Although China is an appealing market for the commercialization of AI related algorithms, there are challenges. Many of them are related to obtaining IP protection for algorithms in China. Lack of IP protection will expose the development to theft, infringement and misuse by Chinese competitors and it may result in huge economic losses for the developers. Navigating these challenges in the proper way will be key to the selection of the appropriate business model for the exploitation and commercialization of the algorithms (e.g. licensing, assignment, JVs, cooperation and co-development etc.).
This is a guest contribution to the Ethics Dialogues blog, adding perspectives to work conducted in the Human Brain Project’s work package on adaptive networks for cognitive architectures: from advanced learning to neurorobotics and neuromorphic applications (more specifically in task 3.8 looking at the application of RRI to AI and bio-inspired modelling) on the challenges and processes required to ensure responsible international transfer practices of AI related interventions and technologies developed in the HBP and EBRAINS. Dr. Beconcini is the head of the China IP and counterfeiting team of Squire Patton Boggs, a global law firm. Paolo is based on the firm office in Los Angeles but he lived in China for 18 years and his team members are located in the Beijing, Shanghai and Hong Kong offices of the firm . Paolo has headed the firm’s China Intellectual Property team, where he has managed the trademark, copyright and design portfolios of European and US clients, conceived and implemented IP litigation strategies, regularly appeared in Chinese courts and attends IPR administrative and police raids on counterfeiters. Paolo has contributed to the inter-disciplinary discussion in the HBP around the ethical and social challenges that can arise in the context of the exploitation, commercialization, and international technology transfer of AI-related research outcomes and inventions from the HBP.
In the West, China’s AI policies are seen more as a cover for employing AI to curb freedoms and control society, However, the reality is that China is also keen to develop a thriving market for AI applications in both the trade and industrial sectors. An example is the most recent policy paper from the Chinese Ministry of Technology and Innovation “Guiding Opinions on Accelerating Scenario Innovation and Driving High-quality Economic Growth with High-Level Applications of Artificial Intelligence”(in Mandarin 关于加快场景创新以人工智能高水平应用促进经济高质量发展的指导意见). These “Guiding Opinions” aim at providing concrete support in the creation of plausible AI employment scenario and bring businesses together in such projects. They also make it clear that major scenarios will be created around the cultivation of high-end and efficient smart economy, the construction of a safe smart society, high-level scientific research activities, major national events and major projects. Encourage in-depth exploration of artificial intelligence technology application scenarios in key industries such as manufacturing, agriculture, logistics, finance, commerce, and home furnishing, and continue to explore opportunities for artificial intelligence application scenarios in areas such as urban management, traffic management, ecological environmental protection, medical health, education, and elderly care. There is indeed practical concern that the development of AI in scenarios such as urban management may present problematic aspects of societal control. Where data gathered, monitored and tracked by sensors, cameras, mobile phones and cashless payment applications could be used for policing and surveillance purposes.
Incentives and government investment plans have been devised to support this new market and incentivize start-ups, thus creating an environment full of investment opportunities, R&D cooperation, and AI algorithm commercialization in China. There is no doubt that China is at the forefront of AI development in voice and visual recognition technologies with companies such as iFlytek being a world market leader , and they are indeed finding numerous commercial applications. China’s most famous search engine Baidu makes extensive use of such technologies. Large amounts of available data, lower enforcement standards when it comes to data and privacy protection, may further drive AI deployment opportunities and commercialization. The downside is that incentives and further developments are strongly linked to governmental support. Although leading AI innovators are private companies such as Huawei, Alibaba or Tencent, government institutions provide substantial financial and infrastructure backing, as well as policy guidance for the practical implementation of AI algorithms developments. This increases the concerns of foreign developers and investors about the ultimate use of such technologies and whether a fair playing ground really exists in the Chinese market. An example was the revelation that iFlytek, world leader in voice recognitions AI systems received subsidies from the Chinese government for an amount higher than half of the company’s yearly net profits. China’s major economic policies from Made in China 2025 onward have incentivized the creation of a more self-relying supply chain and the creation of innovating leading brands and technologies that will ultimately be self-sufficient. At the same time, the Chinese government knows very well that foreign knowledge and investments are key to China’s future AI development and independence. The same happened with other industries in China before. This tension between contradictory interests will affect how AI policies will be practically implemented and will likely lead to disparity of treatment between domestic and foreign investors and stakeholders, toleration of IP theft and heightened risk of forced technology transfer.
