This article is published under a Creative Commons license and not by the author of the article. So if you find any inaccuracies, you can correct them by updating the article.


When the Invented Becomes the Inventor: Can, and Should AI Systems be Granted Inventorship Status for Patent Applications? Creative Commons

Link for citation this article

Lindsey Whitlow

Legal Issues in the Digital Age, Journal Year: 2020, Volume and Issue: №2, P. 2 - 23, https://doi.org/10.17323/2713-2749.2020.2.3.23

Published: Nov. 4, 2020

Latest article update: Aug. 28, 2023

This article is published under the license

Link for citation this article Related Articles


Artificial Intelligence (“AI”) systems have become vastly more sophisticated since the term was first used in the 1950s. Through the advent of machine learning and artificial neural networks, computers utilizing AI technology have become so advanced that a team of attorneys in the United Kingdom claim that their AI machine, DABUS, actually created patentable inventions. The team went so far as to file patent applications with the European Patent Office, the UK Intellectual Property Office, and the US Patent and Trademark Office. All applications named DABUS as the inventor. This sparked a heated debate within academic and legal communities that centered around whether AI can be an inventor, and, if so, what this might mean for the current state of patent law. This paper discusses the purposes of patent law through a brief look at its history, in an effort to highlight why patent law as it stands may no longer be one- size-fits-all. It considers the evolution of AI systems to explain how one might determine that a machine could be “creative” and therefore justifiably named as inventor. It surveys popular opinions and organizes them on a spectrum ranging from those who believe that patent law should stay as it is and that AI cannot be an inventor, to those who, more dramatically, advocate for the abolition of patent protection for AI inventions. This paper suggests that legislators be proactive in traversing this technological minefield rather than reactive, as technology will continue to outpace, and trample, law if left to its own machinations.


USPTO, Patent Law, Artificial Inventor, Emerging Technologies, Inventor, Patent, Trademark, Creativity Machine, Copyright, Artificial Intelligence


In early August 2019, the Artificial Inventor Project, led by an international team of attorneys, filed patent applications for two inventions in both the US and the EU: one for a “fractal container” — the purpose of which is to “improve grip heat transfer in and out of the container” and enable one container to be connected to another — and the other a “neural flame” — a lighted device to be used for search-and-rescue missions1. These inventions seem innocuous enough, and likely to meet the novelty, utility, and non-obviousness requirements for patentability2. So what in these applications could possibly spark controversy among legal academics and practitioners, and have the business and tech worlds holding their breath? One small, seemingly insignificant detail: in the area where an applicant must list the inventor of the product or process for which the patent is sought, these attorneys listed DABUS3, a “Creativity Machine,” rather than a human inventor4.

The patent system5 is built to deal with “inventors” as human beings. Introducing the possibility that a non-human can “create” or author something has either not been contemplated, or sometimes completely refuted, by US law [Кор M., 2019]6. Now, however, the United States Patent and Trademark Office (USPTO) recognizes the need to address this question. On August 27, 2019, the USPTO promulgated a request for comment in the Federal Register for several questions relating to Artificial Intelligence (Al) and the patent system7. Shortly thereafter, what began as a quest to determine how patent law must react to new uses of technology was then expanded to include all of intellectual property law, and the comment period was extended.

There are arguments for and against awarding Al systems the title of “inventor”. Within these arguments and opinions lie core policy questions that the USPTO must address, namely whether inventors can only be human. This commentary endeavors to provide a broad overview of some of the more common positions held by industry leaders and academics on the state of patent law in response to these filed applications. Many such policy opinions and positions rely heavily on the history and purpose of the patent system; the same facts and history construed differently depending on said position.

Part I of this paper charts brief histories of both the patent system and Al, and how we arrived at the current policy discussion. What follows in Part II is a rudimentary, and barbarously simplified, description of how Al systems, and in particular DABUS, work, and how they might share characteristics with a human “inventor”. Part III then introduces industry opinions about the idea of inventorship. Part IV touches on how similar questions have been asked and answered before by describing how different countries have reacted to the use of Al in copyrighted, or copyrightable, works. This paper then concludes by suggesting that, rather than following the legal precedent of being reactionary, legislators should take strides now to push these important legal developments forward.

