Use mainstream value-oriented control algorithms
Lu Xiaohua
"Use the mainstream value orientation to control the'algorithm'" is a requirement clearly put forward by General Secretary Xi Jinping.
Algorithm is a strategic mechanism for describing and solving problems in a systematic way based on data. It is widely used in many fields of economy and social life, and iterates continuously based on accumulated data, becoming a basic mechanism that affects information distribution, service provision, opportunity allocation, and resource allocation. and strength.
At present, it is necessary to grasp the development trend of artificial intelligence, to use it well, and to manage it effectively, to effectively prevent hidden risks caused by the abuse of algorithms, and to achieve mainstream value-oriented control of algorithms by strengthening the comprehensive management of algorithms.
Strengthening the comprehensive management of algorithms is a must-answer question
Since the 21st century, algorithm technology has developed rapidly, and automated decision-making has increasingly gained greater influence and dominance in economic and social life.
Internet platforms gain a competitive advantage by using algorithms to improve the matching of information services with user needs. A single platform even has hundreds of millions of daily active users.
China's huge user base provides a solid foundation for algorithm development.
Supported by the massive increase in user data, the algorithm continues to iterate, and the level of intelligence is gradually improving.
"In front of algorithms, people have become tools?" Sharp questions such as this resonated strongly.
The recommendation algorithm pushes similar information to make users tired, and people are increasingly worried about the various "information cocoons" and ethical issues that the algorithm recommendation service will give birth to.
More importantly, algorithm recommendation services have an increasingly dominant role in social life, public cognition and ideology, and must be included in the scope of comprehensive government governance.
From the perspective of public interest and national security, the recommendation algorithm must not be allowed to determine the information acquisition and value formation of young people, and the recommendation algorithm must not be allowed to determine the information distribution that leads to an imbalance of ideological security, and must not be caused by inadequate legal and technical supervision capabilities. Internet information algorithm recommendation services bring risks at critical moments.
It can be said that strengthening the comprehensive management of recommendation algorithms is a must-answer question.
Multi-party implementation of comprehensive algorithm management
The effective implementation of algorithmic comprehensive governance requires at least four aspects to be strengthened.
First, establish the correct orientation.
If the algorithm is designed by humans, it should be controlled by humans and responsible for it.
The value orientation embodied by algorithm recommendation reflects the values of algorithm recommendation service providers to a considerable extent.
Algorithm application must be upward for good, which is a social consensus; adhere to the mainstream value orientation, optimize the algorithm recommendation service mechanism, and actively spread positive energy, which are the basic requirements for algorithm recommendation service providers; do not use algorithm recommendation services to endanger national security and disrupt the economy For activities prohibited by laws and administrative regulations such as order and social order, infringement of the legitimate rights and interests of others, algorithm recommendation services must not be used to spread information prohibited by laws and administrative regulations. This is the bottom line rule that algorithm recommendation service providers must abide by; establish and improve manual intervention and users Independent selection mechanism, in the first screen of the homepage, hot search, selection, list category, pop-up window and other key links, actively present information content that conforms to the mainstream value orientation, and improve the accuracy and effectiveness of positive energy dissemination, which is algorithm recommendation The value orientation and pursuit of service providers.
In the comprehensive management of algorithms, a variety of rules and methods must be used to force requirements, and effective supervision and implementation must be ensured.
Second, improve laws and regulations.
The "Data Security Law of the People's Republic of China", which came into effect on September 1, 2021, comprehensively regulates data processing activities.
The "Personal Information Protection Law of the People's Republic of China", which will be implemented on November 1, defines the term "automated decision-making" in the legal text for the first time as "automatic analysis and evaluation of individual behavior habits and interests through computer programs." Hobbies or economic, health, credit status, etc. and making decisions” and clearly stipulates that “personal information processors use personal information to make automated decisions”, “should ensure the transparency of decision-making and the fairness and justice of the results”, and provide “ Do not select options based on their personal characteristics, or provide individuals with a convenient way of refusal"; "Use of personal information for automated decision-making", "Personal information protection impact assessment beforehand, and records of processing" should also be made.
