Who is the AI Owner? Hacking the Illusive AI Inventorship Challenge
Giuseppina (Pina) D’Agostino discusses the AI Inventorship Challenge in patent law. This is the fifth post in a EGG commentary series exploring how AI’s development is affecting economic, social and political decision-making around the world.
Should a machine be treated like a human inventor and therefore be allowed to be an inventor on a patent? The paper posits that it is not so much who (human or machine or both) invents or even creates in the case of copyright, but who owns the intellectual property and who has the power to commercialize those assets that matters. Tackling the AI ownership challenge will more effectively ensure the public benefits from the intellectual property rights granted.
The Ownership Question
Dr. Stephen Thaler filed numerous patents worldwide for his AI inventing machine, called DABUS and invented precisely to invent. His filings spurned a significant question in the global intellectual property community: Does an AI inventor meet the various requirements for obtaining a patent? While important, the current debate among the global intellectual property community does not capture the entire challenge. Indeed, while salutary to answer the question of granting patents to an AI inventor, doing so will not solve the more pressing AI ownership challenge. Without considering the entire AI patent system and the humans who enable the commercialization of the AI patent, little benefit is derived for the public. And so, while there are many contrasting opinions on who should be the inventor and/or owner, there is little work on why ownership is a critical consideration.
The importance of the ownership question has applicability to the wider and quickly evolving AI industry worldwide. Indeed, generative AI, a new term to delineate the outputs of AI, is also subject to ownership issues and must be properly understood. While this paper focuses on AI patent ownership, it hopes to prove valuable to assess wider generative AI developments.
The Patent Bargain
The patent bargain is at the heart of the patent system, and any discussion that tackles inventorship and ownership in patents must depart here. The Supreme Court of Canada explained this bargain in Free World Trust v. Électro Santé Inc.: “In return for disclosure of the invention to the public, the inventor acquires for a limited time the exclusive right to exploit it.” The Court elaborated on the concept of the patent bargain in Apotex Inc. v. Wellcome Foundation Ltd., describing a patent as “a method by which inventive solutions to practical problems are coaxed into the public domain by the promise of a limited monopoly for a limited time. Disclosure is what the public gets for allowing owners exclusive rights.
Central to the patent bargain is for society to benefit from the inventor’s knowledge when the government grants a patent for an invention that is new, useful, and not obvious.[i] Through the work of patent examiners worldwide and the courts later testing their work – with much input from scholars and policymakers along the way – doctrines and principles are shaped over time to breathe life into this patent bargain. In one of his many foundational pieces, and long before it was trendy to write about intellectual property and artificial intelligence, David Vaver argued, “No two persons are likely to agree on [what the ideal law on patents for an invention should look like].” Yet, we can agree that the patent system aims overall to improve a nation’s economic performance and contribute to social welfare.
Vaver offers a guiding framework that patent laws should seek to reward the right activity, on the right conditions, to the right extent, with the right level of enforcement, and the right person. He suggests that patents should more precisely match and reward the advance the inventor discloses, not be granted for activities that need no stimulus or are already adequately stimulated by other intellectual property laws, disclose all the inventor knows about the invention to as wide an audience as possible, catch only activities the patentee and the public fairly expect to be included with the patent’s claims, and be enforced in ways that do not unfairly benefit patentees and unnecessarily restrain industry.[ii] Among these guidelines, proper disclosure by the inventor to the public is central.
The disclosure requirement incentivizes innovation: it prevents wasteful duplication of research, allows follow-on innovation by informing patentees about the boundaries of a patented invention, and benefits the public by disseminating new technology. In the context of AI, it may be difficult to describe how the algorithm works and to properly disclose the inner workings of the invention, putting the public in the same position as the machine. On the other hand, if AI-invented patent applications are unlikely to succeed (as is currently the case in various jurisdictions), there is a risk that inventors may choose not to disclose at all and to rely on trade secrets instead, effectively undermining the core rationale of the “patent bargain.” Indeed a trade secret mandates the very opposite of patents, “locking up” the invention and ensuring no one knows about how it was made.