1. Histories Repeat

1.1. Patent Pending — a History

Patents, or the precursor of patents as we know them today, were first introduced in the 1400s [Chirambo C., 2019] and were initially distributed on an ad hoc basis8. The advent of the printing press in the mid-fifteenth century naturally spurred the need for protection of literary works, as dissemination became simpler — however, this event simultaneously created an open market for the imitation of inventions as published news became prominent and new inventions were shared on a wider, and more rapid, scale [Bugbee B., 1967: 17]. In response, inventors became sawier: Filippo Brunelleschi, Florentine architect and engineer,

refuse [d] to make [his] machine available to the public, in order that the fruit of his genius and skill may not be reaped by another without his will and consent; ...if he enjoyed some prerogative concerning this, he would open up what he [wa]s hiding, and disclose it to all [Bugbee B., 1967:17-18].

This was the first “patent” that recognized an inventor’s inherent right to his invention, “and the contractual nature, or quid pro quo, of patent protection”. The preamble to this grant alone was an acknowledgment by the Florentine government of the benefit to the city of Florence, and society, of incentivizing — and thereby stimulating — creativity by its population, and in providing legal protection for the same.

Subsequent to the heavy decline in the granting of monopolies in Florence [Bugbee B., 19-20], Venice was the next to take up the patent mantle. Importantly, the Venetian Senate then passed what is the first-known general patent law, which embodied the basic tenets of Brunelleschi’s patent [Bugbee B., 22]. It read:

We have among us men of great genius, apt to invent and discover ingenious devices ... Now, if provision were made for the works and devices discovered by such persons, so that others, who may see them could not build them and take the inventor’s honor away, more men would then apply their genius, would discover, and would build devices of great utility and benefit to our commonwealth.

This law mirrors the underlying policy introduced by the Florentine government: we are inventors and discoverers, but if we are to put in effort to invent and discover, this work should be protected from theft, and inventors should be rewarded for innovation.

The Venetian system spilled over to other parts of the world throughout the following centuries [Bugbee B., 25-27, 35-43]9. It even extended to colonial America, by way of England and its the Statute of Monopolies [Bugbee B., 36-38]; [Hovenkamp H., 2016: 263, 270]". An ad hoc monopoly-granting process continued in the United States until finally, upon the writing of the U.S. Constitution, there was a federal mandate of intellectual property protection directed to Congress: “Congress shall have Power ... [t]o promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries...”10.

What we see throughout the history of inventive development is a consistent belief that human invention requires incentive, and that the way to incentivize is to offer exclusive rights in the manufacture and production of an entity’s innovation. However, while the Patent Act frames inventorship and patentability in reference to human creation, there is no threshold of human control or input for the inventive process written into the law11. It seems fair to say that these same incentives to create, invent, or discover cannot exist for Al systems. But perhaps it goes further than that: with the need to incentivize human inventors of Al to push their ideas forward. To see how this detached layer of incentivizing human inventors affects the progress of Al development, we will need to similarly follow the advent and history of artificial intelligence.

1.2. They Walk Among Us — From Science Fiction to Science

Al is a complex field, and defining it proves difficult for most scholars and practitioners [Lea G., 2015]. Since Al can comprise innumerable technologies, and combinations of technologies, it is difficult to condense the concept into one singular definition [Marr B., 2018; Pring-Mill D., 2018]. For purposes of this discussion, it will helpful to start with the first recognized uses of the phrase “artificial intelligence”.

Though the idea was bandied about for some time, the concept of “thinking machines” was formally introduced by Alan Turing in his paper Computing Machinery and Intelligence, which asked: “Can machines think?” [Turing A., 1950: 433]. The now-famous “Turing Test” was created as a way to address this quandary — if a machine can answer questions, fooling a human judge into thinking that it is human, then the program is said to be exhibiting intelligence [Hern A., 2014]12. The first known use of the phrase “artificial intelligence” came in 1956, when John McCarthy called to order the Dartmouth Summer Research Project on Artificial Intelligence, where researchers from disciplines as varied as language simulation and complexity theory were to gather and begin a discourse about what we now call the field of AL

Following early successes with computer-based, complex problem solving, like IBM’s Deep Thought chess-playing machine in the 1980s [Higgins C., 2017]. Al researchers began with an assumption: if an Al could solve a more advanced problem, then it should also be able to solve much simpler ones. This turned out not to be the case [Higgins C., 21]13. The realization that the human skills that are the most difficult to reverse engineer are the unconscious ones [Rotenberg H., 2013: 108-109] contributed to a marked cooling off period in Al research. Researchers shifted focus from attempting to solve grand, societal problems to resolving narrow challenges for which there are clear measures of success [Dormehl L., 22].

