As the world wakes up to the potential of AI, China’s patent office has ambitions to become the world’s leading industry hub, write Lusheng’s Landy Jiang and Terry Lu
As a key sector of novel productivity, AI technology has been seen as a new engine for China’s economy. In 2017, the Chinese government enacted a national plan for the development of state-of-the-art generative AI, aiming to develop an AI industry worth more than $140 billion by 2030.
Within six years, China had become a top two country of origin for notable machine learning models (the core tech in AI). This setting developed 15 notable models in 2023 according to Stanford University’s 2024 AI index report. Today, ostensibly, China is the world-leading country in AI research and innovation.
The surge of domestic AI innovation has raised new questions for the current IP protection system, especially the patentability of AI-related technology. For example, many AI patents relate to new algorithms or models which, in essence, belong to mathematic methods.
To offer a patent-protecting level compatible with the pace of AI innovation, the China Intellectual Property Administration amended the Guidelines for Patent Examination (GPE) at the end of 2023, which became effective on 20 January 2024. The amendment improved the granting standard for software patents in the AI sector.
It is argued that this new GPE broadens eligible subject matters related to AI technologies and makes inventiveness standards better accommodate the situation of AI R&D processes.
In this article, we shed light on the outcomes generated by the new guidelines.
The new guidelines appear to favour granting more AI patents
Machine learning is a fundamental branch of AI, and its related patents are mostly software orientated. We studied the number of granted machine learning patents within the six months of 2024 and compared it with the same period in 2023 to reveal the influence of the new GPE.
Figure 1 shows that the total number of machine learning patents granted between January and June 2024 increased by 30% compared to the same period in 2023, from about 94,000 to 127,000. In particular, the granting number almost doubled during the first three months after the new GPE enactment. This figure indicates that the new GPE is fostering patent protection for AI innovation.
Figure 1: Monthly filing number of machine learning patents in China between January and June of 2023 and 2024
Figure 2: The number of machine learning patents granted in China before and after the new GPE become effective
More AI patents to be granted with ‘computer program product’ claim
Before the GPE amendment, the ‘computer-readable medium’ defined by process features of a software was a common subject matter for product claims. At that time, the ‘computer program product’ claim was rarely granted as the software-related solution to be combined with hard devices or be used as a part of the system.
However, the new GPE directly identifies computer program products as the subject matter of claims. That is, the invention mainly related to software innovation can be granted as an independent product claim, it no longer needs to be re-written as a computer readable medium.
According to Figure 2, CNIPA granted more than 1,500 patents in the machine learning field as of 30 June 2024, the number nearly tripled from the total granted number in the whole year of 2023.
It is shown that the new GPE gives the patent applicant more flexibility to structure a proper claim for AI patents.
The new guidelines eliminate some obstacles for granting AI patents
Before the 2023 reform, there were two eligibility issues that may deny AI-related patents from granting:
- The diagnosis method issue: If the AI processes data relating to human health, the processing method would be examined as a diagnosis method even if no doctors were involved during the whole process. So, a subject matter was not eligible for a patent according to Article 25 of the Chinese patent law.
- Whether AI related patents constitute a technical solution: The 2019 amendment of the GPE already addressed that when examining whether the invention was a technical solution and eligible for a patent according to Article 2 of the Chinese patent law, the algorithm features should be considered together with other features in a whole solution. However, in practice, many AI or big data inventions were still rejected because the inventions only used a new algorithm to improve the internal performance of a computer system or the reliability/accuracy of a big data algorithm, which did not mention any application for solving technical problems or the whole solution has a technical effect.
To offer more protection for AI innovation, the new GPE further broadens the eligible subject matters for AI patents, specifically:
- The information processing method where all steps implemented by computers or other devices are no longer considered as the diagnosis method, no matter whether the processed data relates to human health or not. Accordingly, the AI+ health or ‘smart medical care’ inventions that typically use software to analyse human health big data, such as DNA sequencing, and provide a health risk assessment on individuals will be eligible for granting patents.
- AI inventions that do not specify their application, such as deep learning, clustering algorithms and neural networks, are eligible for a patent if the invention can improve the internal performance of the computer system or increase the accuracy or reliability of data analysis. This changed the former rules that required the AI algorithm to be applied in certain technical sectors.
- An AI invention dealing with big data of a specific application is eligible for a patent if it uses technical means, such as classification, clustering, regression analysis and neural networks, to find out the inherent correlation in the data and improve the reliability or accuracy of big data analysis.
The CNIPA is shifting to the new criteria for eligibility of AI patents during the patent examination. We set out three typical cases to illustrate the trend.