In sum, aside from the ethical doubts connected to commercializing or co-developing AI algorithms in China in relation to human rights abuses or dual use (civil/military), there is also another level of complexity that an AI developer must consider: the Chinese Intellectual Property System. Obtaining IP rights protection is the only way for a smoother and effective commercialization of an algorithm and its related software and will also affect the model of business exploitation of these inventions. The Chinese IP system and China’s business practices do pose strategic risks for the inventors and right holders when they seek to secure proper IP protection and effectively exploit their inventions. This risk is particularly acute in case of international cooperation projects involving Chinese businesses, research institutions, or even the Chinese government.
Dealing with copyright and patent co-ownership over algorithms
Ownership of IP rights over algorithms is always a tricky issue in China. Even before dealing with such issue, an inventor must select the appropriate type of IP protection for its AI-related algorithm. The choice is between patents and copyright. Art. 25.1(2) of the Patent Law of China expressly prohibits the patenting of rules and methods for mental activities because they are not created to solve a technical problem. Pure algorithms or mathematical rules are therefore excluded from patentability. These algorithms would normally be protectable by registering a copyright in China. However, patenting an AI algorithm would be much more desirable, in that patents offer more effective protections than copyrights against infringers and reduce the playroom for competitors to invent around the patent. However, if an AI-related algorithm is proved to be capable of solving a technical problem by employing technical means that produce a technical effect, such AI algorithms can be registered as a patent in China. Recent statutory amendments have introduced specific rules for the examination of AI algorithms by the Chinese patent examiners (outlined in Section 6 of Chapter 9 of Part II of the China Patent Examination Guidelines as amended in August 2021). An example of patented AI-algorithms is those developed by Baidu to run health screening at airports or other public places during the zero-Covid policy. In such cases AI algorithms were used to power measuring machinery such as scanners or thermometers .
Patenting AI inventions remains challenging because the applicant, when drafting the patent application, will have to focus on structures rather than functions. For example, in case of a supervised learning model for image recognition, the applicant will have to describe the relation of the algorithm to input devices such as sensors, or scanners, where data are uploaded and stored afterwards and, most importantly, how it comes to its final image choice and how it labels it. The same questions arise with other AI learning processes. This specific aspect, although related to US patents, has been very well explained by Wen Xie in Forbes Magazine.
If an AI-algorithm is not eligible for patent protection, it can still be protected by registering a copyright. China requires registration of copyrighted software for it to be enforced and exploited in China). This means that an EU or US registration will not be sufficient to enforce the copyright in China. The registration process in China requires the disclosure of the source code. Foreigners are often reluctant to provide the source code to Chinese examiners at the Chinese IPO (known in English with the acronym of CNIPA) and opt to keep the matter secret due to the risk of forced technology transfers (we shall see below what this entails). Registering a copyright or a patent for the algorithm, will be the best way to protect it against theft and infringement and to secure better licensing deals and the payment of the ensuing royalties in China. An unregistered Ai algorithm and software will be easy prey to theft and misuses and will cause an obvious economic damage to its developer.
Once the developers know what rights they can file in China for an algorithm they must decide who owns it. This is a complex issue when there are multiple inventors and developers from different countries and working for different legal entities. China has its own ownership rules, and they can be complicated especially if no detailed written agreement was concluded among the co-developers to supersede them. Specific rules exist for instance, to regulate ownership for a copyrighted work created during the course of employment or by using the resources of the employer. In such cases, the copyright is owned automatically by the employer by operation of law (Art. 18 China Copyright Law). Another case is that of copyrighted work created upon commission or hire. Unless otherwise regulated by a separate agreement among the parties, the copyright is owned by the commissioned party by operation of law (art. 19 Copyright Law). In case of a work created jointly by two or more authors, copyright of the work is jointly enjoyed by those authors (Art. 14.1 Copyright Law). Co-authorship cannot be claimed by anyone who has not participated in the creation of the work. Therefore, joint ownership of copyright cannot be created by agreement (Art. 14.1 Copyright Law). If a jointly created work can be divided for separate uses, each author can enjoy an independent right as long as its exercise does not infringe the entire copyright (Art. 14.3 Copyright Law). It is also important to remark that without a derogating agreement; each right holder has the same right to the work as a whole and any act of exploitation he intends to carry on over the right must be agreed by the other co-owners (Art. 14.2 Copyright Law). If the parties want a co-ownership by shares, this must be stipulated by an agreement. Given the above complexities, it is always recommendable to have a co-ownership agreement in place before filing a copyright registration for a software in China, especially when the development of the algorithm is the result of complex R&D constellations with multiple developers and researchers, involving Chinese parties.