One area to which this focus was applied was the advent of “expert systems”. These involved Al systems operating as problem-solving tools alongside human counterparts. Reasoning was not enough to solve real-world problems; it had to be combined with knowledge in order for it be useful. If, for instance, programmers wanted a computer that would be useful in the field of neuroscience, the system would need to be deeply familiar with every facet of the field in the same way that a studied neuroscientist would be. Programmers and developers “suddenly had to become ‘knowledge engineers,’ capable of taking human experts in a variety of fields and distilling their knowledge into rules a computer could follow. The resulting programs were called ‘expert systems’” [Dormehl L., 23-24].

Creating these systems seemed like a stroke of genius — until the process began hitting roadblocks. Counterintuitively, these expert systems would become less accurate as more rules were introduced: the more “knowledge rules” that were incorporated into the system, the more undesirable interactions between those rules would crop up. Essentially the rules, when interdependent, would cause the system to break if a contradictory rule was programmed. The process was complex, and cost prohibitive [Dormel L., 27].

But things changed again in 1996, when two students at Stanford University built a “smart web catalogue” [Dormehl L., 28] — a promising algorithm that became the boost Al needed in the court of public opinion, and which contributed to the rapid growth of the field. Within a decade, Geoff Hinton introduced the concept of “deep learning [Dormehl L., 49]”. Scientists moved away from programmable knowledge as seen in expert systems, and attempted to model the neural pathways of the human brain inside a computer by creating artificial neural networks [Dormehl L.,31]. In other words, these artificial neural networks are the computational approximation of the human brain [Hawkins J., 2004: 18-21]14. Though neural networks had been the rejected sibling to traditional Al research, suddenly, this became the only way forward [Dormehl L., 29, 49].

1.3. Crossing the Streams

Here is where we arrive at the crux of our discussion: the point at which patent principles and Al overlap and/or intersect. As quoted above, the provision in the Constitution that gives Congress the power to promote science and the useful arts allows for securing to authors and inventors the exclusive right to their works15. It has long been understood that “authors” and “inventors” refer to humans. In early August 2019, a team from the UK University of Surrey decided to test the boundaries of the patent system: the team filed two patent applications in both the USPTO and the European Patent Office (EPO) claiming an Al system called DABUS as the inventor16. The filing tabled the question of how would patent offices react to the claim that the invented has become the inventor.

In response, on August 27, 2019, the USPTO sent out a call requesting comments for patenting Al inventions17. The request states that the USPTO “is interested in gathering information on patent-related issues regarding artificial intelligence inventions for purposes of evaluating whether further examination guidance is needed to promote the reliability and predictability of patenting artificial intelligence inventions18”. The request asked questions like: “What are the different ways that a natural person can contribute to conception of an Al invention and be eligible to be a named inventor?” and “Do current patent laws and regulations regarding inventorship need to be revised to take into account inventions where an entity or entities other than a natural person contributed to the conception of an invention?”19 Within two months, a second call was made that expanded the request to also include copyright and trademark20.

2.How Might Al be Considered an Inventor?

While the USPTO asked several questions surrounding Al inventions in its request for comment, the application filing that sparked debate among legal practitioners and business entities centers around one particular question: whether an Al system can be considered an inventor for patent-filing purposes. The team that filed the application supports Al inventorship21. A cursory scan of some of the comments the USPTO received evidences that industry leaders oppose it22. The trouble with a blanket statement that Al can or cannot be granted inventorship is that there are several different types of Al, all with varying capabilities, all trained in different ways.

2.1. Machine Learning

Before Al can perform any great inventive or creative steps, it must be built, programmed, and trained. The first two steps of this process are wildly outside the scope of this commentary. However, it is relevant to discuss the training process for Al algorithms to better understand why one might make a policy argument in favor of an Al system being awarded inventor or author status. This training can be done through machine learning — “the ability for computers to learn without being programmed” [Wallace L., 2019]. Based on the documentation surrounding DABUS and the type of learning typically used in Creativity Machines, here we focus on two machine learning methods: supervised learning and unsupervised learning.

