The core and path of algorithmic governance in the era of artificial intelligence

At the end of December 2021, the four departments jointly signed and announced the "Internet Information Service Algorithm Recommendation Management Regulations" (referred to as the "Algorithm Recommendation Regulations"), which has been officially implemented on March 1, 2022. Based on this, it is necessary to further clarify the basic core of algorithmic governance and explore new paths for governance in the algorithmic era.

1. The current legal system for algorithmic governance in my country

The current legislative system for algorithmic governance in my country has been initially established, establishing a legislative system with wide legislative levels and multi-departmental linkage. , a rapidly expanding legal system. Legislative supervision has shifted from the previous focus on network security and data information protection to the current in-depth governance, that is, algorithmic governance in the era of artificial intelligence.

In terms of top-level design, the "Implementation Outline for the Construction of a Legal Society (2020-2025)" proposes to improve standardized management methods for the application of new technologies such as algorithm recommendation and deep forgery. In addition, the "14th Five-Year Plan for Digital Economy Development" points out the acceleration of the construction of a national integrated big data center system with synergy of computing power, algorithms, data, and application resources.

In terms of legal and regulatory basis, the Civil Code, Network Security Law, Data Security Law, Personal Information Protection Law and Internet Information Services Management Measures respectively cover personality rights, network security, data security, Coordinated regulations have been carried out from the perspectives of information protection and utilization, Internet services, etc.

In the direction of specialized regulation of algorithms, there are departmental normative documents "Guiding Opinions on Strengthening the Comprehensive Management of Internet Information Service Algorithms" released in September 2021 and the 2022 "Algorithm Management Regulations" on algorithm-related regulations Comprehensive and detailed specifications were carried out.

In terms of other normative documents or national standards, many departments have indirect regulations on machine learning, artificial intelligence ethics, information synthesis, platform supervision, etc., such as the "State Council Anti-Monopoly Commission on Platform Economic Fields" "Anti-Monopoly Guidelines" "The State Administration for Market Regulation, the Cyberspace Administration of China, the National Development and Reform Commission, the Ministry of Public Security, the Ministry of Human Resources and Social Security, the Ministry of Commerce, and the All-China Federation of Trade Unions on Implementing the Responsibilities of Online Catering Platforms and Effectively Protecting Food Delivery Workers Guiding Opinions on Rights and Interests" "Ethical Code of New Generation Artificial Intelligence" "Regulations on the Management of Deep Synthesis of Internet Information Services (Draft for Comments)" "Specifications for Security Assessment of Machine Learning Algorithms of Information Security Technology (Draft for Comments)" "Information Security Technology Personal Information Safety Specifications" etc.

Although we have legislated at multiple levels, there are still problems in the current algorithm-related legislative system. First, the legislative level is fragmented, focusing mainly on departmental normative documents. The time cost of formulating laws and regulations is significantly higher than that of departmental regulations and various normative documents. This has led to the current emerging issue of algorithms being mainly addressed in departmental normative documents and national standards, which is prone to insufficient enforcement and compromised enforcement and supervision effects. Problems such as unclear division of department responsibilities. At the same time, multi-departmental normative documents have also caused platform companies to be unable to adapt, have inconsistent standards, and require special action-style emergency responses. Second, the supervision of platforms is mainly passive after the fact, and there is a lack of refined platform supervision regulations. The supervision of platforms mainly adopts administrative punishment measures based on the faults, behaviors and responsibilities of the platform. However, this supervision model lacks prior process supervision. Even if there is an algorithm filing system, it mainly stays at the algorithm filing in specific important areas. The algorithm review logic and standards for registration also need to be adjusted in a timely manner according to the algorithm classification system. Third, there is little supervision of the technical specifications of algorithms, and legislation lacks a return to the origins of algorithms. Algorithm is a technical concept, which is a "method of calculation" or "method of processing data". At the same time, the algorithm also has a certain learning ability and can continuously evolve based on the existing algorithm foundation and data. There is still a lack of legislative specifications for the technical specifications of these computer instructions. At present, the main regulations are regulated from the perspectives of network security and legal risks.

2. The core, path and system construction of algorithm governance

In order to promote the improvement of algorithm-related legislative systems and achieve precise governance of algorithms, the author believes that the core of algorithm governance lies in data information security. On the one hand, algorithms are a series of program logic constructed on the basis of natural language, which are essentially logical operations of AND, OR, and NOT. But no matter how complex the algorithm is, its essence is also a "model trained with data", that is, the continuous operation and evolution of the algorithm is achieved by continuously feeding data. Algorithms are inseparable from the support of data. When there is a problem with data processing activities, there will inevitably be problems with the algorithm. Therefore, the essence of paying attention to the governance of algorithms is the security and reasonable processing of data.

On the other hand, legal risks such as "big data killing" caused by automated decision-making algorithms have attracted more and more social attention, which shows that the essence of algorithm governance lies in the reasonable use of information. In addition, the soul of an algorithm lies in its positive values. The use and processing of data information need to pursue positive values ​​and gradually realize algorithms that are verifiable, auditable, supervised, traceable, predictable, and trustworthy, while also being inclusive, fair, and non-discriminatory.

It should be noted that data information security includes two major parts: data security and information security. Data security is to regulate data processing activities, ensure data security, safeguard the interests of all parties, and ensure data development and utilization and industrial development; information security is to The "Personal Information Protection Law" is the main body that regulates personal information processing activities, promotes the rational use of personal information, and strictly protects personal privacy.

