Abstract: The term “patent” has dual meanings of “publicity” and “exclusivity”. Therefore, “public for protection” has become a classic expression of the patent system: the patent applicant discloses a technical solution to the society to obtain the protection of the patent law. If the technical information disclosed by the patent applicant is insufficient, the technical solution cannot be protected by the patent system. This is to ensure that the patented technical solution can be fully disclosed to the society to promote technological development. In recent years, the number of patent applications for artificial intelligence (AI) algorithms has increased rapidly. In terms of the disclosure of AI algorithm patents, the following three factors lead to their mismatch with the existing patent disclosure standards - computer software patents and patents containing algorithm features: the black box attribute of AI algorithm, the low predictability of the algorithm technology field, and the easy hiding of useful technical information by patent applicants. In the field of AI algorithm patents, this mismatch causes a conflict between the “public interests corresponding to the full disclosure” and the “private interests directed by monopoly protection” pursued by the patent system: an AI algorithm technology is protected by both the patent system and trade secrets, which limits the public interests. Reviewing the evolution of the patent system, disclosure theory has become one of the justifications of the patent system. For example, biotechnology is highly unpredictable, and in order to better realize the knowledge spillover effect of the patent system, the disclosure standard of biotechnology patents is high. Considering of the comparability between China’s biotechnology patents and AI algorithm patents, in order to better implement the “public for protection” concept required by the patent disclosure theory, the AI algorithm patent disclosure standards should be improved, and the technical information should be fully disclosed when the patent is granted. The AI algorithm patent applicant shall disclose the following four contents: (1) No less than 3 embodiments; (2) Disclosure of algorithm source code; (3) Disclose the algorithm training method; (4) Disclose algorithm training data. Improving the disclosure standard of AI algorithm patents will lead to AI algorithm technology being protected as trade secrets, and increase the patent application cost of patent applicants. Incentive measures can be considered to eliminate the negative effects of changes in patent disclosure standards. And incentive measures can be divided into two types: inside the patent system and outside the patent system. The former includes reducing the patent application fee, reducing the annual patent fee, and giving priority to examination, while the latter mainly includes government grants, financial support, and tax relief. In view of the relatively low implementation cost of incentive measures in the patent system and the better incentive effect for patent applicants, it should be taken as a solution. Only in this way can patent applicants be encouraged to disclose algorithm source code, to train data and other information, and to better realize the social function of patent system to promote knowledge spillover and technology development.

摘要： 近年来人工智能（AI）算法专利申请量迅速增长。但AI算法黑箱属性、算法技术领域可预测性低、专利申请人易隐藏技术有用信息，使现行专利披露标准难以较好平衡“充分公开对应的公共利益”和“垄断保护指向的私人利益”。为更好践行专利披露理论所要求的“公开换保护”理念，应提高AI算法专利披露标准，确保在授予专有权的同时充分披露技术信息。鉴于提高专利披露标准会导致AI算法技术多被作为商业秘密保护，可通过费用减免和优先审查等专利制度内的激励措施，来鼓励专利申请人披露算法源代码、训练数据等信息的积极性，更好实现专利制度促进知识溢出和技术发展的社会功能。