In supervised machine learning, the goal of training the algorithm is to uncover insights — i.e., recognize patterns or categorize information — from data, by telling the system the desired output23. This process typically has three phases: training, validation, and testing. In the training phase, the algorithm is given “inputs” — data from which it can draw its conclusion — and is told the desired output from this set of data [Ashley K., 2017]. Because the algorithm has the input, and correct output, it can “learn” how variables assigned by the trainer relate to the target output; this helps the system to recognize patterns and make predictions based on the given input24. The goal is for the machine to be able to make these categorizations correctly with every piece of data, based on its prior “learning”. Once a machine or algorithm has “learned” what it must do, in theory the Al can then take over on its own [Nielsen M., 2019].

Unlike supervised machine learning, when conducting unsupervised machine learning, the algorithm is not given a predetermined set of outcomes25. Because there is no desired output, the algorithm cannot classify the data inputs; rather, the goal of this type of training is instead to learn more about the data itself [Brownlee J., 2016]. The important difference between this type of machine learning and supervised learning is that, within the zetabytes of information available on the internet, there are an impossible number of unlabeled or incorrectly labeled data sets. In unsupervised learning, no labels are provided to the machine; it simply has input with no explanation [Dormehl L., 50]. This allows the machine to sift through the data for patterns that a human, given the same magnitude of data, could not possibly see.

Both of these types of machine learning lead to machines being able to make predictions based on vast amounts of input. As machine learning continually evolves and increases its prediction accuracy, machines become better able to perform tasks that would previously require human input [Agraval A.et al, 2018].

2.2. DABUS, or the “Device for the Autonomous Bootstrapping of Unified Sentience”

Law is reactive, and often slow to be so. As discussed, the idea of an Al invention then inventing things itself is not new, and neither is the discussion about how patent law will have to deal with it when it happens. While inventions have been deemed “creative” before August 2019, intellectual property law — at least in the United States26 — has not changed with the growth of Al technology. The team that initiated the Artificial Inventor Project27 seems to have done so to force the hands of patent offices to confront the question of whether Al can be considered an inventor.

According to the attorneys who filed the applications, DABUS is a “Creativity Machine” — “a particular type of connectionist artificial intelligence”28. Connectionist Al systems operate through the use of artificial neural networks, which are modeled after the way that neurons and synapses are thought to fire in the human brain, and are used to mimic the way humans learn [Dormehl L., 2019]. There are two neural network layers at play in Creativity Machines: the first network, and what this paper will call the “novelty network,” is trained using general information from a variety of fields of knowledge, and is made up of a series of smaller neural networks29. It is tasked with generating novel ideas in response to disruptions, or “selfperturbations,” in the way the smaller neural networks weigh and interpret statistical data from new inputs30. A second, overarching network monitors the novelty network for any ideas that are “sufficiently novel compared to the machine’s pre-existing knowledge base”31. It responds by increasing or decreasing the perturbations to which the smaller neural networks react when interpreting data, in order to “form and ripen ideas having the most novelty, utility, or value”32. Through this “learning” process, one may start to see how by analogy, an Al system may be considered inventive.

The Artificial Inventor Project maintains that DABUS was trained only on general knowledge “in the field”; as its training was likely unsupervised, there would be no expected outcome from its machinations33. Its goal was simply to create something novel. From this training, DABUS came up with these inventions independently, and was able — on its own — to designate them as novel34.

In some cases of invention by an Al system, a human may still be considered the inventor, perhaps because she exhibited inventiveness in creating a program to solve a specific problem, or carefully curated the information provided to the machine, or even identified the machine’s output as novel and inventive35. The Artificial Inventor Project team argues, however, that no human may be considered the inventor of these inventions36. DABUS was not created to solve any particular problem, it was not given training data specifically relevant to its creations, and the machine itself identified that the inventions were novel within the scope of prior art of which it was “aware”37.

These are all arguments made to support the idea that DABUS itself can and should be the only available option to list as the inventor of these innovations: it exhibited the inventiveness required of a human creator, and no human can claim to have had a hand in its processes. But convincing a patent office of these things is no small task. When so much needs to be proved just to show that an Al system is even capable of creative invention without human intervention, it seems like a monumental, and fruitless, effort for the Artificial Inventor Project team to push the need to list DABUS as the inventor of these products. However, the Manual of Patent Examining Procedure (“the Manual”)38, used by all patent examiners when determining whether or not to approve a patent application, gives some insight as to why the team is driving these test cases forward.

