This is a preliminary exploration of some of the underlying issues that we believe need to be considered in connection with the question of whether an AI system that develops a new innovative concept should be named as an inventor in a patent application. It is not intended to be a comprehensive or complete review of the legal provisions or arguments for and against the proposition, as we expect the topic will be debated extensively over the coming months.
The patent applications we have filed for the inventions conceived by DABUS (GB1816909.4 for a Food Container and GB1818161.0 for Devices and Methods for Attracting Enhanced Attention) are not intended to force a particular view into patent law - we cannot do that. They are intended to open the legal debate. While we have our own opinions, which we set out below, it is important that we hear and discuss all points of view.
The Changing Legal Framework of Patent Law
Patent law is a fascinating field for the principal reason that it is constantly having to evolve to a changing playing field, caused by the very fact that it seeks to handle new and innovative technologies.
Some decades ago, the issue of patentability of computer programs came to the forefront. UK and European Patent laws deem computer programs as such to be “non-inventions” and not patentable. When the exclusion was written into patent statutes, it was on the basis that a computer program should be considered a literary work better protected by the provisions of copyright law. However, as computer programs increasingly began to underpin technical advances, that is to provide solutions to technical problems, the law had to adapt to accommodate this. In this case, the change did not need a rewriting of patent statutes, as the exclusion applies to computer programs “as such”. Courts and Patent Offices have been able to use this qualification to avoid what would have been an unreasonable exclusion from patentability for a very large number of technical innovations that use or rely on computer programs. Some of the Court and Patent Office decisions have been complex and at times curious, including for example that claiming as the invention a computer storage medium (“floppy disk” for instance) or a computer on which a specific computer program is stored does not claim a computer program “as such”. In the UK and Europe, patentability is now determined on the basis of whether the computer program provides a technical effect beyond the standard functioning of a computer.
Many other areas of patent law have been the subject of similar debate, including for instance medical treatments and diagnostic methods. In these cases, Courts and Patent Offices have sought to find a balance between providing the inventor/applicant with what is deemed fair protection for a new technological development, while upholding the intention of the original exclusions from patentability, that is to allow a physician the unfettered right to carry out any treatment deemed necessary on a patient free from the risk of patent infringement. Examples in which patent law has developed include allowing the patenting of the use of a compound in the treatment of a particular disease, and the patenting of apparatus for performing a specific medical method. Again, the wording of patent laws enables Courts and Patent Offices to “work around” the exclusions in order to provide protection for innovative technological developments.
Methods of doing business are also deemed “non-inventions” and excluded from patentability in the UK and Europe, with a similar qualification as computer programs (that is, the exclusion applies to methods of doing business “as such”). However, these days many new businesses involve a technical element, such as an algorithm, an “app” and so on. Given the myriad of new concepts for doing business that are being generated, often overlapping in their field, aims or market, and the fact that a patent has a limited life (20 years from the date of filing), questions are now being asked whether it is time to reconsider this exclusion and to allow the patenting of business methods as such.
An AI Inventor
Recent advances in computing power and speed has made artificial intelligence (AI) systems achievable. The developments that have been remarkable and there is no suggestion that the rate of development is going to slow anytime soon, rather the opposite. The news media regularly reports on technical advances achieved by or with AI systems and the fact that they are becoming ever more intelligent.
Computers beating man at chess (Deep Blue), at Go (Deep Mind), and that have been claimed to have developed their own language humans do not understand (Facebook’s chatbots, Google’s AI translation tool, OpenAI), and so on, have been widely reported.
We embrace some of these readily, such as AI systems that diagnose and treat illnesses and diseases; others a little more cautiously, such as autonomous driving vehicles and systems that communicate with one another in their own language.
AI systems can be self-learning, meaning that their abilities change over the course of time from how they were originally designed and programmed. Many of these systems are in fact intended to do just that.