The "Guiding Opinions on Strengthening the Comprehensive Management of Internet Information Service Algorithms" promulgated by the National Internet Information Office and other nine departments has made a top-level design for good management, good use, and development of good algorithm applications. The goal of establishing a comprehensive governance structure for algorithm security with a sound governance mechanism, a complete regulatory system, and a standardized algorithmic ecology, and a systemic arrangement for improving the algorithmic security governance mechanism, building an algorithmic safety regulatory system, and promoting the development of algorithmic ecological specifications.
The "Internet Information Service Algorithm Recommendation Management Regulations (Draft for Comment)" issued by the State Cyberspace Administration of China first distinguished five types of directions: generation synthesis, personalized push, sorting and selection, retrieval and filtering, and scheduling and decision-making. The algorithm technology for users to provide information content, and detailed regulations have been made on the responsibilities and obligations of algorithm recommendation service providers, algorithm recommendation service announcements and algorithm filing systems, and algorithm recommendation minor models.
This management regulation clearly will implement classification and hierarchical management of algorithm recommendation service providers based on the public opinion attributes of algorithm recommendation services or social mobilization capabilities, content categories, user scale, data sensitivity processed by algorithm recommendation technology, and degree of intervention in user behavior, etc. , And made specific provisions for algorithm security evaluation and supervision and inspection.
The implementation of the Data Security Law and the Personal Information Protection Law, the issuance of guidance and management regulations for public comments have provided a rule basis for the comprehensive governance of algorithms and laid a foundation for the rule of law.
The third is to improve supervision capabilities.
Algorithmic comprehensive governance requires professional supervision capabilities.
The algorithm's constant iteration and even self-renewal are a manifestation of its level of intelligence.
People have many views on the "interpretability" of algorithms, and they are deeply worried about the "uninterpretability" of algorithm recommendation results, so the algorithm is called the "black box".
Algorithm supervision requires not only professional supervision teams and continuously improved legal supervision capabilities, but also continuous iterative detection, verification, evaluation tools and technical supervision capabilities, so that comprehensive algorithm governance can be implemented.
The regulatory authority should have the professional ability to comprehensively test the algorithm with certain rules, specific input conditions, and output results without grasping the internal structure and core content of the algorithm; or according to the relevant information provided by the algorithm user. Recommend professional capabilities for comprehensive verification of service results; and professional capabilities for evaluating algorithm security in accordance with relevant regulations and verifying its self-assessment report.
The third part of the "Guiding Opinions on Strengthening the Comprehensive Management of Internet Information Service Algorithms" "Building an Algorithm Security Supervision System" puts forward requirements for "actively carrying out algorithm security evaluations", including "organizing and establishing professional technical evaluation teams, and in-depth analysis of algorithm mechanisms. Mechanism, evaluate the defects and loopholes in application links such as algorithm design, deployment, and use, study and judge security risks such as ideology, social fairness, and moral ethics arising from algorithm application, and propose targeted countermeasures.”
Such professional technical evaluation is the material basis for effective implementation of algorithmic comprehensive management.
Fourth, optimize the coordination mechanism.
Establishing a sound algorithm security governance mechanism with laws to follow, diversified coordination, and multi-party participation is an important guarantee for China's organizational advantages, institutional advantages and important guarantees for comprehensive algorithmic governance.
The basis for the construction of a collaborative mechanism is the allocation of rights, obligations, and responsibilities.
Through laws, regulations, rules and related regulations, further clarify the rights, obligations and responsibilities of governments, enterprises, industry organizations and netizens in the governance of algorithm security, and in particular, strengthen the main responsibility of enterprises for the results of algorithm applications. Government supervision, corporate responsibility performance, industry self-discipline, and social supervision are algorithmic security and multiple co-governance situations.
Therefore, enterprises are required to establish an algorithm security responsibility system and a scientific and technological ethics review system, improve algorithm security management organizations, strengthen risk prevention and control and hidden danger investigation and management, and improve the ability and level of responding to algorithm security emergencies.
The effective operation of the coordination mechanism also requires the coordinated operation of various relevant departments in accordance with laws and regulations, attracts a team of professionals, and gathers resources from multiple sources to provide strong support for algorithm security governance.
(The author is the Dean of the School of New Media and Communication, Tianjin University, and a researcher at the Research Center for Theoretical System of Socialism with Chinese Characteristics in Tianjin)
Lu Xiaohua