중국 인공지능의 세무 영역 응용 시의 리스크와 법적 규제 연구

As the breadth and depth of artificial intelligence participating in tax administration continue to improve, a series of real-world experiences show that artificial intelligence technology not only brings us convenience and efficiency, but also poses many risks in practice. It appears specifically in three aspects: data management risk, technology gap risk, and application ethics risk. With the advancement of data mining and big data analysis technology, the amount of data is no longer an important factor hindering the application of artificial intelligence. Conversely, data quality directly determines the accuracy of artificial intelligence applications under the influence of factors such as generalization and lack of verifiability of data sources. Incorrect reporting by taxpayers, duplicate collection and omission of data affect the quality of taxable data. Intrinsic bias in data also affects data quality. Tax authorities have expanded data sources for accurate supervision and inevitably expanded the collection and processing of taxpayer data, raising concerns about taxpayer privacy, while its high practicality and scalability make it an important area of data leakage. The additional promotion and application of next-generation artificial intelligence technologies in the tax sector will reinforce natural information asymmetry between tax agencies and taxpayers. Tax algorithms are related to national taxable interests, and confidentiality requirements do not help taxpayers obtain algorithmic information. Since taxpayers cannot understand and return the internal operating mechanisms by which tax agencies make decisions, the taxpayer's ability to question the decision-making outcome is differentiated, affecting the realization of the right to remedy. Artificial intelligence improves the efficiency of tax collection and management, while artificial intelligence decisions tend to 'consistency' or 'standardize' tax decisions, and can lead to unfair decisions about individual taxpayers in disregard of consideration of tax precedents. As for the data management rules, there is no clear regulation on the specificity of information quality standards, and it is true that it is difficult to effectively unify the data quality of each field. If the standards for collecting tax-related data are not clear, the government's information collection behavior cannot be effectively regulated. For example, the algorithmic interpretation rules are not clear, and the Personal Information Protection Act does not set the scope of the interpretation of government agencies. At the same time, countries around the world are currently having difficulty figuring out what 'automatic decision-making' is. It has been shown that algorithmic automation practices in fields such as justice, police, and social credit ratings can be biased like algorithmic decision-making and proceed more stealthily than human decision-making. This is where the dilemma of algorithmic deviation lies. The problem is that correlation is not causality, and that inferring causality only from correlation is likely to prejudice taxpayers. In a broad sense, algorithmic black boxes include both black boxes caused by technical bottlenecks and technical black boxes formed to protect the country's taxable interests. The algorithm's black box not only makes it impossible for taxpayers to directly participate in the formation of tax decisions and obtain a reasonable explanation for their conclusions, but also prevents tax authorities from objectively inquiring automated decisions and has weaknesses that make them vulnerable to the control and impact of automated decisions. There are several things that are necessary to respond to these issues. Data governance can be strengthened in terms of taxpayer rights to further improve the level of administration and law enforcement under the law. In order to establish data quality standards, cross-examination and sharing of data must be realized based on a unified data platform and data quality standards. At this stage, China should clarify data quality standards with a separate 'data quality method' as data quality rules are lacking. Support for data quality management through supervisory authorities helps to expand and maintain the implementation of data quality standards and improve data maturity. The United States has adopted a legislative format to clarify data quality standards, which is an effective way to ensure data quality, which is worth referencing by China. Data quality standards include many aspects such as acquisition, availability, reliability, correlation and performance. Unification of data standards can promote the integration of each data platform. Article 1039 of the Civil Code of China emphasizes that state agencies and employees should keep their privacy and personal information acquired during work confidential. Therefore, the duty of confidentiality of tax accountants has the nature of a civil obligation. In case of violating regulations, the taxpayer has the right to claim civil liability from the tax officer. According to administrative law theory, taxpayers are not allowed to file civil lawsuits against tax authorities, and have no choice but to apply the national compensation system to claim compensation. Currently, infringement of confidential information related to tax is not within the scope of China's 'National Compensation Act'. Therefore, a special compensation mechanism for public data infringement can be specified in the law. At the same time, the notification system of information leakage can be clarified in the 'Tax Collection Management Act' and related laws, allowing taxpayers to track the information security situation in a timely manner. It is also necessary to conduct regular cybersecurity training for tax employees, raise security awareness by publishing cybersecurity guidelines, and increase the effectiveness of data security management. Article 24 of China's 'Personal Information Protection Act' stipulates requirements for transparency of algorithms similar to the 'Regulations on General Data Protection', but the interpretation rules are not specific, so operability is insufficient. For this, the content interpreted by the algorithm must be further legally subdivided. Considering the peculiarity of the tax sector, China can add provisions subject to smart collection management to the 'Tax Collection Management Act'. Tax algorithms are about the country's taxable interests, and algorithmic interpretation does not require tax authorities to disclose the technical details of algorithmic decisions, and only if the taxpayer has objections to the algorithmic decisions can the tax authorities post-mortem the algorithm and provide a reasonable explanation to the taxpayer. In the case of decision-making using an algorithm, the taxpayer must be able to raise questions and apply for manual intervention, a powerful way to break down the algorithm's "mystery." In traditional tax collection management, tax agencies and tax staff must be supervised by taxpayers in accordance with the law. The objective existence of a "technical gap" in the algorithmic era makes it difficult for taxpayers to assess the tax algorithm system and its impact. To solve this problem, a special neutral agency or committee responsible for regular evaluation and monitoring of algorithms can be established within the administrative agency, and if the algorithm does not provide expected results, it must be able to adjust it. Finally, attention should be paid to the administrative retrial of the fairness of the algorithm and the relief effect of litigation. To strengthen algorithmic ethics screening, next-generation artificial intelligence ethics norms have shown that algorithms can regularly evaluate and monitor them from their impact on the economy, society, government management, and citizenship. If an algorithm is found to fail to provide expected results, it can be adjusted and updated in real time to meet the needs of tax law enforcement. In addition, by setting an integrated algorithmic transparency template, an algorithm audit can be performed by a third-party agency. It should be noted that even if the algorithm's decision-making is wrong, the subject responsible for it is still at the tax authorities. Artificial intelligence perfecting tax affairs in combination with collection management is a concrete implementation of the modernization of tax governance in China and must be carried out under the guidance of the principle of tax certainty. Specifically, the application criteria of artificial intelligence can be clarified in the 'Tax Collection Management Act' or related laws. Currently, neither the legislative nor academic level has a unified understanding of the boundaries of artificial intelligence application. China should be able to consider the field of introduction of artificial intelligence due to factors such as whether it affects the rights of the people or the principle of tax fairness.