In 2020, a machine-learning algorithm helped researchers to develop a potent antibiotic that works against many pathogens (see Nature https://doi.org/ggm2p4; 2020). Artificial intelligence (AI), which is used for vaccine development, drug design and materials discovery, ship design, space technology, and ship design, can also be used. Many inventions that use AI could be made within a few years. This is one of the greatest threats to patent systems.
Patent law assumes that inventors are humans; however, it is difficult to handle an inventor who is a machine. This problem is being addressed by courts around the globe as more than 100 patent applications naming an AI-based system as the inventor are being filed. Public consultations are being held by several groups on AI and intellectual Property (IP) law. These include in the United States, United Kingdom, and Europe.
The implications of a court or government ruling that AI-made inventions can’t be patentable could be devastating. Businesses and funders would be less motivated to invest in useful research with AI inventors if the return on their investment is limited. The development of life-saving and worthwhile inventions could be lost on society.
Instead of forcing patent laws to adapt to new technology, we suggest that national governments create a bespoke IP law (AI-IP) that protects AI-generated inventions. To ensure that all laws comply with standardized principles and that disputes can be settled quickly, nations should also establish an international treaty. Both steps must be communicated by researchers.
The Venetian Patent Statute of 1474, which was the first global patent legislation, did not consider machines that could invent. They were not even mentioned in the 1883 Paris Convention for the Protection of Industrial Property which established the foundations of the international system of patents. When the World Trade Organization (TRIPS) finalized its Agreement on Trade-Related Aspects of Intellectual Property Rights in 1994, AI-generated inventions had not been seen before. Today, international patent standards are mandated by the 1994 and 1883 treaties.
The TRIPS agreement protects “any inventions, whether products or processes, in all fields of technology, provided that they are new, involve an inventive step and are capable of industrial application” (see go.nature.com/3n4khc2). It states that “inventions”, “new”, “inventive step”, and “capable of industrial use” are terms of art. Each term has a legal definition. If any of these requirements are not met, then an object is not patentable. (See ‘What is patentable ).
In order to be patentable, an invention must satisfy the following requirements:
* A invention that was created by one or more inventors. This can include products, processes, or methods in nearly all areas of technology.
* Novel. * The invention is not yet known.
* Non-obvious or inventive step. A person skilled in the art would not know that the invention is possible.
* The invention is suitable for industrial use or utility. The invention is possible to be used in industry and can also be claimed as having economic significance.
These principles must be adhered to by all 164 members of the World Trade Organization (WTO) as standardized in 1994’s Agreement on Trade-Related Aspects of Intellectual Property Rights.
The system has been challenged before by new technologies. Some high-profile cases have examined whether patents could be granted for genetic sequences, human-made organisms, and other objects. These cases were centered on the question of whether inventions are possible. For example, after a years-long court battle between the US Association for Molecular Pathology (among others) and molecular-diagnostics firm Myriad Genetics in Salt Lake City, Utah, the US Supreme Court concluded in 2013 that isolated human gene sequences were unpatentable because genetic information is a product of nature rather than a human invention2.
Because AI-generated inventions pose a challenge to the patent system, the question is not about “what was invented” but rather “who did it”. Patent registration offices have had to confront the most important question regarding such inventions: Does the inventor need to be human? One concern is that AIs may soon become so prolific that they overwhelm the patent system with their inventions.
Yet another challenge is more fundamental. A person skilled in the art is not able to recognize an invention as ‘inventive’. This is called an ‘inventive stage’. This hypothetical person is a normal expert in the relevant technical area, with an average skill level and general knowledge. Patentability is one step closer if the patent examiner finds that the invention was not obvious to the hypothetical person.
However, if AIs are more skilled and knowledgeable than everyone else in the field, it is not clear how a human patent examiner would determine whether an AI’s invention is obvious. A system that is designed to examine all information about a technology area before it invents will have a greater knowledge base than any human. Everything would be obvious if it was compared to all existing knowledge. If all people have access to these AI tools in the future, then the criterion for patentability ‘inventive steps’ would be nearly impossible to meet and virtually nothing would be patentable. It would take a complete rethink.
An AI system called DABUS (Device to the Autonomous Bootstrapping of Unified Sentience) was created by Stephen Thaler. Thaler is also the president and chief executive of US-based AI company Imagination Engines. Thaler claims DABUS has invented a new type food container and a flashing light to attract attention in emergency situations.
These inventions are not surprising. They are not surprising. The international legal team of Thaler, headed by Ryan Abbott, an academic at the University of Surrey, in Guildford, UK began submitting patent applications to offices all over the globe in 2018. DABUS was named as the inventor. These cases are believed to be the first to determine whether an AI system is eligible to be considered an inventor under the existing laws. This question has been a subject of patent offices and courts, which have begun to highlight gaps in the law.
