According to the "Guidelines for Patent Applications Related to Artificial Intelligence (Draft for Comments)" released by CNIPA at the end of 2024, the common types of AI-related patent applications can be summarized into the following four categories, each involving specific legal issues and examination requirements:

1. Patents involving the AI algorithm or model itself

These applications focus on innovations in AI algorithms or models, such as new neural network structures or optimized machine learning algorithms. It should be noted that the algorithm itself must meet the requirements of a "technical solution" under patent law, i.e., embody technical features and solve a specific technical problem. During examination, emphasis will be placed on whether the algorithm is combined with a specific application scenario to avoid being classified as an abstract mathematical method.

2. Functional application patents based on AI algorithms or models

These applications emphasize applying existing algorithms or models to a specific field (for example, medical diagnostics, autonomous driving, image recognition, etc.), with clear details of the technical implementation and its practical effects. For example, using an AI model to optimize the efficiency of medical image analysis. The examination will require proof of the technical contribution of the application and ensure that there is an interaction between the function and the algorithmic features.

3. AI-assisted invention applications

These refer to technical solutions completed by humans with the aid of AI tools, such as designing drug molecules or optimizing industrial processes with the help of AI. The inventor must be a natural person who has substantially contributed to the inventiveness, with AI only being used as a tool. It is necessary to clearly state the core role of humans in the invention in the specification to avoid ambiguity in the determination of inventiveness due to AI intervention.

4. Inventions generated entirely by AI

These applications refer to inventions autonomously generated by AI, such as algorithm optimization schemes without direct human intervention. However, according to the guidelines, the inventor must still be a natural person, and AI itself cannot be the subject of the invention. These applications involve significant legal disputes and require a special explanation of the human contribution in AI training, parameter setting, or result selection to meet the patent law's requirements for an "inventor."

Key Legal and Examination Requirements

Identification of the inventor: The inventor must be a natural person in all types of inventions, and AI systems cannot be named.

Sufficient disclosure in the specification: To address the AI "black box" issue, it is necessary to disclose in detail the algorithm logic, data processes, and other technical details related to the invention point to ensure examinability.

Consideration of inventiveness: Algorithmic features must support and be supported by technical features, collectively reflecting the novelty and inventiveness of the invention. For example, algorithmic optimization should be combined with hardware performance improvement or specific application effects.

The CNIPA's current guidelines are still in the consultation stage. As technology advances and the market changes, the content of the guidelines may be further adjusted and published after incorporating public comments. For detailed guidance content, please feel free to contact Chang Tsi & Partners. We at Chang Tsi & Partners will also continue to keep you updated with our interpretations of related news.