First, patent law requires the naming of an actual inventor, or joint inventors, and the applied-for subject matter39. The Artificial Inventor Project asserts that DABUS is the inventor. DABUS’s creator, Stephen Thaler, would not be allowed to list himself as inventor because he had no part in the conception of the products for which the group is seeking patent protection40. Patent seekers may not simply leave this area of the application blank, because, secondly, if an inventor is not listed, or is listed incorrectly and is not subsequently corrected, the Manual requires that “Office personnel should reject the claims under 35 U.S.C. 101 and 35 U.S.C. 115”41. If Thaler were to claim to be the inventor, but a patent office deems that not to be the case, any applications or claims may be rejected on that fact alone, leaving aside the discussion of whether Al can be a “creator”. Because the patent system requires the disclosure of how an object is made or a process is completed in order to be patent-eligible42, once the information is on record, it becomes public knowledge. The risk, particularly for businesses, is that naming the wrong inventor would result in no one receiving a patent [Crouch D., 2018], but with a chance that now the process is public information and the invention can no longer be monetized on the same scale.

Through this official interpretation of the patent-granting system, it makes sense that there are diverging opinions on how the USPTO should deal with Al as inventors.

3. The Al-Space Continuum

How industry leaders and legal practitioners have reacted to these claims seems to fall on a varied spectrum, though many of the ideas and opinions have areas of overlap. While not all responses to the USPTO’s request for comment have been published as of this writing, several industry leaders and associations have weighed in, sharing their views of how the patent system should react to patents filed with an Al inventor moving forward.

3.1. Legal Gymnastics and the “Law of the Horse”

On one end of the spectrum are proponents for leaving the law in its current state. The Institute of Electrical and Electronics Engineers (IEEE)44 and the American Intellectual Property Law Association (AIPLA) submitted comments to the USPTO for each of its inquiries45. When answering the question “Do current patent laws and regulations regarding inventorship need to be revised to take into account inventions where an entity or entities other than a natural person contributed to the conception of an invention?” AIPLA’s position is a clear “no.”46 The letter states that the law requires an inventor to be a natural person, and though DABUS is currently a case testing this principle, it is unclear whether this Al system “is truly ‘inventive AL’”47 Even if inventive Al does exist in the future, AIPLA says, that still does not dictate that Al should be granted automatically the title of inventor; rather, “it will be necessary to consider what types of activities by Al entities would be considered as inventive contributions to the claimed invention”48.

Similarly, the Intellectual Property Owners Association (IPO A) emphasizes that “inventors” must be natural persons, and notes that “if non-natural entities were afforded inventor status, additional downstream issues would also need to be addressed”49. For instance, a common concern is that a machine cannot fulfill the requirement for an inventor to sign an affidavit affirming that she is the original inventor, and that she understands that a false statement to that effect has legal consequences50. In addition, IPO A cites the fact that it would impossible to depose a machine to determine its inventorship status51.

Interestingly, however, when answering the question “Are there any other issues pertinent to patenting Al inventions that we should examine?” AIPLA recognizes that depending on the type of Al and the way it analyzes data and “learns” from it, the connection between human inventor and Al system may become more tenuous as Al systems become more “intelligent”52. We begin to see areas in which relying on the patent system in its current state may cause problems for future Al innovations.

3.2. Change is Inevitable, and the Sooner the Better

While industry associations seem to agree that changing the law to include Al in the definition of inventor is unnecessary and problematic, legal practitioners differ even amongst themselves as to the impact Al will have on the current patent regime in the United States. In an interview with legal practitioners, Law. com asked Kathi Vidal of Winston & Strawn and John Dragseth of Fish & Richardson, whether current laws and regulations regarding inventorship should be revised in light of the DABUS patents [Graham S., 2019]. Where Dragseth is unconcerned because society is not at a place where “an entity or entities other than a natural person [can] contribute to the conception of an Al invention”, Vidal looks ahead: in short, “adding Al to the [already divided incentives for the current patent system in other industries] may stretch our one-size-fits-all’ patent system to its breaking point”. She continued by saying that “very rule needs to be rethought when it comes to Al and all the data the Al analyzed to come up with the invention...if we are going to reward Al inventions, we need to make sure the public receives the appropriate quid pro quo”.

