CHANG TSI
Insights
(The 1st Article) Changes in Examination Rules for Artificial Intelligence-Related Patent Applications, Case Studies, and Recommendations
On April 30, 2025, the China National Intellectual Property Administration (CNIPA) issued a notice soliciting public opinions on the "Draft Amendments to the Patent Examination Guidelines (Solicitation for Comments)." The notice was accompanied by a comparison table of the amendments and explanatory notes, inviting feedback from the public.
Among the revisions, the most notable updates include changes to the examination rules for the artificial intelligence field in Section 6, Chapter 9, and the introduction of examination rules for the video codec field in Section 7, Chapter 9. Specifically:
1. In the first and second parts of Section 6, Chapter 9, the examination criteria for artificial intelligence-related patent applications have been revised, with corresponding case studies provided.
2. In the third part of Section 6, Chapter 9, new requirements for drafting specifications for artificial intelligence-related patent applications have been introduced, along with relevant case studies.
3. In Section 7, Chapter 9, new provisions have been added for the examination of invention patent applications containing bitstreams.
This series, authored by Irene, provides a comprehensive interpretation of the changes to the examination rules in the fields of artificial intelligence and video codec technologies. The series consists of three articles, each corresponding to the amendments outlined in points (1)-(3). This first article focuses on an in-depth analysis of the changes to examination criteria for artificial intelligence-related patent applications, explores relevant case studies, and offers practical recommendations.
1. Changes to the Chapter Title and Examination Criteria
1.1 The Chapter Title
The title of this chapter has been changed from "Inventions Involving Algorithm Features or Business Methods and Rules" to "Inventions Related to Artificial Intelligence and Similar Technologies."
1.2 Examination Criteria
The examination criteria now emphasize that, for applications in the AI field, examiners must consider the content of the specification when necessary as part of the basis for conducting the patent examination.
- Regarding the scope of examination for AI-related applications, the modifications are as follows (underlined text indicates newly added content):
The examination should focus on the solution being claimed, i.e., the solution defined in the claims, and, when necessary, examiners must consider the content of the specification as part of the basis for conducting the patent examination. During the examination, technical features, algorithm features, or business rules and methods should not be examined in isolation. Instead, all content recorded in the claims should be considered as a whole, and the technical means involved, the technical problems solved, and the technical effects achieved should be analyzed comprehensively.
2. Newly Added Prohibitive Provisions and Corresponding Examination Cases
The newly added prohibitive provisions specify that if the application involves aspects such as data collection, label management, or rule configuration that violate laws, social morality, or harm public interests (e.g., unlawfully obtaining personal information or violating ethical decision-making), the application shall not be granted a patent under Article 5 of the Patent Law.
- Regarding the newly added prohibitive provisions, the modifications are as follows (underlined text indicates newly added content):
6.1.1 Examination Based on Article 5, Paragraph 1 of the Patent Law: For invention patent applications involving algorithm features or business rules and methods, if they contain content that violates laws, social morality, or public interests—such as data collection, label management, rule configuration, or recommendation decisions that violate laws, fairness, justice, or exhibit discrimination or bias—then, under Article 5, Paragraph 1 of the Patent Law, such applications shall not be granted patent rights.
- New Case Examples (All Newly Added):
Case Example 1: Big Data Sales System for Shopping Malls
This example underscores the importance of compliance with the *Personal Information Protection Law*, which requires explicit individual consent for data collection.
Key Point:
(1) Invention patent applications that include algorithmic features or commercial rules and methods violating laws, social morality, or harming public interests cannot be granted patent rights.
Overview: The invention patent application provides a solution for a big data-based mattress sales assistance system in shopping malls. The system utilizes camera modules and facial recognition technology to collect facial feature information and identify customer identities without their knowledge. The collected data is analyzed to assist businesses in targeted marketing.
Claim of the application:
Analysis and Conclusion:
The relevant provisions of the "Personal Information Protection Law of the People's Republic of China" stipulate that the installation of image acquisition and personal identification equipment in public places shall be necessary for maintaining public safety, comply with relevant national regulations, and include prominent warning signs. The collected personal images and identity identification information can only be used for the purpose of maintaining public safety and shall not be used for other purposes; unless the individual's separate consent is obtained.
This invention applies image collection and facial recognition methods in commercial venues such as malls for precise mattress marketing, which clearly does not fall under the scope of maintaining public safety. Moreover, the collection of facial information and identification of customer identity is conducted without their knowledge and without obtaining individual consent. As a result, this invention violates the law and, according to Article 5, Paragraph 1 of the Patent Law, cannot be granted patent rights.
Case Example 2: Emergency Decision-Making Model for Autonomous Driving
This example emphasizes that if the implementation of technology violates ethical principles (e.g., sacrificing the interests of a minority), it is deemed to violate social morality and cannot be granted.
Overview:
The invention patent application provides a method for establishing an emergency decision-making model for an autonomous vehicle. It uses pedestrians' gender and age as obstacle data and employs a trained decision-making model to determine the protected party and the impacted party in situations where obstacles cannot be avoided.
