The era of "strict supervision of algorithms" is coming: the new regulations of the four ministries and commissions refer to the platform "big data killing, two choose one"

In the era of big data, algorithms have become hubs connecting people and the virtual world. The irregular use of algorithms has also led to problems such as algorithm discrimination, information cocooning, big data killing, rice circle culture, and inducing addiction.

On January 4, the State Internet Information Office, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation jointly issued the Provisions on the Recommendation and Administration of Internet Information Service Algorithms (hereinafter referred to as the Provisions), which came into effect on March 1, 2022.

It is worth noting that the algorithm recommendation service has been stripped separately from the Internet application to formulate a separate management specification for it, which is the first time in China.

Wang Yuan, a partner at Beijing Gaoqin Law Firm, told the China Times reporter: "The algorithm determines the content of the information to a certain extent, determines what information the user can see, and will have an impact on public opinion and values." Therefore, it is necessary for the algorithm to be stripped out separately for specification. Over the next 3 years, algorithms will be a regulated focus. ”

"The algorithm is stripped out separately to regulate, that is, the objective needs of the standardized algorithm recommendation industry, and it is also consistent with foreign practice." Deng Ruohan, a researcher at the Digital Economy and Rule of Law Research Center of Southwest University of Political Science and Law, told this reporter.

In addition, the Provisions also make corresponding provisions on issues such as "big data killing, inducing addiction, and choosing one of the two".

Why is the algorithm stripped separately and regulated?

It is the first time in China that the algorithm has been stripped out separately for standardization.

It is understood that the current algorithm recommendation technology refers to the use of algorithm technologies such as generating synthesis classes, personalized push classes, sorting selection classes, retrieval filtering classes, and scheduling decision classes to provide information to users. But the nature of the algorithm itself leads to a series of problems.

The algorithm pushes a large amount of homogeneous content to the user according to the user's interest, and the user can only obtain limited sources and single tendency information for a long time, resulting in the user being tied to the "information cocoon"; and the imperfection of the algorithm mechanism leads the takeaway rider to fall into the dilemma of "food delivery can only be faster and faster, more and more disregard for their own safety" under the "persecution" of the algorithm; and the rise of the "rice cooker test" is to expose the algorithm discrimination on the e-commerce platform.

Deng Ruohan said: "Algorithmic systems have a certain degree of independence because of their many characteristics at the level of formation, operation and renewal. ”

He pointed out to reporters from the technical level: "The cultivation of machine learning algorithms uses more 'designer input training data', Therefore, even the designer of the algorithm can only grasp the data of the input algorithm, but cannot grasp the results of the algorithm learning; secondly, the algorithm usually updates itself according to the new information generated by the market and the data it enters in the process of operation, thereby ensuring its adaptability to changes in the market environment, but this also makes the update and development of the algorithm to a certain extent not under the control of the algorithm designer and operator. The existence of the algorithm black box makes the designer and operator of the algorithm only grasp the input data and output results of the algorithm, and the specific analysis and decision-making process is difficult for the designer and operator to know. ”

"If the algorithm is not regulated separately, and corresponding regulatory measures and risk prevention rules are designed for the characteristics of the algorithm system, then the regulatory model that only restricts the algorithm operating enterprises is more likely to fall into the dilemma of 'fighting cattle from the mountains', that is, enterprises not only do not know how to constrain the algorithm, but also tend to blame the problems caused by the algorithm on 'technical failure', thereby exempting themselves from their own management responsibilities." Deng Ruohan pointed out.

In addition, Deng Ruohan also believes that "the problems of information cocooning and thinking induction caused by algorithm recommendation have gradually affected the public opinion, culture and values of the entire public, and then have certain public attributes, which is also an important reason why the regulatory authorities should directly regulate the algorithm." ”

It is forbidden to "kill big data, induce addiction, choose one of the two" and so on

As a product of the Internet era, the double-edged nature of the algorithm has brought positive and negative influences, and how to use the strengths to avoid the shortcomings and exert its positive effects has become the focus.

At present, China has made relevant top-level designs for network security, data security, personal information security, etc.