One way in which scholars propose to reform the patent system is to increase the patentability threshold as it pertains to Al-created inventions. They suggest a recalibration of the definitions for an inventive step, prior art, and non-obviousness, and standardizing, or at least balancing, guidelines regarding who or what may be considered a “person having ordinary skill in the art”. [Кор M., 16; Ramalho A., 2018]. For example, a person operating with an Al would likely embody a different definition of “ordinary skill” from someone doing the same research or production without the use of a machine. This would keep the current patent system in place generally, while changing the quintessential questions asked by examiners with regards to Al inventions.

Change to the patent system, it seems in this view, is necessary, whether or not that means a complete overhaul of the legal regime. There are questions and answers that cannot be addressed or even contemplated by the current state of the law. Technology evolves too quickly for whatever adaptations are made in how the current legal system approaches the awarding of patents. Proponents of these changes, however, do not necessarily advocate for excommunicating Al to another legal realm entirely. They merely push for stretching the metaphorical rubber band to capture that which is not currently suited for traditional notions of patent law protection.

3.3. And Now for Something Completely Different

A subset of the academic population believes that if these issues are not resolved satisfactorily prior to Al technology becoming fully capable of creative invention, this would then stifle the incentive for the investment in Al research and innovation. The proposed resolution for this group of scholars often involves a complete change to the patent system, or another regime for regulation entirely.

While it is widely agreed that an Al system would not need to be incentivized to invent in the same way the patent system rewards human inventors, the computer scientists, trainers, and machine learning experts behind these technological marvels may still require the incentive to even create the Al that can accomplish those tasks. Ryan Abbott, theoretical front man of the Artificial Inventor Project, asserts that lack of protection will lead to lack of innovation53. He states that if “outdated IP laws around the world don’t respond quickly to the rise of the inventive machine, the lack of incentive for Al developers could stand in the way of a new era of spectacular human endeavor”54.

Other scholars take this view a step further, asserting that traditional patent law is outdated — but even if it were to take into account recent technological development, it cannot be applied to Al-inventions [Yanitsky- Ravid S., Xiaoqiong L., 2018: 2215-2231]. These academics go so far as to advocate for abolishing patent protection for Al inventions. They assert that traditional view that a human person must be identified as inventor for AI- generated inventions is an unrealistic threshold: “Al systems can produce a surprisingly large number of inventions, write and submit numerous patent applications, and even evaluate (or monitor) the risk of patent claims”. Al systems can be programmed to fulfill necessary patent-eligibility requirements with regard to its inventions; if the law was not intended to cover human inventors alone, it is theoretically possible that Al systems could be entitled to patent rights in their creations.

The state of the technology may call for a completely new regime for protection, one outside of intellectual property law. Diametrically opposed to Judge Easterbrook’s discussion of the law of the horse, perhaps the introduction of legislation and regulation specific to Al should come under the purview of an entirely new agency, rather than existing within current frameworks that need be stretched and manipulated to account for the unique effect Al has on just about every aspect of modern life. The logistics of creating such a new regime have not been contemplated in modern literature, likely because each agency touched by Al reacts to developments seemingly within their purview as they appear. A new system would require the development of an agency that would hardly be able to build from models that already exist, and it is possible that agencies would not let go of this type of regulation to another agency quietly.

3.4. International Approaches and Recent Decisions

Other jurisdictions are facing these same questions. Most jurisdictions represented in the IP555 specifically restrict inventorship rights to natural persons56. Europe, however, does not follow this pattern. Article 60(1) of the European Patent Convention (EPC) simply states that “the inventor or his successor in title is entitled to the right to a European Patent”57. “The EPC does not define the term ‘inventor’ or construe inventorship rights as being limited to only natural persons. Thus, Inventive Al may be recognized by the EPO as an inventor”58.

Despite the wording, or lack thereof, surrounding inventorship in the European Union, the applications naming DABUS as inventor have been denied in both the EPO and the UK Intellectual Property Office (UKIPO; [Nurton, 2020]). The EPO has not fully explained its reasoning, but relying on Article 81 and Rule 19 of the EPC, stated that the applications “do not meet the requirement of the European Patent Convention (EPC) that an inventor designated in the application has to be a human being, not a machine”59. As stated above, this requirement is not clear in the wording of the EPC rules and articles applicable to inventorship; this interpretation of those rules may be the first of its kind.