Claim of the application:
A method for establishing an emergency decision-making model for an autonomous vehicle, characterized by:
obtaining historical environmental data and historical obstacle data for an autonomous vehicle, wherein the historical environmental data includes the vehicle's driving speed, distance to obstacles in a lane of the vehicle, distance to obstacles in adjacent lanes, movement speed and direction of obstacles in the lane of the vehicle, and movement speed and direction of obstacles in adjacent lanes; the historical obstacle data includes pedestrians' gender and age;
extracting features from the historical environmental data and historical obstacle data as input data for the decision-making model, and using the vehicle's historical driving trajectory in situations where obstacles cannot be avoided as output data for the decision-making model; training the decision-making model based on the historical data, wherein the decision-making model is a deep learning model;
obtaining real-time environmental data and real-time obstacle data, when the autonomous vehicle encounters a situation where obstacles cannot be avoided, the trained decision-making model is used to determine the vehicle's driving trajectory.
Analysis and Conclusion:
This invention involves a method for establishing an emergency decision-making model for an autonomous vehicle. Human life holds equal value and dignity, regardless of age or gender. If the emergency decision-making model for autonomous vehicles selects the protected party and the impacted party in unavoidable accidents based on pedestrians' gender and age, it conflicts with the ethical principle of equality for all lives.
Moreover, this decision-making approach reinforces existing biases related to gender and age in society, raises public concerns about the safety of autonomous transportation, and undermines public trust in technology and social order. Therefore, this invention contains content that violates social morality and, in accordance with Article 5, Paragraph 1 of the Patent Law, cannot be granted patent rights.
Summary:
Case Example 1 emphasizes that invention patent applications containing algorithm features or business rules and methods, particularly those involving the legitimacy of personal information collection and the purpose of using personal information, must strictly comply with the *Personal Information Protection Law of the People’s Republic of China*.
Case Example 2 highlights that invention patent applications must adhere to ethical principles related to human rights, the right to life, and non-discrimination. They must take into account the impact on the public's perception of fairness and morality.
3. Enhanced Standards for Inventiveness Examination and Corresponding Case Studies
- New Requirement for Evaluating Technical Contributions: Algorithm features must interact with technical features and solve specific technical problems.
- Newly Added Relevant Examination Case Example (all newly added content):
Case Example 18: Method for Identifying the Number of Ships
This example emphasizes that merely using conventional deep learning without demonstrating technical improvements lacks inventiveness.
Overview:
The invention patent application provides a method for identifying the number of ships by obtaining images of ships, training a detection data model through deep learning, and addressing the technical problem of accurately identifying the number of ships in a given sea area.
Claim of the application:
A method for identifying the number of ships, characterized by:
obtaining a dataset of ship images and preprocessing image information within the dataset to mark position and boundary information of ships within the image information, and dividing the dataset into a training dataset and a testing dataset;
using the training dataset for deep learning to build a training model;
inputting the testing dataset into the training model to obtain ship testing result data;
multiplying the ship testing result data by a preset error parameter to determine actual number of ships.
Analysis and Conclusion:
D1 discloses a method for identifying the number of fruits on a tree, specifically disclosing steps such as obtaining image information, marking the positions and boundaries of fruits in the images, dividing the dataset, training the model, and determining the actual number of fruits.
The solution proposed in the present invention patent application differs from D1 only in the type of object being identified. Although ships and fruits differ in appearance, size, and environment, for a person skilled in the art, the steps required to identify the actual number—such as marking positions in images, dividing datasets, and training models—target the positional relationships of objects in the images. The claim does not reflect any changes in the deep learning process, training methods, or model hierarchy due to differences in the objects being identified. Marking ship data in images and marking fruit data in images to obtain training datasets and train models do not involve adjustments or improvements to deep learning, model construction, or training processes.
Case Example 19: Waste Steel Grading Model
This example demonstrates that algorithm adjustments (e.g., convolutional layer structure optimization) directly linked to technical effects possess inventiveness.
Overview:
When storing waste steel, it is necessary to classify grades based on the average size of the steel material. However, waste steel is often stored in a disorganized manner, piled on top of each other, making manual measurement and grading inefficient and inaccurate. The invention patent application provides a method for establishing a neural network model for waste steel grading using convolutional neural networks to create a model with grade classification outputs, thereby improving the efficiency and accuracy of waste steel grading.
Claim of the application:
A method for establishing a neural network model for waste steel grading, wherein the model is used to classify the grades of stored waste steel, the method comprising:
obtaining multiple images, determining grades of waste steel in the images, preprocessing the images, extracting image data features of different grades, and performing convolutional neural network learning on the extracted image data features of different grades to form a grade classification neural network model with grade classification outputs;
the extraction of color, edge features and texture features of objects in the image, and the extraction of correlation features between edges and textures of objects in the image, which are composed of multiple line convolution layers or convolution layers plus pooling layer calculations output by the set;
wherein the extraction of the color and edge features of the object in the image is composed of the set output of the calculation output of three line convolution layers plus pooling layers, including the first line one pooling layer, the second line two convolution layers and the third line four convolution layers from left to right; the extraction of texture features in the image is the extraction of the set output of the extraction of the color and edge features of the object in the above image, which is composed of the set output of the calculation output of three line convolution layers, including the first line zero convolution layer, the second line two convolution layers and the third line three convolution layers from left to right;
the number of circuits calculated by the convolution layer for extracting the correlation features between edges and textures is greater than the number of circuits calculated by the convolution layer for extracting the color, edge and texture features of objects in the image.
