This content is from: China (PRC)

Protecting AI-related innovations in China

Lin Li, Yujie Jin and Yali Shao of Liu Shen & Associates address the requirements AI inventions must meet in order to be protected under patent, trade secret and copyright law

Artificial intelligence (AI) is a booming area in this high-tech era. It has led to new challenges regarding intellectual property protection of AI-related innovations in many countries. As one of the most active countries in developing AI-related technologies, China has been making a lot of effort to improve IP protection in relation to AI-related innovations. This article addresses the patent protection of AI-related inventions and also other types of IP protection including trade secrets and copyright for AI technology in China.

Patent protection

The latest revision to the Patent Examination Guidelines, which came into force from February 1 2020, focuses particularly on emerging technologies such as AI, big data etc. and provides more specific guidelines on how to get patent protection for AI-related inventions in China. Hereinafter, some main issues the revision addresses, including eligibility, novelty/inventiveness, and sufficient disclosure, are introduced.

Eligibility issue

As defined in Article 2.2 of the Patent Law of China, invention is a new technical solution relating to a product, a process or improvement thereof. To constitute a technical solution, a three-element criteria must be met: a solution should resolve a technical problem by using technical means and achieve a technical effect.

AI-related inventions usually involve improvements in AI algorithms which are implemented in conjunction with computing devices in the form of computer-executable instructions. However, AI algorithms or AI mathematical rules themselves are excluded from patent protection as being rules and methods for mental activities under Article 25 of the Patent Law of China, since they generally do not solve any technical problem and therefore are not technical solutions. But there is an exception for an invention relating to computer programs. That is, if the solution of an invention application relating to computer programs involves the execution of computer programs in order to solve technical problems, reflects technical means by the laws of nature through computers executing programs to control and process external or internal objects, and thus technical effects in conformity with the laws of nature are obtained, the solution is a technical solution and is the subject matter of patent protection.

Many intellectual properties related to AI, including technical manuals, product design drawings, computer software codes, algorithms and platform codes, processed data, etc. can be protected by copyright. However, regarding works created by AI, a lot of questions remain unanswered by copyright laws

The 2020 revision of the Examination Guidelines clearly specifies examination rules and regulations for inventions containing algorithm features (i.e. AI-related inventions).

It is notable that the 2020 revision of the Examination Guidelines has eased the difficulty of drafting claims to meet the requirements of eligible subject matter for patent protection. As long as the claim contains not only algorithm features but also technical features, the claim as a whole does not fall within the scope of rules and methods for mental activities. On the other hand, all the features described in the claims should be considered as a whole, and if the claim solves technical problems by adopting technical means and achieves technical effect in conformity with laws of nature, the solution defined in the claim is a technical solution specified in Article 2.2.

The eligibility of subject matters of AI-related inventions can be further understood from the following examples.

Example 1

A method of establishing a mathematical model, including:

  • training an initial feature extraction model according to feature values in training samples to obtain a target feature extraction model;
  • processing feature values in each training sample according to the target feature extraction model to obtain extracted feature values;
  • combining the extracted feature values and label values into extracting training samples to train an initial classification model to obtain a target classification model; and
  • combining the target classification model and the target feature extraction model into a data model.

Example 1 provides a method for establishing a mathematical model, wherein none of the processed objects, processing procedures and processing results involve any specific application fields, thus the method belongs to the rules and methods for mental activities, and is therefore not an eligible subject matter of patent protection.

It should be noted that, the examination criteria on patent eligibility of AI-related inventions seems to be stricter than before. Some applications containing claims like Example 1 might have passed the examination on eligibility and even been granted. Nowadays this kind of claims is troublesome and will normally be rejected.

Example 2

A training method for a convolutional neural network (CNN) model, the method comprising:

  • obtaining initial model parameters of the CNN model to be trained, the model parameters including:
  • obtaining multiple training images;
  • performing a convolution operation and a maximum pooling operation on each of the training images to obtain a first feature image of each training image on each of the convolution layers;
  • using the model parameters obtained when the number of iterations reaches a preset number as the model parameters of a trained CNN model.