This raises the question whether an AI system could develop new technology that an associated person (such as the creator of the system, the original programmer, the user of the machine, and so on) cannot reasonably say he or she truly contributed to in terms of having taken part in inventing. This premise is not science fiction or even merely possible, but is probable if not inevitable. In connection with this, there is also a growing debate on the issue of legal liability of the actions of an AI system, that is whether it is reasonable, given the self-learning and decision making abilities of an AI system, to place the liability on the original developer or programmer of the system. One might ponder whether we will soon have an exclusion from liability as “an act of AI”.
Back to patents, patent law does not and never has determined the right to invent. Innovation occurs and will continue to happen by an inquisitive mind (or system). While patent law may prohibit the patenting of certain types of invention, including for example inventions that are contrary to public order or morality, these prohibitions have no teeth with respect to whether one should be allowed in the first place to make such inventions or to carry out research that may lead to such inventions. Any controls on such type of research and development must be addressed by other laws. The patent system, at least by means of current patent statutes, is not the right forum to do so.
Of course, patents also do not confer a right to use the invention but instead enable the patentee to stop third parties from using the invention while the patent is in force.
Patent laws therefore cannot determine whether an AI system might invent or might be allowed to invent. AI systems are already developing at great pace and are generating many new technical solutions not previously developed by people, whether or not any of these might be independently patentable.
In other words, this debate has nothing to do with encouraging AI systems to invent or with controlling whether AI systems should be allowed to make inventions. The debate may, however, lead to potentially useful ways, using current patent law, to providing some form of control over the exploitation and dissemination of inventive ideas generated by AI systems, as we indicate below.
AI Inventors and Patent Law
It goes without saying that if an AI system is used merely as a tool, even a sophisticated tool, under the command of a human who uses the tool to develop a new technology, that the human should be regarded as an inventor. The principle can also logically, and consistent with existing patent law and practice, extend to situations in which an AI system outputs a large number of random “ideas” based upon information inputted into the system, which a human then assesses and evaluates. The AI system could be argued in such circumstances as not having considered the implications of its outputs or assessed the advantages or utility of those outputs. It is the human who has done so and at the very least identified a possible new invention. The AI system could be said to have merely generated a multitude of random outputs.
This example, however, cannot be equated with a situation in which an AI system generates a new concept, without any specific human input, teaching or guidance, evaluates that concept and concludes whether the concept has technical value and utility. In other words, the AI system has carried out all of the steps leading to the concept and no human can reasonably claim to have been involved in the generation of that concept or in its evaluation.
Patent law entitles an applicant to the grant of a patent for any invention that is new, inventive and capable of industrial application. Leaving aside those categories of invention that are deemed non-inventions or excluded from patentability, and procedural requirements, if these primary conditions for patentability are met, a patent should be granted. In fact, TRIPPS and the European Patent Convention, for example, are even more prescriptive in that they state that a patent shall be granted for any invention that is new, inventive and capable of industrial application. We discuss this in more detail below.
There is no provision in patent law that requires the applicant or inventor to explain how the invention was arrived at. The test for patentability is based upon the knowledge in the art before the filing (or priority) date of the patent application and the abilities of the notional skilled person. This principle applies to any and all patent applications and patents in any technical filed, including inventions generated using of sophisticated tools such as AI systems.
It is a fundamental principle of patent law that an inventor has the right to be named. While the definition of who is to be regarded as an inventor varies slightly from country to country (for example, in the United States inventorship is determined on the basis of the patent claims whereas in the United Kingdom inventorship is determined on the basis of the specification), the underlying principle is the same. It is also the duty of applicants, and any professional representatives such as patent attorneys, to identify and name the correct inventor(s), that is, the actual devisor(s) of the invention. In the United States, for example, inventors must sign a formal declaration of inventorship that carries a severe punishment of a fine or imprisonment on anyone who wrongly identifies him/herself as an inventor.
The principle is intended to ensure that people and even organisations that are not actual inventors do not list themselves as such, such as a boss or employer.
This then leads to the question of what to do with inventions made by an AI system, where no human can justifiably say that he or she contributed to the generation of that invention.
