算法透明原则作为算法治理的基本原则之一，目前已在国际层面形成共识。我国法律体系中的算法透明散见于多部法律文件中，主要有确立算法透明原则地位、算法透明具体规则和以算法透明为基础方能实现立法目的的三类条文。共同体现着算法透明在保障知情权、反垄断等方面的多维价值。算法透明原则的适用范围包括三方面。一是主体范围至少包括公众、利益相关方、监管机构和第三方机构，在实际场景中有所区别。二是算法范围与三个关键因素有关：自动化决策、是否直接与公众互动以及潜在公共影响。三者满足其一即可纳入算法透明原则的算法范围。三是内容范围主要包括“说明义务约束下的信息”和“技术层面的信息”两类。其中，“技术层面的信息”包括但不限于源代码、数据和参数。综合实证研究和法律分析，算法透明的三个子原则可确定为：可获取性、可解释性和可理解性。可获取性是算法透明原则的程序要求，与易得、及时等具体要求关联。可解释性是算法透明的实质要求，满足可解释性的要求能使得算法透明的实质内容得到充实。可理解性是算法透明的主观要求，在破除算法技术壁垒上表现出了人文关怀。从技术、专业和监管三个路径构建算法透明规则体系的建议是：技术路径上，“析理型透明”是必要的。开放源代码等技术路径应采取不强制披露但同时鼓励披露的态度。专业路径上，算法影响评估和算法审计作为典型代表制度有易满足可理解性的优势，应考虑解决规制范围不广、未建立持续性的评估机制等问题。关于监管路径，应进一步制定对备案内容的规范性要求，考虑以算法备案作为切入点，通过算法备案的实质审查和建立算法备案的动态监管的方式对算法进行穿透式监管。
The principle of algorithmic transparency, as one of the fundamental principles of algorithmic governance, has currently formed a consensus at the international level. The algorithmic transparency in China‘s legal system is scattered in multiple legal documents, mainly including three types of provisions that establish the principle of algorithmic transparency, specific rules for algorithmic transparency, and provisions based on algorithmic transparency to achieve legislative purposes. Together, it embodies the multidimensional value of algorithmic transparency in safeguarding the right to know and combating monopolies.The application scope of the principle of algorithmic transparency includes three aspects. One is that the scope of the subject includes at least the public, stakeholders, third-party institutions, and regulatory agencies, which may differ in actual scenarios. The second is that the scope of the algorithm is related to three key factors: automated decision-making, direct interaction with the public, and potential public impact. One of the three can be included in the scope of algorithmic transparency principle. The third is that the scope of content mainly includes two categories: "information under the obligation to explain" and "information at the technical level". Among them, "technical information" includes but is not limited to source code, data, and parameters.Based on comprehensive empirical research and legal analysis, the three sub principles of the principle of algorithmic transparency can be determined as: accessibility, interpretability, and understandability. Accessibility is a procedural requirement of the principle of algorithmic transparency, which is related to specific requirements such as easy to access and show in time. Interpretability is the essential requirement of algorithmic transparency, and meeting the requirement of interpretability can enrich the substantive content of algorithmic transparency. Understandability is a subjective requirement for algorithmic transparency, demonstrating humanistic care in breaking down technological barriers to algorithms.The suggestion for constructing an algorithmic transparency rule system from three paths: technical, professional, and regulatory is that "reasoned transparency" is necessary in the technical path. Open source code and other technological paths should adopt an attitude of not mandatory disclosure but also encouraging disclosure. On the professional path, algorithm impact assessment and algorithm audit, as typical representative systems, have the advantage of being easy to understand. It should be considered to address issues such as limited regulatory scope and the lack of a sustained evaluation mechanism. Regarding the regulatory path, it is necessary to further develop normative requirements for the record-filing content, considering algorithm record-filing as the entry point, and conducting penetrating supervision of algorithms through substantive review of algorithm record-filing and dynamic supervision of establishing algorithm record-filing.