In the "Implementation Outline for the Construction of a Rule of Law Society (2020-2025)" issued in 2020, it is proposed to formulate and improve the normative management measures for the application of new technologies such as algorithm recommendation and deep forgery. Subsequently, laws such as the Cybersecurity Law, the Data Security Law, the Personal Information Protection Law, and the Guiding Opinions on Strengthening the Comprehensive Governance of Internet Information Service Algorithms were successively promulgated.

The "Specification" has formulated detailed provisions from the aspects of information service specifications and user rights and interests protection to avoid the negative effects of algorithms.

The relevant person in charge of the State Internet Information Office said: "The promulgation of the Provisions is mainly based on the following two considerations: First, the need to further promote the comprehensive governance of Internet information service algorithms. The second is the need to actively promote the healthy development of algorithm recommendation service specifications. ”

In response to the issue of "choosing one of the two", article 16 of the Provisions clarifies that algorithm recommendation service providers must not use algorithms to unreasonably restrict other Internet information service providers, or obstruct or disrupt the normal operation of Internet information services they lawfully provide, and implement monopoly and unfair competition.

The issue of "big data killing" is also clearly stipulated in Article 21, where algorithms recommend service providers to sell goods or provide services to consumers, they shall protect consumers' rights to fair trade, and must not use algorithms to implement unreasonable differential treatment and other illegal acts on transaction conditions such as transaction prices based on consumer preferences, trading habits and other characteristics.

In addition, for the provision of algorithm recommendation services to minors, the elderly, workers and consumers, articles 18, 19 and 20 of the Provisions clarify specific requirements, such as not using algorithm recommendation services to induce minors to indulge in the Internet, the elderly shall be facilitated to safely use algorithm recommendation services, and relevant algorithms such as platform order distribution, remuneration composition and payment, working hours, rewards and punishments shall be established and improved, and must not be based on consumer preferences. Characteristics such as trading habits use algorithms to implement unreasonable differential treatment on trading conditions such as trading prices.

The algorithm-driven industry will usher in a reshuffle

"The promulgation of the Provisions means that the relevant industries of algorithm recommendation have begun to enter the era of 'strict supervision of algorithms'." Deng Ruohan said.

Algorithm-driven companies are bound to face cost increases and compressed profits. He said: "The algorithm security main responsibility, algorithm evaluation and verification mechanism, information security management mechanism and other algorithm management mechanisms proposed in the Provisions, as well as the obligations of algorithm interpretation and the obligation to provide natural search results, will certainly greatly increase the compliance costs of algorithm-driven companies." The curbing of problems such as big data maturity and information cocooning has also limited the profit space of algorithm-driven enterprises to a certain extent. ”

"Enterprises are shifting from data-driven to algorithm-driven, and it's not how much data is mastered, but whose algorithms are more advanced and effective." Wang Yuan pointed out to this reporter the current situation of such enterprises.

How to find new and standardized profit growth points has also become a difficult problem facing algorithm-driven companies.

Deng Ruohan believes that "it is foreseeable that related companies will inevitably usher in 'pain', during which enterprises with weak capital foundations, immature profit models and imperfect algorithm systems may face the possibility of being eliminated." ”

Deng Ruohan also added: "The promulgation of the Provisions is only the first step out of the arithmetic regulatory system, in which some mechanisms such as algorithm evaluation and verification, algorithm interpretation and other mechanisms have vague provisions. To give full play to the institutional efficacy of the Provisions, standardization will inevitably become an urgent problem to be solved in the next step. For example, what the 'basic principles' of algorithm recommendation services include, to what extent they should be explained, and what methods the 'appropriate methods' include, etc., all need to be further improved. ”

Li Keshun, president of guizhou data treasure product research institute, told the "China Times" reporter: "Market supervision needs to be effective, and the entire operation process and data of the enterprise must be sufficiently transparent. The technical diversity of algorithm design, the complexity of the calculation process, and the poor explanatory performance of some algorithms cannot be sufficiently transparent, which will affect the effectiveness of market supervision. ”

However, the reporter learned that at present, there are no corresponding standards and implementation guidelines for algorithms in China. Therefore, the era of "arithmetic system" has arrived, but there is still a long way to go.

Responsible Editor: Xu Yunqian Editor-in-Chief: Gong Peijia