Case 1 (reference 201910223555.5)
The invention relates to a prediction method for death due to fatal illness based on an attention mechanism sequential convolutional network algorithm. The method of claims 1 – 9 includes collecting and analysing the fatal illness and surgery multi-source monitoring data, thereby predicting the death risk due to fatal illness by calculating the death risk coefficient. The CNIPA decided this invention was not eligible for patent and then rejected it on 3 August 2021 because the claimed method is working on the living human and belongs to the diagnosis method. The applicant filed a re-examination request and CNIPA dismissed the rejection according to the new criteria in January 2024, ruling that the claimed method is an information processing method rather than a diagnosis method because all steps, including collecting data, modelling, calculating and predicting, are performed on a computer device.
Case 2 (reference 202410501018.3)
The invention is directed to a computing resource distribution method, which uses clustering algorithms to classify computing resources and label the corresponding task type for each computing resource and finally adjusts the resource allocation plan and resource pool size according to the current status data in the resource status database and the predicted data of the prediction model. The results of this invention maximise the utilisation of the computing resources of a computer system, without specifying any application of this algorithm. However, this patent application was directly granted in June 2024 and no office action was issued to challenge the eligibility of this invention.
Case 3 (reference 202410195936.8)
A new patent application filed in May 2024 discloses a dialogue training method based on big data. The invention discloses and uses a multi-level information fusion algorithm to pre-train the dialogue intention recognition model. This algorithm can solve the problem of low recognition accuracy in traditional neural network training methods. Although the invention does not mention or specify the technical meaning of the big data, it is granted directly and the examiner did not challenge the eligibility of this application.
Overall, the new eligibility test for AI patents is more friendly to innovators. The granting rate for AI patents is very likely to increase in a long term.
The new GPE considers an algorithm’s contribution during the inventive-step examination if it improves the internal performance of the computer system. In the 2019 amendment of the GPE, the CNIPA stressed that the algorithm features should be considered in a whole solution together with other technical features, rather than reviewing the algorithm feature separately, when evaluating the inventive step of the patent.
This amendment only gives a common rule for inventive-step analysis on inventions including algorithm features but does not mention how to evaluate the effects or inventive-step of a general algorithm innovation that can be applied in different fields.
The new GPE provides a specific rule for assessing the inventive step of a general algorithm invention. It states that algorithm features should be considered if they work together with the internal structure of a computer system, improve the internal performance of the system and achieve good outcomes, such as reducing data storage or transmit volume and improving processing speed.
The CNIPA has started to use this new rule to re-assess the inventiveness of the invention concerning algorithm features. For example, in a re-examination decision issued in 2024 (Case no 1533508), the CNIPA re-assessed the contribution of the algorithm features on the inventiveness of the invention and overruled the rejection where it was decided that the algorithm features are common selection based on prior arts.
The re-examination decision assessed the function of the erasure code algorithm together with the field programmable gate array (FPGA) in the whole solution that reduces the occupation of CPU resources, thereby improving the input and output rate of the CPU.
Although the rejection implied that the erasure code algorithm was common knowledge in the art, the re-examination decision did not treat the algorithm feature separately from other features and found that no prior art disclosed or taught that erasure code algorithm together with FPGA can increase the speed of CPU.
Soon, the CNIPA will examine AI patents from the function and outcomes of the algorithm features in the whole solution, avoiding any underestimation of AI innovation.
The new GPE introduces the improvement of user experiences as a key factor in the inventiveness assessment.
This is the first time that the GPE expressly requires that the improvement of user experiences should be considered during the inventiveness assessment if it originated from technical features, or the algorithm features together with other technical features.
In practice, when the patent application mentions that the invention improves the user experience, the CNIPA would consider this effect during the examination. However, the implementation of the new GPE will undoubtedly make AI invention more likely to meet the inventive step after user experiences become an effect that must be considered during the inventiveness evaluation over prior arts.
For example, an invention (Case No.1534581) was finally granted after a review of re-examination considering the prior arts did not disclose or provide any technical teaching for using the distinguishing features to improve the user experiences. This invention relates to a data processing method that can be used in a generative AI to answer a user’s questions. It determines whether the target question statement is a valid question and solves the low-efficiency problem of providing proper answers by determining whether the knowledge graph of the target object satisfies the first condition.
The CNIPA dismissed the rejection decision considering that the prior arts 1 – 3 did not solve the low answering efficiency problem nor improve the user experiences.
In summary, CNIPA is trying to keep pace with the rapid development of the AI industry in China via the new update of examination guidelines. In response to the actual needs of AI innovators, the new GPE adjusts the standards for eligibility and inventiveness, making them more consistent with the R&D practice.
Facing today’s increasingly fierce competition in the AI field, the CNIPA is working hard to help China become the preferred jurisdiction for AI innovations around the world.