Analysis and Conclusion
D1 provides a method for identifying the types of scrap steel based on a convolutional neural network model to solve the problem that the sources of renewable resources are complex, the types are diverse, and the materials are very different. It is necessary to accurately identify whether the scrap steel belongs to material beans, stamping material residues, bread iron or other types to improve the recycling rate of renewable resources. It specifically discloses the steps of obtaining multiple image data of the scrap steel types that have been determined, preprocessing the image data for feature extraction, and using the convolutional neural network for training and obtaining the product model.
The difference between the solution of the present invention patent application and D1 is that the training data and the extracted features are different, and the number of lines and the level settings of the convolution layer and the pooling layer are also different. Compared with D1, it is determined that the technical problem actually solved by the invention is how to improve the accuracy of scrap steel grading. D1 uses scrap steel image data of a certain type to extract features and conduct model training. In order to classify scrap steel according to its average size, the present invention patent application needs to identify the shape and thickness of the scrap steel from the disordered and overlapping scrap steel images. In order to extract the color, edge and texture features of the scrap steel in the image, the number of lines and the level settings of the convolution layer and the pooling layer are adjusted during the model training process. The above algorithm features and technical features support each other in function and have an interactive relationship, which can improve the accuracy of scrap steel classification. The contribution of the algorithm features to the technical solution should be considered. The above adjustments to the number of lines and the level settings of the convolution layer and the pooling layer have not been disclosed by other prior art documents, nor are they common knowledge in the art. The prior art as a whole does not provide inspiration for improving the above D1 to obtain the technical solution of the present invention patent application, and the claimed technical solution to be protected possess inventiveness.
Summary:
By combining Case Example 18 and Case Example 19, it is evident that during the inventiveness examination of artificial intelligence-related patent applications, factors such as whether the model is merely applied or whether the model structure brings actual improvements are considered.
The newly added inventiveness examination case examples provide practical guidance for drafting AI-related patents, emphasizing the collaborative interaction between algorithm features and technical features.
1. Mitigate Ethical and Compliance Risks (Corresponding to Newly Added Prohibitive Provisions)
- Legitimacy of Data Sources:
Before filing a patent application in the field of artificial intelligence, ensure that data collection and processing comply with the Personal Information Protection Law and ethical standards. For example, explicitly indicating that "individual consent has been obtained" (refer to Case Example 1). It is necessary to avoid using covert data collection methods (e.g., undisclosed facial recognition). it is possible to consider specifying the data anonymization process to further ensure compliance.
- Algorithm Value Review:
Exclude discriminatory decision-making logic based on sensitive attributes such as gender or age (e.g., the autonomous driving model in Case Example 2).
2. Strengthen Technical Integration and Inventiveness (Corresponding to Revised Inventiveness Examination Standards)
- Highlight Collaborative Innovation Between Algorithms and Hardware:
1) Avoid Simple Algorithm Transplantation:If a general model (e.g., CNN) is applied to a new scenario (e.g., ship identification in Case Example 18), it is necessary to demonstrate how structural adjustments to the model address specific technical problems in the field (e.g., interference from lighting, overlapping targets).
2) Quantify Technical Effects: Provide comparative experimental data in the specification to demonstrate significant improvements in accuracy or efficiency brought about by algorithm optimization (e.g., the waste steel classification model in Case Example 19).
- Define a Technical Problem-Oriented Approach:
Describe in the claims how algorithm features drive improvements in physical devices (e.g., "optimized convolutional layer structure → enhanced image processing speed → reduced delays in industrial sorting systems").
Distinguish the invention from purely commercial methods, emphasizing breakthroughs in technical bottlenecks (e.g., resolving sensor noise or insufficient real-time performance).
The latest draft revision has significantly strengthened the ethical and compliance examination standards for AI patents (e.g., data legality and algorithm fairness) while raising the bar for technical innovation requirements (e.g., collaborative problem-solving between algorithms and hardware). These changes reflect China's commitment to ensuring that AI-related inventions not only comply with legal and ethical norms but also contribute genuine technical advancements to the field.
It is important to note that this draft is still in the consultation phase (published on April 30, 2025), and the final rules may be subject to adjustments. As the regulatory landscape for AI patents continues to evolve, staying informed about these changes is crucial for applicants seeking protection for their innovative technologies in China.
At Chang Tsi & Partners, we specialize in navigating the complexities of AI-related patent applications. Whether you are interested in learning more about the examination standards for AI patents in China or have plans to file AI-related patent applications, we are here to assist you.
For more detailed updates and insights into AI patent regulations in China, stay tuned as we continue to monitor developments in this dynamic area of intellectual property law
