On December 6, 2024, the China National IP Administration (CNIPA) announced the publication of a draft for the AI-related Invention Patent Application Guidelines (Guidelines).[1] AI technology has made significant advances globally, and the number of patent applications has increased considerably. In response to this trend, the Guidelines represent a strategic effort to rigorously and comprehensively construct a policy framework for the patent examination of inventions in the AI realm. Specifically, the Guidelines aim to clarify and improve the quality of the current examination process, address key challenges encountered by innovators, and more importantly, serve as a standardized policy interpretation to help applicants understand the current examination norms within the existing legal framework.
Chapter 1 summarizes the four common types of AI-Related patent applications. They are: (1) applications for AI algorithms or models themselves, (2) applications related to functions or industrial implementations based on AI algorithms or models, (3) applications for AI-assisted inventions, and (4) applications for AI-generated inventions. They are further elaborated as follows.
In the first category, AI algorithms or models covers fundamental AI components such as machine learning, deep learning, neural networks, fuzzy logic, genetic algorithms and model optimization. These fields are the core of AI. Correspondingly, these types of applications may also involve the improvement or optimization of algorithms or models such as model structure, compression and training, among others.
The second category concerns applications relating to AI functions and industrial implementations, meaning the integration of AI into specific functionalities or industries. These functions may be natural language processing, computer vision, speech processing, knowledge presentation and reasoning, or data mining. Implementations of AI in various industrial fields such as transportation, telecommunications, life sciences, medicine, security, business, education, entertainment and finance in order to promote the advancement of innovation is also a key area of applications.
For the third category, that of AI-assisted inventions, AI technology serves as an auxiliary tool in the inventive process. An example would be the use of AI to identify specific protein binding sites ultimately resulting in the acquisition of a novel drug compound.
The last category concerns AI-generated inventions, those which are created autonomously by AI without any substantial human contribution. An example of this is a container autonomously designed by AI systems.
Chapters 2 to 5 of the Guidelines present five major legal issues.
Inventorship (Chapter 2)
The Implementing Regulations of the Patent Law of the PRC require that an inventor must be a person who makes creative contributions to the essential features of the invention. For AI-assisted and AI-generated inventions, AI systems participate in the invention process to varying degrees. Whether an AI system can be named as an inventor has long been a subject of discussion.
CNIPA's new Patent Examination Guidelines specifically stipulate that the inventor must be an individual and the application form must not include an organization or the name of the AI system. This is a clear manifestation that, unlike a human being, an AI system cannot be a subject with entitlement to either property rights or moral rights. Therefore, an AI system cannot be named as an inventor.
For inventions made with the assistance of AI, a natural person who has made creative contributions to the substantive features of the invention can be named as the inventor. On the contrary, as previously explained, it is illegal to grant inventorship to an AI system for an invention generated by AI.
Subject matter (Chapter 3)
Patentable subject matter must direct to a technical solution, which is a collection of technical means that utilize the laws of nature to solve technical problems. When a claimed invention includes both the rules of intellectual activities and technical features, the invention shall become patentable subject matter. In the area of AI, the question of patent eligibility depends on whether, for example, the execution of the algorithm or model employed in the invention utilizes natural laws to solve a particular technical problem. When using AI algorithms or models to analyze and predict big data in various fields, the question becomes how one is able to judge whether the intrinsic correlation between the mined data conforms to the laws of nature. For example, a method for establishing a general neural network model based on an abstract algorithm or a method for training a general neural network using an optimized loss function to accelerate training convergence would not be considered mere abstract mathematical algorithms and would thus not be eligible for a patent if no technical features are involved.
The Guidelines illustrate three specific scenarios in which an invention embodies a technical solution. Firstly, the AI algorithms or models process data with specific technical content. Secondly, the AI algorithms or models have specific technical correlations with the internal structure of computer systems. For example, the invention solves technical issues by improving hardware computing efficiency or execution -- by reducing data storage volume, reducing data transmission volume, or increasing hardware processing speed. Thirdly, the AI system mines big data in specific industrial fields to find intrinsic correlations based on an AI algorithm. Such intrinsic correlations must conform to the laws of nature rather than socio-economic norms. Taking, for example, a method for estimating a regional economic prosperity index using a neural network to determine the intrinsic correlation between economic data and electricity consumption data then using this correlation to make a prediction, such a correlation is regulated merely by economic norms. Hence, the claimed method does not constitute a technical solution and is thus not eligible for a patent.
Sufficient Disclosure (Chapter 4)
"Black box" is a feature of AI algorithms and models. Sufficient information must be provided for the purposes of full disclosure. The specification must clearly and comprehensively describe the technical means embodying the concept of the patent invention to the extent that the person with technical skill in the art can enable it. Moreover, the specification must clearly and objectively state the beneficial effects of the invention.
For example, for a claimed invention whose inventive concept lies in the training of AI models, the specification generally requires, among others, a description of the algorithms involved and the specific steps performed. To give another example, for a claimed invention whose inventive concept lies in the construction of an AI model, the specification generally requires a description of the necessary module structure, hierarchical structure or connective relationship, among other things. If necessary, the achievable technical effect should be corroborated through experimental data, analytical statements, or other means. Lastly, for a claimed invention whose inventive concept lies in the implementation of AI in various industrial fields, the specification generally requires, inter alia, a description of how the model is integrated into a specific industrial scenario, or of the method by which the input or output data is set.
Inventiveness (Chapter 5)
In principle, when evaluating inventiveness, the technical features and the algorithmic features that functionally support and interact with those technical features should be evaluated as a whole.
When using AI algorithms or models for specific functions or in specific industrial fields, the contribution of algorithmic features to the solution should be evaluated. That is, it is necessary to state the technical problems solved, the technical means adopted that conform to the laws of nature, and the technical effects achieved. If the solution involves adjusting the existing AI algorithm processes or model parameters, and the adjustment solves the technical problems in question and produces beneficial technical effects, it can be determined that the algorithm features and technical features support and/or interact with each other.
When using AI algorithms or models for different functions or in different industrial fields, factors to be evaluated in terms of their favorability for inventiveness include, among others, the closeness of the technical fields, the existence of any corresponding technical inspirations, the difficulty of using them in different fields, whether there are any technical difficulties that need to be overcome, and whether they bring unexpected technical results.
Improvement in the internal performance of computer systems is another positive factor in inventiveness evaluation. Examples are the support or optimization of the operation of specific algorithms or models by adjusting the hardware system architecture and the optimization of hardware resource scheduling in the computer system through the execution of algorithms or models.
Another significant positive factor is the improvement of the user experience. A claimed invention aims to enhance online customer service by addressing the underutilization of Chabot resources through the optimization of dynamic allocation between robotic and human customer service with the use of a genetic algorithm to reduce customers' waiting time. It indicates optimistic potential for being deemed inventive.
Ethical issues (Chapter 6)
Article 5 of the Patent Law of the PRC stipulates that no patents shall be granted to inventions that violate the law or social ethics or that harm public interests. This is the emperor's clause of patent practice. In the world of AI inventions, particular attention must be given to issues of algorithm ethics, data safety and data compliance. Inventions with AI used in a particular industrial scenario must avoid any legal violations. For inventions applying AI to obtain and utilize data, caution must be exercised regarding the source, destination, privacy management and usage regulations of the data. The collection, storage and processing of data must also comply with the relevant laws and regulations.