The UKIPO, on the other hand, published a decision that called other facets of the patent application into question: it accepted that DABUS was the creator, but as a machine, it could not legally be considered the inventor [Nurton J., 2020: 112]. Further, the machine cannot have rights in the invention, and therefore could not possibly assign those rights to the designated “owner” of the invention listed in the application. The Hearing Officer also raised a question that will likely spark further debate in the future: “given that an Al machine cannot hold property rights, in what way can it be encouraged to disseminate information about an invention?”

4. Deja Vu All Over Again

As with many other questions raised by the USPTO, there is a substantial likelihood that, though these applications have been rejected, they have succeeded in their purpose: to begin a conversation that will likely be happening for years to come. Of course, this is not the first discussion regarding Al-created intellectual property, specifically within copyright law. The EU and the United States both hold that “there can be no copyright [in the case of purely Al Created works] because of inter alia the absence of a human author’s own intellectual creation as an extension of his personality”. This has been the assumption across all intellectual property forms in these jurisdictions.

Contrary to this, however, the United Kingdom implemented a computer-generated works (CGW) regime, which “stretches human authorship towards algorithmic authorship”. The rights given to CGW are slightly different than typical copyright protection; for instance, no moral rights are awarded, and the protection term is shortened to 50 years60. This seems odd, considering the UK’s decision to reject the Artificial Inventor Project’s application. In theory, the same type of concerns would be brought forth in both IP regimes: if a machine can be considered an author, there is still a question of whether the machine can assign its ownership rights over to the copyright registration applicant.

It is curious that this concern was considered more of an issue in the patent context. Speculating, one could say there are two reasons for this divergence. First, it is possible that the decision with regards to the DABUS patent applications is an effort by the UKIPO to roll back some of its policies on computer-generated works. Second, the office may have found something so uniquely present in the patent system and inventiveness requirements that the analogy to copyright rulings cannot apply.

If either reason, this shift toward allowing certain of CGW or AI-Gener- ated Works to be attributed to the Al itself in copyright law still has the potential to carry over to the patent side of intellectual property. Because the requirements for invention are more strenuous than the requirements for works of authorship, the speed of Al development can only tell how these offices will approach these questions moving forward.


The law is reactive. As it stands, there is heated debate as to whether Al is in a place to purely be creative or inventive — but regardless, the time seems to be coming. The Artificial Inventor Project’s goal in submitting applications naming an Al as inventor appears to be an attempt to begin the conversation surrounding these systems, and how the law can and should react to these innovations before they exist, knowing this technology is coming.

The dominant view is that, for now, the law should remain as it is. However, regarding the speed with which Al develops, legislators and regulators heeding this advice may be faced sooner than they think with Al that can do what DABUS is purported to have done, with more alacrity and even less human intervention. It may be time for legislation and regulation to attempt to work ahead of technological innovations to continue to provide incentives for the development of these Al systems.


  1. Agrawal A., Gans J., Goldfarb A. (2018) Prediction Machines: The Simple Economics of Artificial Intelligence. Cambridge (Mass.): Harvard Business Review Press, 272 p.

  2. Ashley K. (2017) Artificial Intelligence and Legal Analytics: New Tools for Law Practice in the Digital Age. Cambridge (UK): University Press, 446 p.

  3. Brief History of Patent Law of the United States (2014) Available at: https:// ladas.com/education-center/a-brief-history-of-the-patent-law-of-the-united- states-2/(accessed: 15.08.2020)

  4. Brownlee J. (2016) Supervised and Unsupervised Machine Learning Algorithms. Available at: https://machinelearningmastery.com/supervised-and- unsupervised-machine-learning-algorithms/(accessed: 15.08.2020)

  5. Bugbee B. (1967) Genesis of American Patent and Copyright Law. Washington: Public Affairs Press, 208 p.

  6. Chirambo C. (2019) Introduction to the Patent System: Challenges Facing Small Offices. Available at: https://www.wipo.int/edocs/mdocs/pct/en/wipo_pct_gbe_19/wipo_pct_gbe_19_topic_1b.pdf (accessed: 15.08.2020)