The applications from the United Kingdom, United States of America, Europe (in both European Patent Office and Germany), South Korea and Taiwan, New Zealand, and Australia have been rejected by patent registration offices. Challenges to these decisions have for the most part failed, with courts concluding that inventors are presumed to be human (see, for example, go.nature.com/3fjwd9n). Germany’s court agreed that patentability could be possible if Thaler was named the inventor who prompted DABUS into creating the inventions. This compromise acknowledged Thaler’s contribution. However, at this stage, the tide is almost against patenting AI systems as inventors.
Patent registries and judges are currently required to interpret and apply existing law in order to determine how AI-generated inventions should be assessed. This is not ideal. It would be more beneficial for governments to develop legislation specifically tailored to AI inventions. This goal can be achieved in three steps, according to our proposal.
The first step is to conduct a systematic examination of the evidence, viewpoints, and issues facing national governments and multilateral agencies involved in patent policy (like the World Trade Organization). They should consult all stakeholders, including national governments and multilateral bodies involved in patent policy (such as the World Trade Organization), professional bodies representing scientists and engineers, consumer advocacy groups, bodies for business development/commercialization and IP law professionals. In response to the rise of the internet and digital economy, previous inquiries of this type have resulted in changes to many countries’ IP laws. The Australian government’s 2014-18 public consultations on online piratery resulted in laws that allow courts in Australia to block access from websites that infringe copyright.
Several countries have already begun preliminary investigations of this kind relating to AI-generated inventions, as has the European Union (go.nature.com/3j6qgu3) and the World Intellectual Property Organization (go.nature.com/3nc79cr). This is a great start and something that all patent systems should follow.
These questions must be rooted in the basics. They should assess whether IP protection for AI-generated inventions encourages society to produce useful inventions, just as for patentable goods. Some IP protection can be obtained by programmers of AI systems through copyright in computer code and patents on the functionality of the software. For political or pragmatic reasons, some people might prefer that AI-generated inventions remain in the public domain. Others call for IP protection to increase (refs 6-8).
There are many grey areas that can be created by tinkering with existing law protections. It is better to have a more comprehensive law reform. A sui generis law, a unique form of IP that governments can create, is an ideal solution. These custom-built laws cover creative outputs that are not covered by the ‘big 4’ IP doctrines (copyright, industrial design, trademarks, patents, and patents. They are already used to incentivize investment in circuit layouts, new varieties and plants, and in some cases, databases.
There are some critics who might be concerned about the proliferation of IP that is topic-specific. A distinct AI-IP doctrine could have the advantage of being tailored to the conditions under which AI creativity takes place. Legislators might decide that AI-IP should be protected for a shorter time than standard patents’ 20-year terms. This would encourage others in the future to improve upon inventions that have expired. Patents are usually awarded to the inventor. However, legislators could decide to split the rewards of an AI-generated invention between the AI developer, the AI director, and the data owner.
Countries with AI-IP would attract more investment in research and technology. However, royalty payments for the invention may make it less accessible. Today, the same thing happens with vaccines and drugs: while patents may attract investment that allows them be developed, people in countries that cannot afford them or can’t pay royalties for their manufacture lose out. One reason is that only 16% of low-income people have received at least one dose of COVID-19 vaccines 18 months after they were first made available. This could be due to patent restrictions. Some higher-income countries are now offering fourth doses (see Nature 603, 764 and 202). AI-IP must be balanced to avoid inequity.
An importer country might find it beneficial to not protect AI-generated inventions. Instead of paying royalties, it could sell cheap copies of a new expensive drug an AI has created elsewhere. It might not attract industry-building investment.
This ‘freerider problem’ is addressed by patent treaties. Because countries can’t opt out of IP protection through patents, the TRIPS agreement was created. International conventions govern the use and protection of designs, trademarks, and copyright, as well as other IP areas.
An international treaty for AI-generated inventions is also essential, according to us. It would establish uniform principles for protecting AI-generated inventions across multiple jurisdictions. It could be accomplished by either negotiating a new international IP treaty or by adding these rules to an existing one.
This treaty would be a long-term, ambitious plan. Critics might object to countries losing the ability to decide their own domestic policies regarding AI-generated inventions. This is especially true when the full potential of AI remains unclear. Because it would prevent disputes and uncertainties down the road, we believe that the international benefits of an agreement would outweigh the cost and time it takes to negotiate.
A treaty on AI-IP should avoid a major limitation to the patent system. Patents are separately registered in each country. Enforcement disputes must be settled by the law of the country where they were registered. This can result in patent holders running similar legal cases in many countries — as when the technology firms Apple and Samsung spent 7 years battling more than 50 lawsuits about phone and tablet design and functionality (see go.nature.com/3lfzpej). This system is cumbersome and costly, so some people can’t afford to defend or enforce their rights under the patent law.
An AI-IP treaty might instead include dispute-resolution mechanisms. This could be adjudicated perhaps by an international specialist court. Inspiration can be found in Europe’s new Unified Patent Court, which is due to commence in the next year (www.unified-patent-court.org), and various arbitration courts around the world.
It will be difficult to create a bespoke law or an international treaty, but it will not be impossible to not make them. AI is revolutionizing science and inventing. To ensure that it serves the public interest, we need a fit-for-purpose IP Law.