In assessing the various policy implications and unintended consequences of ascribing AI inventorship and ownership, we go back to trying to sort out the desired patent system. It is typical to lean on theories underpinning IP and patent law. Among the strongest and most widely adopted justification for IP protection is the Economic theory or incentive theory. In a nutshell, encouraging individual effort by personal gain is the most effective way to advance public welfare and granting rights flowing from these acts as an incentive. Although AI cannot be motivated to invent by the prospect of a patent, we can incentivize developers to develop creative machines. And having more creative machines and their patented outputs is good for society.
Natural law or labour theory is often adopted in IP, especially in copyright law. Drawing from John Locke, a person is said to hold a natural right to the fruits of their labour. So, if you put work into something, you deserve to get something out of it. According to this theory, ownership of AI-generated inventions should be awarded to the person who laboured to plant the seed, AI being the tree that bears fruit in this metaphor.
Personality theory draws from Kant and Hegel to say that a person’s personality becomes fused with their inventions, thus warranting protection. Incorporating AI complicates the infusion of personality, but to what extent? Perhaps it could be argued that the developer’s personality becomes fused with an inventing machine.
The theories which underpin patent law – economic theory, natural law theory, personality theory – as well as the still useful but perhaps less popular theories – contract theory, Foucauldian, feminist, Marxist, and so on – should help guide determinations of who qualifies as inventor or owner. A human touch is necessary to ensure the patent bargain continues to function. We must also ensure we are rewarding the right humans. Perhaps that means those who can be incentivized to invent to advance the public good, those who put work into the invention, or those whose personalities are reflected in the AI or the final product. It is also essential that whichever human is granted patent protection have the willingness and ability to properly meet their disclosure obligation to fulfill their end of the bargain.
The AI Inventor v the AI Owner
The inventor is the person or persons who conceived of the invention and who is responsible for the inventive concept. It has been argued that AI should be excluded from being recognized as an inventor because modern AI processing is distinct from the human mental act of conception. Corporations have been barred from inventorship status for a similar reason: “people conceive, not companies.”[iii] It has been argued in support of awarding computers inventorship that AI may function independently, and it is only sometimes the case that substantial insight is needed to identify and understand a computation invention.
In 2019, a representative of the European Patent Office stated, “It is a global consensus that an inventor can only be a person who contributes to the invention’s conception in the form of devising an idea or a plan in the mind.” She added that, “The current state of technological development suggests that, for the foreseeable future, AI is… a tool used by a human inventor… Any change… [would] have implications reaching far beyond patent law, i.e., to authors’ rights under copyright laws, civil liability, and data protection.” The implications for patent law and beyond, however, are more closely tied to who is awarded ownership than who is awarded the title of inventor.
All computer work is appropriated. Computers cannot be incentivized. The effectiveness of an intellectual property regime in incentivizing innovation and creation rests solely on its ability to impact humans and, therefore, its ability to allow the right humans to reap the rewards of AI-generated inventions or works. Further, it is human owners who will play a role in ensuring the existing model of tort liability remains functional in the context of AI-generated inventions.[iv]
Just as important as determining who is an inventor (or author, in the copyright context) – whether a human or a machine – who is the owner, and how and why, is vital. The AI patent or copyright owner will be instrumental in giving life to the right. The owner defines where, how, and when commercialization occurs. The owner enjoys the rights and incentives associated with patent protection. Ultimately the owner or owners and many other commercialization stakeholders (i.e., investors) are material to the innovation ecosystem and part of the realization of the patent bargain’s public benefit.
AI’s Challenges to the Patent System
The ownership considerations of the questions raised by AI inventors should not be overlooked as we work to bolster our AI innovation ecosystem for the global public good. Will the owners of the most powerful AI or those with the means to purchase the most powerful computers become patent monopolists? How can we ensure the right individuals are rewarded and have access to file a patent application in the first place? How can we not only incentivize AI innovation but also ensure this innovation takes place in a manner that fulfills the patent bargain? Which model of ascribing patent ownership concerning AI-generated inventions is put in place is closely tied to these questions.
The challenges to the patent system are complex and cannot be attributed solely to its constituent laws but also to its institutions. The entire innovation system needs to work from the pre-patent inventive stage to the post-patent commercialization stage. For instance, the speed of technological developments can render patent examiners’ knowledge dated. This is exacerbated in the AI context, where a reasonable onlooker or patent examiner may find it difficult to explain the inner workings of AI.
