency of recommendation algorithms are also disclosed in the "Guiding Opinion of the Chinese Cyberspace Administration (CAC) <19> on Strengthening Comprehensive Management of the rules of information Service Algorithms on the Internet" (Internet Information Service Algorithms Regulations) <20> dated September 29, 2021.
--------------------------------
<19> Together with the Propaganda Department of the CPC Central Committee, the Ministry of Education of the People's Republic of China, the Ministry of Science and Technology of the People's Republic of China, the Ministry of Industry and Information Technology of the People's Republic of China, the Ministry of Public Security of the People's Republic of China, the Ministry of Culture and Tourism of the People's Republic of China, the Main State Administration for Market Control and Regulation of the People's Republic of China and the General Directorate for Radio and Television Affairs.
<20> Notification of the release of "Guiding Opinions on strengthening comprehensive rule management of information Services algorithms on the Internet."
A separate section of the Regulation focuses on protecting the rights of users. It is established that the provider of algorithmic recommendations services must clearly inform users about the provision of algorithmic recommendations services and properly publish the basics, purpose and main working mechanism of the algorithmic recommendations service (Article 16 of the Regulation). In art . 17 Provisions separately emphasize that recommendation systems cannot base their proposals on the characteristics of the user's personality, which makes it possible to establish a non-discriminatory mechanism for user access to various types of information services and products. Users should also be provided with convenient options for disabling algorithmic recommendation services. If the user decides to close the algorithmic Recommendation service, the algorithmic recommendation service provider must immediately stop providing these services. Users of many popular mobile apps in China, such as Toutiao, Douyin (, Tik-tok), Kuaishou, Ele.me , Taobao and Meituan Waimai , at the moment, the function of disabling personalized recommendations is already available.
Special rules have been established to create recommendations for minors and senior citizens.
Article 21 of the Regulation requires service providers of algorithmic recommendations that sell goods or provide services to consumers to respect consumers' right to fair trade and not to use algorithms in relation to terms of transactions that may lead to discrimination and differential definition of terms of the transaction depending on the characteristics of consumers. It should be emphasized that these rules correlate with art. 24 of the Law on the Protection of Personal Information, according to which if the operator of personal information uses personal data for automatic decision-making, he must ensure transparency of decision-making and the fairness of their results. This rule is aimed at avoiding unjustified discrimination in matters of trade, for example, determining the price or terms of purchase of goods based on the financial situation of the user, evaluated by algorithms.
Following the general trend, this issue has also been addressed in other regulatory documents of China. For example, in art. 21 Regulations on Unfair Competition on the Internet (Draft for public discussion) <21>, developed by the General State Administration of the People's Republic of China for Market Supervision and Regulation, provides that an economic entity should not use data, algorithms and other technical means based on the collection and analysis of information about the counterparty's transactions, the content viewed by him and the number of committed the number of visits, as well as the brands and cost of the equipment used by them when concluding transactions, it is unreasonable to provide the counterparty with different information on the same transaction. This article also states that information about transactions includes the history of transactions, the willingness of a potential client to pay, his consumer habits, individual preferences, solvency, credit status, etc. This article describes the characteristics of "cheating familiar customers with big data" (<22>), as well as the composition of such an action. In addition, the above criteria can be considered as a clarification of the "characteristics of the user's personality" mentioned earlier in the Regulation.
--------------------------------
<21>
<22> This is a well-known term in China to describe price discrimination. The literal translation is "to kill" acquaintances, which means "to charge old customers more than new ones."
Development of legal policy in the field of application of recommendation systems in Russia
Currently, the legal relationship between digital platforms and their users is regulated by civil law. This is usually done through user agreements, general terms and conditions of digital platforms. These are any terms, conditions or specifications, regardless of their name or form, that govern the contractual relationship between the information intermediary and users. Recently, codes of ethics and state standards have been added to regulators in the digital sphere <23> (for example, GOST R 59276-2020 "Artificial intelligence systems. Ways to ensure trust. General provisions"). The Code of Ethics for Artificial Intelligence, adopted by digital companies in Russia in 2021, includes ethical and technical community standards that have reached a high level of detail on communication platforms.
--------------------------------
<23> Kharitonova Y.S., Savina V.S., Pagnini F. Bias of artificial intelligence algorithms: issues of ethics and law // Bulletin of the Perm University. Legal sciences. 2021. N 53. pp. 488 - 515.
According to the current acts in Russia, recommendation systems and intelligent decision support systems are classified under the section of subcutaneous digital technology (sub-CT) "Artificial intelligence and neurotechnologies" <24> along with technologies of computer vision, natural language processing, speech recognition and synthesis. One of the key tasks of regulatory regulation in this area is linked by the ideologists of digitalization of the economy and public administration with ensuring access to data against the background of maintaining an effective balance between the interests of companies developing and implementing artificial intelligence and the interests of society. However, this is where the consolidation at the regulatory level of issues directly related to recommendation systems ends.