  7. Crouch D. (2018) When Co-lnventors Fail to Cooperate — Nobody Gets a Patent. Patently-О. Available at: https://patentlyo.com/patent/2018/05/inventors- cooperate-nobody.html (accessed: 15.08.2020)

  8. Dormehl L. (2017) Thinking Machines. New York: Tarcher Perigee, 288 p.

  9. Dormehl L. (2019) What is an artificial neural network? Here’s everything you need to know. Available at: https://www.digitaltrends.com/cool-tech/what-is-an-artificial-neural-network/ (accessed: 15.08.2020)

  10. Easterbrook F. (1996) Cyberspace and the Law of the Horse. University of Chicago Legal Forum, no 4, p. 207.

  11. Graham S. (2019) Can Als Hold Patents? Experts Answer USPTO’s Questions about Artificial Intelligence. Available at: https://www.law.com/therecord- er/2019/08/29/can-ais-hold-patents-experts-answer-usptos-questions- about-artificial-intelligence/(accessed: 15.08.2020)

  12. Hawkins J., Blakeslee S. (2004) On Intelligence: How a New Understanding of the Brain Will Lead to the Creation of Truly Intelligent Machines. New York: Holt, 272 p.

  13. Hern A. (2014) What is the Turing test? And are we all doomed now? Available at: https://www.theguardian.com/technology/2014/jun/09/what-is-the-alan- turing-test (accessed: 15.08.2020)

  14. Higgins C. (2017) A Brief History of Deep Blue, IBM’s Chess Computer. Available at: https://www.mentalfloss.com/article/503178/brief-history-deep-blue- ibms-chess-computer (accessed: 15.08.2020)

  15. Hovenkamp H. (2016) The Emergence of Classical American Patent Law. Arizona Law Review, vol. 58, p. 270.

  16. Кор M. (2019) Al & Intellectual Property: Towards an Articulated Public Domain. Unpublished manuscript. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3409715# (accessed: 15.08.2020)

  17. Lea G. (2015) The Struggle To Define What Artificial Intelligence Actually Means. Available at: https://www.popsci.com/why-we-need-legal-definition-artificial- intelligence (accessed: 12.08.2020)

  18. Marr B. (2018) The Key Definitions of Artificial Intelligence (Al) That Explain Its Importance. Available at: https://www.forbes.com/sites/bernard- marr/2018/02/14/the-key-definitions-of-artificial-intelligence-ai-that-explain- its-importance/#1a0646664f5d (accessed: 15.08.2020)

  19. Nielsen M. (2019) Neural Networks and Deep Learning. Determination Press. Available at: http://neuralnetworksanddeeplearning.com (accessed: 15.08.2020)

  20. Nurton J. (2020) EPO and UKIPO Refuse Al-Invented Patent Applications. Available at: https://www.ipwatchdog.com/2020/01/07/epo-ukipo-refuse-ai-in- vented-patent-applications/id=117648/ (accessed: 15.08.2020)

  21. Pring-Mill D. (2018) Everyone Is Talking About Al — But Do They Mean the Same Thing? Available at: https://singularityhub.com/2018/03/15/everyone-is-talk- ing-about-ai-but-do-they-mean-the-same-thing/(accessed: 15.08.2020).

  22. Ramalho A. (2018) Patentability of Al-generated Inventions: Is a Reform of the Patent System Needed? Institute of Intellectual Property Journal, no 2, pp. 25-26.

  23. Rotenberg V. (2013) Moravec’s Paradox: Consideration in the Context of Two Brain Hemisphere Functions. Activitas Nervosa Superior, vol. 55, pp. 108-109.

  24. Turing A. (1950) Computing Machinery and Intelligence. Mind, vol. 59, pp. 433- 460.

  25. Vertinsky L, Rice T. (2002) Thinking About Thinking Machines: Implications of Machine Inventors for Patent Law. Boston University Journal of Science and Technology Law, no 8, pp. 586-587.

  26. Wallace L. (2018) What You Need To Know About Al. Available at: https://www.forbes.com/sites/forbestechcouncil/2018/01/08/what-you-need-to-know-about-ai/#6a73019253c7 (accessed: 15.08.2020)

  27. Yanisky-Ravid 8., Liu X. (2018) When Artificial Intelligence Systems Produce Inventions: The ЗА Era and an Alternative Model for Patent Law. Cardozo Law Review, vol. 39, pp. 2215-2231.