The peer-to-patent system piloted in the UK, Japan, the US, and Australia is one mechanism of for infusing fresh and accurate expertise into the examination process. More drastic overhauls, for example re-examining the “person having ordinary skill in the art” test, have also been suggested considering the potential high bar to obviousness that may be established as AI advances. Since data is the precondition for AI innovation, there have also been calls to ensure antitrust law and governance principles of open access and data sharing work to prevent consolidation in the data and AI industries.
When assessing the granting of patents to AI-generated inventions, it is important to consider these challenges and the wider socio-economic and cultural framework and how it can best enable the proper application of patent laws. If the computer is designated as the first owner and grants entitlements to individuals through contract, all downstream inventions would require economically inefficient assigning or licensing. If humans are listed as owners of AI-generated inventions, scholars have asked who that human should be – is it the creator of the AI, the person who asked the curious question or happened to press a certain button that set off a chain reaction leading to the invention, or the person who discovered that the AI had generated the invention? The Legal Board of Appeal of the European Patent Office recently remarked they were not aware of any case law which would prevent the user or the owner of a device involved in an inventive activity from designating himself as an inventor under European patent law. Socio-economic and cultural considerations should factor into a determination of who qualifies as user or owner and who between has a superior claim.
Generative AI and Ownership
IP ownership questions are also material in global generative AI developments. Rather than an invention subject to a patent, these generative outputs can be subject to copyright protection. Indeed, Open AI’s ChatGPT, which was downloaded more than 500,000 times in its first month and accumulated more than 30 million users in two months, can generate creative works, including poems, screenplays, stories, term papers, reports, business plans and so on.
Kristina Kashtanova, the author of a comic book containing some images generated by AI, was denied copyright protection by the United States Copyright Office over the specific images generated by the AI as she was “not the author for copyright purposes.” In another instance of an AI-Generated Work, Dr. Thaler filed to register for copyright protection and is now in the process of challenging a refusal.[v] While the AI was named as the author, Dr. Thaler sought ownership of the copyright. Dr. Thaler rooted his entitlement to the work in common law principles of property ownership, including accession and first possession,[vi] as well as in the work-for-hire doctrine.[vii] The outcome of Dr. Thaler’s challenge will have wide-reaching implications, especially as AI-generating machines grow more popular and more advanced. Who owns the generative AI matters for determining commercialization questions and developments of various industries.
Concluding Remarks – The Hack
When attempting to harness the power of generative AI, it is key to develop skillful prompts. The appropriateness of a command impacts the quality and accuracy of the generated output. Similarly, policy questions of great importance, such as how or how not to account for AI-generated inventions and works in existing intellectual property regimes, can be better answered when the question is more pointedly tailored. This paper serves to better prompt the discussion surrounding the AI Inventorship Challenge: it is not so much who invents or creates but who owns the patent or copyright that has the power to ensure the public benefits and the patent or copyright bargain is realized.
Giuseppina (Pina) D’Agostino is an Associate Professor at Osgoode Hall Law School at York University. She is the Founder and Director of the IP Innovation Clinic, the inaugural Vice-Director of Canada First Research Excellence Fund program, Connected Minds: Neural and Machine Systems for a Healthy, Just, Society and the inaugural Co-Director of Centre for AI & Society at York University in Toronto Canada. My thanks to JD student Claire Wortsman for her speedy research assistance.
Photo by Pavel Danilyuk
[i] The seminal test for granting a patent; see the Patent Act.
[iii] New Idea Farm. Equip. Corp. v. Sperry Corp., 916 F.2d 1561, 1566 n.4 (Fed. Cir. 1990).
[iv] Applying the concepts of strict liability and negligence can be challenging in the case of autonomous AI systems: see Akanksha Bisoyi, “Ownership, Liability, Patentability, and Creativity Issues in Artificial Intelligence,” (2022) Information Security Journal 31(4).
[v] Thaler v. Perlmutter et al, “Complaint,” 1:22-cv-01564.
[vi] Ibid, paras 45-51.
[vii] Ibid, paras 52-56.