Only by clarifying the core of algorithmic governance can we focus on the focus of legislative norms and supervision, and have new solutions to the dilemmas of algorithmic governance in current practice. The development of artificial intelligence and even the entire economy and society is inseparable from the filling of massive data and personal information. Automated decision-making algorithms make full use of data information to exert greater economic and social value. Therefore, the author believes that a "two internal and one external" guarantee path for algorithm governance should be constructed. The two internal elements are to strengthen privacy protection and expand the breadth, depth and accuracy of data, and the first extension guarantee is the algorithm security guarantee mechanism.

First, strengthen privacy protection. The protection of privacy rights in the Civil Code is included in the section on personality rights, which is enough to show the importance of privacy protection. Currently, the privacy policies of major platforms are being adjusted and updated, and this round of updates is bound to bring more restrictions to the disorderly development of algorithms on relevant platforms. Privacy protection and algorithm development are relative. Strengthening privacy protection will inevitably hinder the more diversified development of algorithms. However, it is precisely based on the importance of privacy protection that algorithms can avoid infringing on the legitimate rights and interests of others. Strengthening privacy protection can start from the following points:

First, strengthening privacy protection is reflected in legislative content, algorithm design and application, filing and review, law enforcement supervision focus, legal liability, etc. This is the basis of algorithm governance. The basic concept is also the bottom line principle.

Secondly, it is also very important to strengthen the privacy protection of key groups, especially minors under the age of fourteen, the elderly, workers and consumers. Information protection and data processing shall be carried out in accordance with the relevant provisions on privacy protection in the Civil Code and the relevant provisions on sensitive personal information in the Personal Information Protection Law. Personal information processing activities meet the five important principles of personal information processing, as well as comply with the core personal information processing rules of "inform-inform-consent".

Third, privacy protection disputes are mainly resolved through private law remedies, while privacy protection issues involved in algorithm governance will inevitably require more public law relief channels, so more public law governance algorithms At this time, we need to pay attention to the integration of the traditional attributes of privacy protection and public and private law governance.

The second is to broaden the breadth, depth and accuracy of data. Algorithmic governance is by no means overly emphasizing regulatory penalties, but rather emphasizing prior overall management. Automated decision-making such as deep learning requires the feeding of massive amounts of data. Lack of data volume and inaccurate data will cause the algorithm to calculate in the wrong direction. For example, when an enterprise is conducting "user profiling", when the user data base is small or data in a certain dimension is missing, it is impossible to accurately push relevant information or provide corresponding services. When broadening the breadth, depth and accuracy of data, it must be restricted by legal data processing activities. The key points to deal with this problem are as follows:

First, the process of broadening data is to ensure data security. Security can ensure the safety and stability of the algorithm, which is the cornerstone of data processing.

Second, establish a hierarchical and classified management system for important data and data. The influx of large amounts of data may disrupt the basic order of the algorithm, so the hierarchical classification of data is something that major platforms, especially very large platforms, need to standardize.

Third, establish a verification and error correction mechanism in the algorithm, that is, verify the quality of the data, such as random inspection mechanism, result warning, etc. to discover data defects so that deviations can be corrected in a timely manner.

The third is the algorithm security mechanism. With the first two foundations of privacy protection and data, it is especially important to improve the algorithm security mechanism. The safety guarantee mechanism includes scientific and technological ethics review, legislative guarantee, safety assessment and monitoring, and emergency response to safety incidents, etc., forming multiple guarantees of technology, law, and management. Specific measures include the following:

First, the algorithm is good. The basic point of algorithm registration review is scientific and technological ethics review. The difficulty of this review lies in the unpredictability of the algorithm. Even if the current algorithm rule review is reasonable, as the algorithm itself extends, the results of the algorithm will be inconsistent. Certainty. Therefore, a specialized organization similar to the Algorithm Ethics Working Group should be established, composed of experts in technology, law and other fields, as well as representatives from regulatory authorities and third-party industries, to strengthen regular review and follow-up supervision and strictly prevent problems with algorithmic values.

Second, legislative protection. The current legislative top-level design on algorithm governance has been gradually completed. Next, in addition to algorithm recommendation management, other algorithm activities need to be paid attention to, such as algorithm technology research and development, data mining, rule content, operational support, personnel management and other perspectives. Build a new pattern of algorithmic governance.

Third, improve management systems and technical measures such as safety assessment monitoring and emergency response to safety incidents. For enterprises, it is necessary to implement the main responsibility for algorithm security, guard the first line of defense for algorithm security, and establish and improve algorithm mechanism mechanism review. For regulatory authorities and industry organizations, it is necessary to formulate an industry standard system for algorithm security, promote the basic concepts of algorithm security, and form a multi-channel supervision force for the whole society.

The algorithmic era has had a profound impact on all walks of life, and has also led to new dynamic changes in the current organizational form. Problems caused by algorithm abuse may be huge in the industry. Algorithm governance needs to be carried out simultaneously with algorithm development, effectively building a "two internal and one external" guarantee path for algorithm governance, establishing a good digital business environment, and promoting the steady health of the digital economy and society. develop.

The above is the detailed content of The core and path of algorithmic governance in the era of artificial intelligence. For more information, please follow other related articles on the PHP Chinese website!