The above claim provides a training method for a CNN model applied in a specific application field, i.e. image processing/recognition. In this example, the data processed in the steps of the training method is image data, reflecting that the training algorithm is related to a specific technical field of image processing. The training method in this example solves a technical problem that a convolutional neural network model can only recognise images of a fixed size and achieves a technical effect of recognising images of any size by a technical means of performing different processing and training on images on different convolutional layers. Therefore, the claim constitutes a technical solution stipulated in Article 2.2.

From Example 2, it can be seen that, in order to meet eligibility requirements of patent protection, it is advisable to incorporate an AI solution seeking protection in specific application fields and whereby a technical problem is resolved.

Novelty/inventiveness issue

The 2020 revision of the Examination Guidelines clearly specifies that AI algorithm features are taken into consideration during the examination of novelty and inventiveness. However, the revision does not specify whether AI algorithm features should be treated as technical features or not.

In terms of novelty examination, all features recorded in the claim shall be considered, both technical features and AI algorithm features.

In terms of inventiveness examination, the technical features and the AI algorithm features that support each other functionally and have an interactive relationship should be considered as a whole. "Supporting each other functionally and having an interactive relationship" means that AI algorithm features and technical features are closely combined to collectively form a technical means to solve technical problems and to obtain corresponding technical effects.

For example, if an improved AI algorithm is applied to a specific technical field and can solve a specific technical problem, it can be considered that the algorithm features and the technical features support each other functionally and have an interactive relationship, and such algorithm features become an integral part of the adopted technical means. Therefore, the contribution of the algorithm features to the technical solution should also be considered when evaluating the inventiveness. As a result, the advantageous effects produced by the joint action of the technical features and the AI algorithm features that support each other functionally and have an interactive relationship are more and more critical.

Sufficient disclosure issue

Article 26.3 of the Patent Law of China provides that the specification shall contain a clear and comprehensive description of an invention so that a person skilled in the art can carry it out. Therefore, for an AI-related invention, the adopted solution for solving technical problems, which may contain both technical features and algorithm features, shall be described clearly and completely in the specification.

One particularity of AI is that the rules for the decision-making of an AI model are not based on logic, but relevance. That is, the relevance of cause and effect, in the AI field. In this regard, the relevance may not reflect the regularity of natural laws. Therefore, in order to avoid the so-called "black box" issue, the relevance relationship between the cause data and the effect data must be sufficiently revealed so that those skilled in the art can reproduce the solution.

When describing in the specification an implementation mode of a solution containing an AI algorithm, the abstract algorithm features should be described in combination with specific application fields, and the definition of at least one input parameter and its related output result should be associated with the specific data in the technical field. Any beneficial effects produced by the combination of technical features and AI algorithm features that support each other functionally and have an interactive relationship should be written about clearly in the specification.

Trade secret protection

Clearly not all AI-related innovations are suitable for patent protection. Some innovations with improved AI algorithms, e.g. AlphaZero, may not have a specific application field that fits into the protection of a patent, and some AI-related IP rights are not protected by the Patent Law, e.g. the data itself. In such scenarios, the AI software owner may consider protecting it as a trade secret.

A critical element of trade secrets is confidentiality. Once the secret is disclosed, the protection of such a secret will be lost and trade secret laws do not prohibit the discovery of trade secrets through fair and honest measures. This requires AI owners to carefully choose the intellectual property to be protected by trade secrets. For example, a module that interacts closely with the internet may not be suitable for trade secret protection because it may be relatively easier to be learnt through reverse engineering.

Another challenge caused by the confidentiality of trade secrets is the difficulty in proving violation and defending rights. The revision of the Anti-Unfair Competition Law of China in 2019 made some changes favourable to right holders of trade secrets in this regard. All of the changes in the 2019 revision, including that the definition of trade secret and the scope of the infringement responsible subjects are expanded, and the situations that violate trade secrets, the penalty/damage, and the burden of proof of the accused infringer are increased, are aimed to enhance the protection of trade secrets. Among others, the newly added Article 32 clarifies the rules of proof in trade secret infringement cases. The plaintiff bears the initial burden of proof for its infringement claim and the defendant is responsible for providing evidence to prove its non-infringement defence. This greatly reduces the plaintiff's burden of proof and provides a solution to the problem of difficulty in proving violation and defending rights.