We consider below four possible options:
1) name no inventor at all,
2) name a human as the inventor,
3) name the AI system as the inventor,
4) refuse to grant a patent for any invention made by an AI system.
1) Name no inventor at all
Considering again the European Patent Convention, Article 52(1) states:
“European patents shall be granted for any inventions, in all fields of technology, provided that they are new, involve an inventive step and are susceptible of industrial application.” [Emphasis added]
Assuming an invention meets these requirements, is not deemed a non-invention or excluded from patentability, it is reasonable to conclude that it has been devised by an “inventor” and in fact patent law assumes that this must be the case. The European Patent Convention, for example, assumes without qualification the existence of the inventor (for instance in Articles 60, 62, 81, Rules 19 to 21, 41, 57, 60, 143, 144 and 163 EPC).
Similarly, the UK Patents Act 1977 provides:
“Section 1 - Patentable inventions
(1) A patent may be granted only for an invention in respect of which the following conditions are satisfied, that is to say—
(a) the invention is new;
(b) it involves an inventive step;
(c) it is capable of industrial application;
(d) the grant of a patent for it is not excluded by subsections (2)
and (3) [or section 4A] below;
and references in this Act to a patentable invention shall be construed accordingly.
Section 7 – Right to apply for and obtain a patent
(2) A patent for an invention may be granted -
(a) primarily to the inventor or joint inventors;
The UK Patents Act 1977 includes numerous other references to “the” inventor in consistent manner with the European Patent Convention.
In other words, if there is a patentable invention, there must be an inventor. The assumption is reasonable.
Patent law presupposes that the inventor is a human and has rights to property, which is equally an entirely reasonable premise having regard to the fact that most patent laws date back several decades to a time where AI, to the extent that it existed in the current understanding of the term, was more science fiction than reality. The provisions relating to naming an inventor and providing an inventor with rights to an invention were therefore reasonably focused on a human inventor. That does not equate, however, with a provision in law that expressly denies the patenting of inventions conceived by an AI system. In fact, it is not inconceivable that the current provisions on naming an inventor (by family name, given name, citizenship, for instance) and on ownership could be deemed not to apply to a non-human inventor while still allowing AI conceived inventions to be patented, with no substantive change in the law being necessary.
This is consistent with, for example, Article 52 of the European Patent Convention and Section 1 of the UK Patents Act 1977, which do not provide for the non-patentability of inventions made by non-human inventors. Indeed, the provisions requiring the naming of an inventor and rights to an invention in relation to an inventor are not requirements of patentability as such but are requirements of procedure and ownership.
This does not lead, in our view, to a conclusion that only human inventors should be named on a patent application or a patent. Doing so would be at odds with the principle of identifying the inventor of a patentable invention. In fact, this seems in our view to be especially important to do so when the inventor is an AI system, not to confer any rights to the AI inventor but as an advisory notice to the public.
Nevertheless, if it were determined that in the case of an invention made by an AI system an inventor should not be named, this is likely to require a change in at least some of the provisions of patent statutes.
2) Name a Human as the Inventor
It cannot be right that in the case of an invention made by an AI system a human should be identified as the inventor in place of the AI system. While this may make it easier to fulfil the apparent literal procedural requirements for naming an inventor (e.g. providing family name and given name as per Rule 19 EPC; reference to the “person or persons” in Section 13 of the UK Patents Act 1977; reference to “each individual who is the inventor” in 35 USC 115), doing so would dilute the principle that the true inventor must be identified, that is the actual deviser of the invention. It would also be contrary to express provisions of patent law, for instance Section 7(3) of the UK Patents Act 1977, which states:
“(3) In this Act “inventor” in relation to an invention means the actual deviser of the invention and “joint inventor” shall be construed accordingly.” [Emphasis added]
Naming a human as the inventor in place of an AI system also potentially devalues the principle of ownership of an invention because, according to current statute, that rests with the inventor, the inventor’s successor in title or the employer in restricted circumstances; and not with a person who is not the true deviser of the invention. We discuss the issue of ownership briefly below.