Protecting trade secrets
Some algorithms may not be mature or suitable for copyright or patent registration. In other cases, their developer simply prefers secrecy to disclosure. In all these cases, algorithms can be protected as trade secrets. Trade secrets are good as long as they stay secret. The moment they are disclosed or published they are no longer protectable. The right holder’s only redress will be to stop further spread and to obtain damages compensation and injunctions for cessation of use by infringers and third parties. That is why it is important to educate developers about secrecy, especially when dealing with Chinese research and academic institutions, as well as with state owned labs and enterprises. These cooperation and R&D situations always present risks for a foreign IP holder. Secrets are stolen at meetings, during dinners and ever more frequently, through hacking of phones and computers. The most recent example of theft, this time in the west by infiltrated Chinese researchers, is that attempted against General Electric.
However, most secrets are stolen from unfaithful employees and business partners. I have handled several cases where employees of the client Chinese subsidiary had been unfaithful and shared company secrets with Chinese competitors. Although many of these cases originate in the occasional greed of an employee, sometimes Chinese subsidiaries were purposefully infiltrated by trade secret thieves with the expressed purpose to steal certain secrets to obtain competitive advantages. Therefore, in light of all these challenges, the first and most important job of an AI-algorithm developer is to set up processes that allow to reduce such risk and timely detect instances of theft as well as to have legal tools to address liabilities and damages. Safety processes inside the company or research institute must limit and track access to secret information and documents, while cooperation, development (co-development) and employment agreements must be drafted in a way that is suitable for the relevant markets so as to provide a legal cushion and reduce the burden of proving a violation.
Training of employees and partners dealing with Chinese counterparts (either from the private or public sector) is also a critical piece in the prevention strategy. Leaving laptops in a meeting room while going for lunch is never a good idea. The same is when your party is left alone in a room for internal talks. A voice recorder may be left behind in a bag on a nearby chair. Informal chats about the project at the hotel bar with a likable counterpart is also a no go. Such rules must be laid out for every carrier of the company secrets when working in cooperation with Chinese counterparts. Prevention is the best way to reduce the fallout from trade secret theft. Litigating trade secret theft in China will not only be challenging but also expensive. This said, if it comes to trade secret litigation in China, a foreign right holder should not shy away from such challenges. There are strong legal tools available and, if well engineered, they can lead to a successful case. It is not a good strategy to give up litigation in China when that is the best and only way to stop a leak and obtain compensation. Doing nothing or initiating palliative actions in the US or the EU would only show weakness. Foreign judgments would not find likely execution in China and would leave the infringers free to continue infringing. If you want to learn more, I have written about this in my essay The State of Trade Secret Protection in China in Light of the US-China Trade Wars.
A note on forced technology transfer and ethical challenges in China, and double or dual use of technology:
One of the recurring US and EU grievances that have led to recent WTO infraction proceedings against China and ushered the trade and tariff war between China and the US is that China uses administrative procedures to force foreign investors and IP holders to transfer secrets and know-how to Chinese private or governmental entities . For example, in an import proceeding a Chinese custom official or an official with any governmental agency in charge of homologation or inspection could make the issuance of an import license, or a homologation (the administrative process by which a government authority verify whether a product complies with the national mandatory quality standards and approves its commercialization) or a permit dependent on the foreign right holder disclosure of proprietary information and document, beyond what is normally required by the law. Even after the Economic and Trade Agreements signed by China with the US (2018) and the EU (2019) where China committed to eliminate and reduce such practices, nothing has really changed. Foreign right holders are always at risk of being extorted know how in exchange of an otherwise legitimate right to obtain permission to conduct business in China. If you want to know more about forced technology transfer I can recommend my book on trade secret protection in China, or a blog post from January 2021 on international challenges that can help China and the EU find an agreement on technology transfer.