Copyright protection

Many intellectual properties related to AI, including technical manuals, product design drawings, computer software codes, algorithms and platform codes, processed data, etc. can be protected by copyright. However, regarding works created by AI, a lot of questions remain unanswered by copyright laws, including authorship of AI, eligibility of AI-created works, ownership of the rights in AI-created works, and infringement of AI-created works. A recent case, Shenzhen Tencent v Yingxun, which was decided by the Nanshan District Court of Guangdong province of China tried to answer one of the questions.

Dreamwriter is a robot writer developed and owned by the plaintiff Tencent, a famous Chinese company. By collecting and analysing a large number of financial articles, Dreamwriter has learnt to form article structures according to the instructions of a human creative team, and use data collected from stock markets to finish writing and publishing financial articles within two minutes from the end of the stock market. Tencent sued another company Yingxun for infringing its copyright by copying one of the articles created by Dreamwriter and publishing it on its website without authorisation.

Are AI-created works eligible for copyright protection? This is the first question that the court needs to answer. In this case the court said yes. The court is of the opinion that the specific form of the article involved originates from the creative team's personalised selection and arrangement, so the article involved was generated by the plaintiff's creative team using Dreamwriter software, and the article conforms to the formal requirements of written works and has certain originality, therefore it belongs to the written works under the Copyright Law of China. The court ruled that infringement was established.

This case is particular because human beings had a certain degree of participation in the creation of the works. The question of whether works created solely by AI can be copyrighted and other issues are still under discussion. A piece of good news for AI assets owners is that a generally accepted opinion is that AI-created works should be legally protected to a certain extent because they have distribution value.

Nowadays, AI is not just confined to imagined plots in sci-fi movies, but is applied practically in real life. An advanced and appropriate patent examination system in turn promotes the innovation and development of AI technology. Trade secrets and copyright also provide unique protection for AI IP assets. Even though many AI-related IP protection issues are still under discussion, it is certain that more reasonable and effective IP protection of AI advanced technology in China is to be expected.

Lin Li
Lin Li is a patent attorney who joined Liu Shen & Associates in 2003. He has accumulated experience in various patent-related matters, including patent prosecution, re-examination, invalidation, litigation, patent analysis and client counselling, with a focus on electrical engineering and automatic control, communication and computer science, networking and e-commerce, and artificial intelligence. Li works in the firm’s electrical and electronics department. He is a member of the All-China Patent Agents Association.

Yujie Jin
Yujie Jin is a partner at Liu Shen & Associates. Jin’s practice mainly focuses on patent application, patent invalidation and patent litigation in the technical fields of electronic engineering, communication technology, computer science, semiconductor technology, automatic control, image processing, video and audio codec, the internet, e-commerce and AI.

Jin has handled a large number of cases related to patent applications and examination as well as patent re-examination since 2006, and is good at dealing with difficult problems in patent prosecutions. Jin also has experience in patent relevance and stability evaluation, has engaged in many commercial patent v targeted product analyses, communication SEP v standards analyses, and stability analyses for both commercial patents and SEPs.

Yali Shao
Yali Shao is a partner at Liu Shen & Associates. She is also double qualified as a patent attorney and lawyer, with years of experiences in patent prosecution, re-examination, patent invalidation, patent licensing, IP litigation, and client counselling, with a focus on electrical engineering, automatic control, telecommunications, electronics, computer science and internet and e-commerce.

In recent years, Shao has been leading teams in studies of patent-related regulations and worked on suggestions for revision of the regulations. She has carried out all types of analysis including pre-litigation analysis, pre-invalidation analysis, and searching analysis. She has also worked on patent invalidation, patent infringement litigation relating to standard essential psatents (SEPs) and non-SEPs, and patent protection studies for specific technical fields such as AI, the IoT, and 5G telecommunication.

Shao is a member of both the Programme Committee and Patent Committee of AIPPI.

The material on this site is for law firms, companies and other IP specialists. It is for information only. Please read our Terms and Conditions and Privacy Notice before using the site. All material subject to strictly enforced copyright laws.

© 2020 Euromoney Institutional Investor PLC. For help please see our FAQs.


Instant access to all of our content. Membership Options | One Week Trial