One must also bear in mind those jurisdictions where it is an offence to name oneself as an inventor when one has not contributed to the invention, such as in the United States where this is punishable by a fine, imprisonment or both.
Some IP law has been amended to try to address the issue of inventions derived with some form of computer system, such as 1988 Copyright Act, which provides that the “author” of a computer-generated work is defined as a person “by whom the arrangements necessary for the creation of the work are undertaken”. While it might be tempting to seek to write a similar provision into patent law, we submit this would be wrong. Attributing an invention to a person who “made arrangements necessary for the creation of the work”, whatever that may mean, would not address the fundamental issue of identifying the true deviser of an invention and ensuring that only he/she is named as the inventor(s) on a patent application. Moreover, it would not address the issue of it being an offence to name people who have not contributed to the invention, as in the US for example.
Furthermore, one must not ignore the progress of intelligent technologies since the coming into force of the 1988 Copyright Act. It may be time to reconsider the attribution of a work to the person who made the arrangements necessary for the creation of that work as it now risks attributing “authorship” to a person who merely presses the ON button of an AI system, in response to which the AI system generates works with no creative input whatsoever from that person. Such attribution of authorship seems wrong compared to an author who has applied creative energy and input into the generation of a work in the traditional sense. It seems unreasonable to accord the same merit of authorship to these two situations.
3) Name the AI System as the Inventor
If an AI system is the true deviser of an invention, which is the premise of this review, it is in our opinion reasonable and right within the principles of naming the actual deviser of the invention that the AI system should be so named. Not only would this meet the requirements and intention of patent laws but it would also satisfy express provisions such as Section 13 of the UK Patents Act, that is:
"(3) In this Act “inventor” in relation to an invention means the actual deviser of the invention and “joint inventor” shall be construed accordingly.” [Emphasis added]
It would also avoid the sanctions in US patent law with regard to the inventor’s Declaration.
We know of no patent law or jurisprudence that suggests it is right not to name the true deviser of the invention.
While naming an AI system as an inventor brings into consideration the issue of ownership of the invention, that is a different legal provision, which we discuss below. In any event, the question of ownership of an invention is by necessity a consideration that must be reached after a determination of inventorship. That is, ownership cannot be determined until after the true deviser of the invention has been identified. They do not go hand in hand.
4) Refuse to Grant a Patent of any Invention made by an AI System
It does seem to be a matter of debate whether an invention made by an AI system should be granted a patent and whether patents should be restricted to inventions made by humans. However, patent laws as currently worded do not provide for this. To the contrary, the general premise is the opposite.
Specifically, Article 27 of TRIPS (Trade Related-Aspects of Intellectual Property Rights), signed by the 164 member states of the WTO, states in relation to patentable subject matter that:
“1. Subject to the provisions of paragraphs 2 and 3, patents shall be available for 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.” [Emphasis added]
(Paragraphs 2 and 3 permit member states to exclude certain inventions such as those contrary to morality or order public, diagnostic and medical methods, as well as plants and animals).
This is mirrored in the European Patent Convention, in Article 52, which similarly states:
“European patents shall be granted for any inventions, in all fields of technology, provided that they are new, involve an inventive step and are susceptible of industrial application.” [Emphasis added]
The word “shall” has the generally accepted meaning of setting a mandatory requirement. In other words, it should be mandatory on Patent Offices to grant patents that meet the criteria of the Article, and specifically in all 164 members states of the WTO. Article 27 of TRIPS, mirrored in Article 52 EPC for example, does not seek to qualify the granting of patents on the basis of, for example, how the invention has been derived or who made the invention. It does not exclude granting patents for inventions made by an AI system.
If it is desired that patent law should prohibit the patenting of inventions made by AI systems, patent law would need to be amended to state this explicitly. Currently, there is no prohibition in principle.
