Global Intellectual Property Systems Are Not Prepared for Artificial Intelligence
By Ryan Abbott
On July 6, 2022, the Boards of Appeal for the European Patent Office (“EPO”) published an important decision (J 0008/20) about whether a machine could be an inventor for the purpose of the European Patent Convention. The decision was significant for several reasons, including because whether a machine can be an inventor is critical to whether an invention made by a machine can be patented. That is because, at least in the United States, patents require an inventor that must have “conceived” of the patented invention. Thus, if an invention is made by a machine in the absence of a human inventor, and if a machine cannot be an inventor, then the machine’s output is unpatentable.
The past few years have witnessed courts and patent offices around the world struggling with these issues, due in part to a legal test case brought by an international group of patent attorneys which I am leading called the Artificial Inventor Project. The project involved filing two patent applications for inventions made by an AI in the absence of a traditional human inventor (an “AI-Generated Invention”)—one for an emergency beacon and one for a beverage container based on fractal geometry. The applications first cleared preliminary substantive examination in the United Kingdom for novelty, inventive step, and industrial applicability before they were eventually rejected for failing to name a human inventor. The case is now pending in 16 jurisdictions including the European Union, United States, United Kingdom, Germany, Australia, etc.
Whether an AI-Generated Invention can receive patent protection is going to have a significant impact on innovation and industrial development. AI is increasingly stepping into the shoes of human researchers and automating tasks that used to make a person an inventor. It is doing this, in some instances, without a human being exhibiting any inventive skill. This may be the case where a pre-trained machine learning algorithm is used to solve a well-known problem, such as optimizing an industrial component, and where the algorithm (or another algorithm) is evaluating the AI output for fitness and value. Today, this degree of autonomy may be rare and restricted to certain fields like materials engineering and drug discovery and repurposing, but AI’s capabilities continue to expand.
The history of AI-Generated Works, namely outputs that would be eligible for copyright protection but that lack a traditional human author, is instructive. People have claimed that AI has been generating AI-Generated Works for decades. The United Kingdom has even explicitly provided statutory protection for AI-Generated Works since 1988. But until very recently, whether to protect AI-Generated Works was a toy problem because such works had very little value. A mediocre song or a semi-sensical story generated by a machine was just a novelty. By contrast, this year has witnessed an explosion in AI-Generated Works by systems such as DALL·E 2. It has also involved numerous high-profile sales of non-fungible tokens (NFTs) based on AI-Generated Works.
Meanwhile, the United States Copyright Office has a policy of applying a “Human Authorship Requirement” to copyright registrations, such that they will not register a copyright in an AI-Generated Work. This policy is based on judicial language about the nature of creativity from the Gilded Age. But if AI-Generated Works lack copyright protection in the United States then the people making and using AI to create such works cannot commercialize them the same the way as human generated works. That is a bad outcome if the purpose of the copyright system is to provide incentives to generate more socially valuable works and to encourage the dissemination of new works. It is also a problematic outcome for a global economy that is supposed to have an internationally harmonized intellectual property system. Someone in the United Kingdom causing an AI-Generated Work to be created can sell or license that work domestically, but not in the United States where anyone can use it without permission.
We have not yet seen a similar explosion of AI-Generated Inventions, but it is far more difficult to generate a patentable invention than a copyright protected work. Yet AI is continuing to improve while human inventors are not, which suggests that such an explosion is inevitable.
The Artificial Inventor Project test cases were filed in part to stimulate a dialogue on how the law should deal with AI stepping into the shoes of people and generating intellectual property, and in part to generate guidance for industry and applicants. It was also done to promote the normative position that AI-Generated Inventions should be protected, with AI owners as the owners of any patents made by their systems, because this will encourage people to make and use AI to generate valuable inventions, to disclose those inventions rather than protecting them as trade secrets, and to promote commercialization of new inventions as marketed products.
To date, courts and patent offices worldwide have come to different conclusions about the patentability of AI-Generated Inventions and AI Inventorship. In July 2021, South Africa granted a patent for our AI-Generated Inventions with an AI listed as the inventor and the AI’s owner listed as the owner of the patents. A few days later, Justice Beach in the Federal Court of Australia issued an extensive reasoned decision explaining why an AI-Generated Inventions should be protectable under Australian law and why an AI could be a legal inventor.
Justice Beach’s decision has since been overturned by a full panel of the Federal Court of Australia. That panel adopted a far more textualist interpretation of the Australian Patent Act as opposed to the purposive approach adopted by Justice Beach. The decision is now under submission to the Australian High Court to decide whether to accept a further appeal.
Meanwhile, the Eastern District of Virginia has ruled in our cases that an AI-Generated Invention is unpatentable. That case is under appeal to the Court of Appeals for the Federal Circuit and a written decision is pending.
The recent EPO decision rejected formally listing a machine as a patent inventor but suggested it would be appropriate for the AI’s owner to list themself as the patent inventor then to explain that the AI had actually conceived of the invention in the text of the patent. This displays an impressive level of pragmatism and avoids running afoul of procedural requirements of the European Patent Convention which require a listed inventor to be a natural person. However, it is likely not an approach that could work in jurisdictions such as the United States because of the requirement that an inventive “conceive” of an invention. Other jurisdictions, like the United Kingdom, also require that an inventor must have actually “devised” an invention (as opposed to just owning an inventive AI).
The Board of Appeal decision is noteworthy because of the importance of the underlying issues together with Europe’s importance as a patent jurisdiction. It is also noteworthy because of how the Board attempted to resolve conflicts between textualist and purposive interpretations of statutory law, conflicts between substantive and formalities rules, and conflicts between differing rules of national laws within a regional framework. It offers insight into whether and how existing laws can accommodate technological change while achieving their underlying goals.
In the next few years, how the law treats AI generating creative and inventive output is likely to continue to evolve, with important implications for industry and global trade. But technology is not standing still, and there is a risk that if the laws too far behind technology it will fail to adequately mitigate risks and that we will all miss out on some very tangle benefits. The law should not just be reactive, and how the law deals with AI automating traditionally human activity is going to be a critical issue in many areas.
Written by Ryan Abbott. Ryan is a Patent Attorney, Solicitor, Physician and an Advisor with The Cantellus Group.
These blogs by TCG Advisors express their views and insights. The strength and beauty of our team is that we encompass many opinions and perspectives, some of which will align, and some which may not. These pieces are selected for their thoughtfulness, clarity, and humor. We hope you enjoy them and that they start conversations!