--------------------------------
<24> Roadmap for the development of "end-to-end" digital technology "Neurotechnology and artificial intelligence".
Thus, so far, the development of legislation in this area has followed the path of self-regulation and technical standardization, taking into account the ethical constraints tested in society. The emergence of clear rules at the level of the law can lead to a reduction in uncertainty only if these rules fully meet the expectations of the market and society.
According to statements by representatives of the Committee on Information Policy and Information Technologies of the State Duma of the Federal Assembly of the Russian Federation in the media <25>, the developed bill focuses on transparency of recommendation algorithms, the responsibility of resource owners for identifying and suppressing the use of non-standard issuance algorithms and monitoring the work of recommendation systems by the Federal Service for Supervision of Communications, Information Technology and mass communications. As the experience of the European Union and China shows, strict requirements for content moderation in the law should and can relate to the suppression of the distribution of illegal content. Including in relation to crimes and terrorism, money laundering, targeting minors and socially vulnerable groups of people. However, in the sphere of interaction of most users with digital platforms, the use of private law regulatory mechanisms is preferable.
--------------------------------
<25>
Experts say that the "black box" approach is not applicable in this case, especially since the human equivalent of making recommendations in itself is a transparent social process. So, collaborative filtering is called a social filtering algorithm, because it is modeled on the basis of a time-tested social process of getting recommendations by contacting friends with similar tastes with a request to recommend movies, books or music that they like. The recipient of the recommendation has several ways to decide whether to trust this recommendation: (a) a thorough analysis of the similarity of tastes between the recipient and the recommender; (b) an assessment of the success of previous offers from this recommender; (c) ask the recommender for more information about why the recommendation was made <26>. Based on this thesis, a legal requirement for transparency of the recommendation system can be formulated: a digital platform should offer the user several ways to assess the relevance of recommendations or completely abandon such a function of the platform. This approach will also correspond to the Russian Concept of the development of regulation of relations in the field of artificial intelligence and robotics technologies until 2024, which indicates the need for a full explanation of the decision made by artificial intelligence systems <27>.
--------------------------------
<26> Sinha R., Swearingen K. The Role of Transparency in Recommender Systems // CHI'02 extended abstracts on Human factors in computing systems (Minneapolis, Minnesota USA, April 20 - 25, 2002) / L. Terveen, D. Wixon. New York: Association for Computing Machinery, 2002. P. 830 - 831.
<27> Decree of the Government of the Russian Federation No. 2129-r dated August 19, 2020 "On approval of the Concept for the Development of regulation of relations in the field of artificial intelligence and robotics technologies until 2024".
It is advisable to accompany the introduction of rules on ensuring transparency of recommendation algorithms with requirements for the interface of resources using recommendation systems <28>. The user's behavior when choosing depends on how highlighted or hidden, understandable or incomprehensible certain functions are <29>. This rule, proposed in the EU Digital Services Act, suggests setting general requirements for the interface design specification.
--------------------------------
<28> Hu Y., Ogihara M. NextOne Player: A Music Recommendation System Based on User Behavior // ISMIR. 2011. Vol. 11. P. 103 - 108.
<29> Kuang C., Fabricant R. User Friendly: How the Hidden Rules of Design are Changing the Way We Live, Work and Play. Random House, 2019.
The most important parameters of recommendation systems should be available to the user of the platform in an accessible and understandable form. All the possibilities that can be used to change or influence the most important parameters should also be indicated. This approach, enshrined in the law and in the user agreement, will ensure the autonomy of the user's will.
The introduction of rules on transparency of recommendation systems raises the question of the responsibility of digital platforms for compliance with regulations. The peculiarity of the platform's responsibility as an intermediary lies in the fact that the legislator intends to impose the analysis and suppression of the use of unfair and prohibited practices of recommendations on the platform. Digital platforms will need to analyze and minimize the risks of using algorithms by various subjects of the information space. This position is also taken by China today.
Conclusions
We believe that a cautious approach to the regulation of recommendation systems seems to be very productive. The rules are being created gradually, in conjunction with the norms of public legislation in the field of ensuring citizens' rights to personal data protection. The state's focus on developing targeted solutions to identified and recognized problems will help maintain a balance between the interests of the state, society and business.
The transparency of algorithms is becoming a kind of form of government control over content moderators. Continuing the global trend of delegating state functions to digital platform operators, assigning the responsibility to businesses to identify and suppress the use of recommendation algorithms that contradict established rules will allow them to respond most quickly to emerging deviations. In this case, one should also strive to achieve a balance between innovative algorithmic solutions and explicit and implicit violations of the principles of fair application of recommendations.
