New practice on patentability of AI inventions in China

More AI patents are being allowed due to revised guidelines introduced this year that set out specific criteria, says Hengwei Zhou of CCPIT Patent and Trademark Law Office.

With the rapid advancement of artificial intelligence (AI) technology, the world is undergoing unprecedented transformations. When innovators in AI technology seek to protect their intellectual achievements through patents, a primary challenge arises—the definition of patentable subject matter.

As stated in Article 2.2 of China’s Patent Law, patentable subject matter excludes non-technical solutions. There was a lack of clear criteria in China’s patent practice for determining whether an AI invention constitutes a technical solution, given the abstract, complex, and unobservable nature of AI algorithms.

This situation has changed with the revised Guidelines for Patent Examination coming into effect on January 20, 2024, which introduced three specific criteria for evaluating the patentability of an AI invention. This article introduces these criteria as well as our observations of their practice over the past six months, aiming to clarify the latest patentable boundaries of AI inventions in China.

Defining accepted data types

The first criterion focuses on the nature of data processed by an AI algorithm. To put it simply, if data employed by the AI algorithm defined in the claims possesses a clear technical connotation, that data is deemed technical and the AI invention utilising this data is a patentable technical solution.

Over the past six months, we observed that diverse types of data are recognised as technical, including images, texts, audio data, measurement data, cellular communications, network packets, etc. Even generalised concepts such as sensor data and media data are accepted by the Patent Office. However, purely mathematical constructs like vectors, features, or graph topologies, are still deemed non-technical.

We also observed that the Patent Office does not require all data processed by the AI algorithm to be technical. It is sufficient to define some technical data in the claims, regardless of the technicality of other data, such as intermediate data or output data, of the AI algorithm. 

The Patent Office also accepts general claim language such as ‘an AI model for image processing’, implying the data being processed includes images, which are deemed technical data.

The first criterion is quite straightforward. When preparing patent applications, applicants are advised to include in the specification a comprehensive list of potential technical data involved in their AI model, from which they may choose to define desired data types in the claims to avoid a potential non-patentable subject matter issue. In the case of a broadly applicable AI model, various technical data types may be relevant. While seeking protection for such an AI model, applicants may submit divisional applications for different technical data types.

Technical association

The second criterion involves assessing a particular technical association between an AI invention and the internal structure of a computer system. If such an association exists, and if the AI invention addresses a technical problem within the computer system and achieves performance improvements, it could be accepted as a patentable technical solution.

The key point is to present the technical association. In our observations, a technical association may manifest itself through the implementation of the AI invention that either depends on or impacts the internal structure of the computer system. 

For example, an AI invention that adjusts execution parameters of its model based on the number of processing cores would be considered to have a direct technical association with the internal computer structure.

An AI invention that solely enhances the AI algorithm, even if such enhancements reduce execution time of the algorithm, would not be deemed to be a technical solution. Put differently, if an AI algorithm is independent of the computer device on which it runs and can be executed on any general-purpose computer and achieve the same effect, that algorithm would not meet the second criterion for being accepted as a technical solution.

To satisfy the second criterion, it would be beneficial for applicants to provide clear explanations in the specification regarding the specific technical association between the AI invention and the internal structure of the computer system. The internal structure may involve hardware, software, or both. Additionally, the applicants may provide an explanation of how the technical association leads to desired technical effects. Exemplary technical effects include reducing data storage, decreasing data transmission demands, enhancing hardware processing speed, etc.

Natural correlation

The third criterion focuses on identifying a natural correlation through big data mining. If an AI invention mines an inherent correlation within big data from specific fields and the correlation aligns with natural principles and potentially resolves technical issues pertaining to the reliability or accuracy of data analysis, leading to potential technical effects, that AI invention could be considered a technical solution.

The essence of this criterion is to present a mined correlation conforming to natural principles. In China’s practice, it has been observed that ‘natural principles’ generally encompass physical principles, chemical phenomena, and biological characteristics, excluding purely mathematical theorems, business logic, and man-made rules.

For instance, a correlation between air humidity and paint quality in industrial factories might be considered to conform to natural principles. An AI model trained on effective court decisions to predict litigation outcomes is not deemed technical, because it generally reflects man-made laws.

The third criterion is relatively subjective. Applicants are advised to draft their specifications to demonstrate the potential natural attributes of the correlation discovered by the AI invention, even if providing a comprehensive explanation may be challenging. This might enable the Patent Office to perceive the underlying natural correlation.

An AI invention is deemed technical and patentable if satisfying any one of the three criteria discussed above. We have seen an increase in the number of allowed AI patents, which would otherwise face subject matter rejections in absence of these criteria.

The introduction of the criteria, which are specifically designed for AI inventions, signifies a relaxation of the restrictions by China’s Patent Office and demonstrates the country’s dedication to fostering AI technological development. For innovators in the AI sector, China not only offers promising business prospects but has the potential to emerge as a pioneer in the protection of IP.

Hengwei Zhou is a patent attorney at CCPIT Patent and Trademark Law Office. He can be contacted at: zhouhw@ccpit-patent.com.cn