It is evident that this affects software and algorithms as well, especially when related to sensitive technological areas whose development is prioritized and strictly controlled by the Chinese government, as it is surely the case with Brain-AI.
Trade secret theft and forced technology transfer are also linked to ethical issues. It is clear that the current major areas for AI algorithms development in China re focused on military and security/police use. Brain-AI is at the intersection of civil and military employments and the instances of theft or forced transfer may be directly or indirectly sanctioned by the government in order to employ the stolen technology
So how should we deal with intellectual property related risks for AI algorithm developments and their exploitation in the Chinese market?
My conclusion is that, while China is an appealing market for the commercialization of AI related algorithms, there are challenges. Aside from the ethical ones, which are only peripherally discussed here, there are several challenges related to obtaining IP protection for algorithms in China. Lack of IP protection may expose the development to theft, infringement and misuse by Chinese competitors and it may result in huge economic losses for the developers. Navigating these challenges in the proper way will be key to the selection of the appropriate business model for the exploitation and commercialization of the algorithms (e.g. licensing, assignment, JVs, cooperation and co-development etc.). Before entering into cooperation agreements with businesses, developers and government institutions, including R&D centers or universities in China, foreign right holders should conduct proper due diligence of their future partners and the related projects, should be secured by registration of any relevant IP (Patents and trademarks in primis) and written agreements to ensure that joint ownership, licensees, pledges or transfer of IP rights derived from the cooperation is properly regulated in order to avoid surprises from the application of unfamiliar Chinese laws and regulations. At the same time, internal safety procedures should be put in place to reduce the risk that secrets leak. Documents must be properly water-marked and classified; clearances implemented in order to determine the flow of information at its possession at all times. Employment or cooperation agreements must ensure confidentiality and proper non-compete-clauses to avoid risk of theft by employees or other unauthorized parties. Additional software and technical measures are also encouraged to track and monitor flow of information and data. If a secret is stolen, or the ownership or its proper apportionment are disputed, or if a third party is trying to copy the software containing the algorithms, in all these cases, the right holders must not shy away from taking all the proper enforcement measures. Given that China has no discovery system like in the US litigation, infringements must be thoroughly investigated by the right holder, evidence secured in the proper forms so that it can be admitted before a Chinese court or accepted by the Chinese police in case of a criminal action. Litigating in China is a viable option and there are competent and experienced Specialized IP courts that can handle very complex patent and copyright infringement matters. AI-algorithms can be successfully protected if all these measures are taken as soon as possible.
Want to know more about this?
- Guiding Opinions on Accelerating Scenario Innovation and Driving High-quality Economic Growth with High-Level Applications of Artificial Intelligence (in Mandarin 关于加快场景创新以人工智能高水平应用促进经济高质量发展的指导意见).
- Wen Xie, How To Patent Artificial Intelligence And Machine Learning Models, Forbes 29 August 2022
- Leo Cheu, Chinese government support benefits domestic AI companies, USCDornsife Global Policy Institute, 8 February 2022
- Victor Liang, Baidu’s AI-related patented technologies: Doing battle with COVID-19, WIPO Magazine, June 2020
- Nicholas Yong, Industrial espionage: How China sneaks out America’s technology secrets, BBC News, 16 January 2023
- Paolo Beconcini, The State of Trade Secret Protection in China in Light of the US-China Trade Wars, University of Illinois Chicago Review of Intellectual Property Law, Volume 20, Issue 2, Winter 2021 pp. 108-144 (https://repository.law.uic.edu/cgi/viewcontent.cgi?article=1490&context=ripl)
- World Trade Organisation’s Dispute Settlement DS542: China — Certain Measures Concerning the Protection of Intellectual Property Rights. 9 June 2021