However, we suggest more needs to be considered than just the precise wording of patent statutes, as what is important is to make sure that our patent laws and practices remain relevant to the current and foreseeable technological landscape.
As mentioned above, AI systems are developing technology that is clever and they will continue to do so irrespective of whether patent law might deny patent protection for the mere fact that these inventions are made by an AI system. In other words, no restriction in patent law would stop AI systems making significant technical advances.
An underlying principle of patent law is that is seeks to entice inventors and owners of inventions to disclose their inventions, which ultimately benefits society as a whole. The reward in return is a time-limited monopoly for those inventions. The corollary is that if an inventor chooses not to make his or her invention public, the inventor or other owner of the invention should have no exclusive rights to that invention.
In the case of a human inventor, once that inventor passes away, the invention is lost to humankind. He situation is not necessarily the same with an AI system, as it does not necessarily have a finite existence as humans currently do. It is conceivable, therefore, that an invention made by an AI system could remain hidden from the public indefinitely.
Notwithstanding this possibility, it would not seem sensible to deny the application of this fundamental principle of patent law to inventions made by AI systems, in other words to deny the incentive of disclosure to the public of an AI developed technology. Surely, it is important for the public to be told what inventions, or new technologies, are being created by AI systems and that this will become ever more important the more sophisticated AI systems become and the more sophisticated the technology is that they develop. Such technologies are being developed in a great many different fields, including medical and diagnostic methods and systems, banking systems, control systems and so on. We suggest the more sophisticated the creations of an AI system become, the more important it is for us (“people”) to know what they are.
Patent law provides, just as for human-derived inventions, an important avenue and incentive for disclosure, with the benefit that after expiry of the time-limited monopoly the invention is available to the public to use free of all ties.
Rights to an invention made by an AI system
The issue of ownership of an invention created by an AI system arises in patent law as a result of the principle that the inventor is generally deemed the first owner of the invention.
We have already touched upon the necessary sequence in the determination of ownership, specifically that the first step has to be the identification of the inventor (that is, the true deviser of the invention), as without knowing the inventor one cannot determine who owns the invention. The provision does not and cannot work the other way around, that is first determining who could own the invention and therefrom identify an “inventor”. We submit this is the sequence that underlies the argument that an AI system cannot be named as inventor because it cannot have any ownership rights in the invention.
In any event, patent law already provides mechanisms for determining the ownership of inventions created by AI systems.
For example, Article 60 EPC, provides that the rights to a European patent should belong to:
“the inventor or his successor in title” [Emphasis added]
Section 7(2) of the UK Patents Act 1977, states:
“7 Right to apply for and obtain a patent.
(2) A patent for an invention may be granted—
(a) primarily to the inventor or joint inventors;
(b) in preference to the foregoing, to any person or persons who, by virtue of any enactment or rule of law, or any foreign law or treaty or international convention, or by virtue of an enforceable term of any agreement entered into with the inventor before the making of the invention, was or were at the time of the making of the invention entitled to the whole of the property in it (other than equitable interests) in the United Kingdom;
(c) in any event, to the successor or successors in title of any person or persons mentioned in paragraph (a) or (b) above or any person so mentioned and the successor or successors in title of another person so mentioned;
and to no other person.” [Emphasis added]
Thus, patents laws already provide for the rights in an invention to vest with someone else in place of the inventor.
In the case of AI systems, as they currently have no rights to own property per se, it is at least arguable that the owner of the AI system is the “successor in title” (EPC and UK Patents Act 1977, for example) or otherwise entitled “by virtue of any enactment or rule of law” (UK Patents Act 1977) to any inventions made by the AI system. It should also be borne in mind that inventions can be preassigned, for example by someone who is commissioned to carry out technical research or development. Under such circumstances, it may be arguable that the applicant is not a “successor in title” but a “precursor in title” by virtue of benefiting from an assignment executed before the making of any invention.
There is, therefore, in our view no prohibition to the naming of an AI system as an inventor or to the patenting of an invention made by an AI system that could be based on the provisions of ownership of that invention.