Finding a Way Forward: Analyzing Approaches to Artificial Intelligence Inventorship

By Zong Wei (Wade) Lin
June 20, 2020

“This article proposes that, to incentivize innovation, embracing the patent eligibility of artificial intelligence-inventions is key. Thus, we suggest viewing AI as a tool—or a pet—to indicate autonomy to some extent.”

artificial intelligence worldEarlier this year, the World Intellectual Property Organization (WIPO) received hundreds of submissions commenting on the Draft Issues Paper on Intellectual Property Policy and Artificial Intelligence. Contemporaneously, the notable DABUS cases were rejected by the EPO, UKIPO, and USPTO on the ground that AI cannot be named as inventor. The uncertainty in the ownership/inventorship of AI technology could impede investment and development of AI technology. This article aims to look into the WIPO submissions and arguments for addressing AI inventorship. Considering balancing the incentive of fostering AI technology and genuine inventorship, this article suggests seeing AI as a tool, or a pet, and that requiring the applicant to disclose any AI technology involved is the better resting place.

Consensus, Complexity and Considerations

Over one hundred of the WIPO submissions commented on the Inventorship and Ownership topic. Among them, we noticed some common grounds. Interestingly, those preliminary consensuses are inevitably contingent with the most contentious and intricate issues. By walking through those consensuses and issues, an overall picture can be established before providing recommendations for addressing AI inventorship.

Are we ready to see artificial intelligence as a human being?

When the law cements that an inventor has to be a natural person, it becomes difficult for patent offices to accept AI as an inventor since that would be tantamount to recognizing AI as a human being. See submission by European Union. For instance, does that mean AI has a personality? Is AI entitled to file opposition or invalidation suits? See submission by Enterprise Ireland.  Would AI infringe others’ patents and bear liability? See submission by AIPLA. The ramifications could be troublesome.

In fact, even the applicant in the DABUS case agreed AI is a machine. That said, although AI is becoming more and more akin to humankind, no, we are not yet in the reality depicted in films like Terminator; I, Robot; Bicentennial Man, or Doraemon. See Elizabeth Wasserman quoting Erik Brynjolfsson, director of the MIT Initiative on the Digital Economy. The situation to be solved is, however, that AI in some ways involves the formation of the conception, see Burroughs Wellcome Co. v. Barr Labs., Inc., 40 F.3d 1223, 1227–28 (Fed. Cir. 1994)., and it seems unfair to name a person as inventor simply because they operated the algorithm.


How Should AI-Invention be Defined?

Defining AI-inventions is the first priority. See submission by United States Copyright Office, opining that “a determination of the line between human and machine authorship would be a ‘crucial question.’” However, defining AI-inventions is not an easy task, especially since the autonomy of AI can be widely varied. Depending on the extent of the autonomy, AI inventions can be generally categorized as AI-assisted inventions and AI-generated inventions. See Michael McLaughlin, Computer-Generated Inventions, 101 J. PAT. & TRADEMARK OFF. SOC’Y 224, 235 (2019). AI-assisted inventions refer to those in which AI is used merely as a tool to hasten the research and development of the invention; AI-generated inventions, on the other hand, mean AI contributes to the conception of the invention no matter whether human intervention also takes part.

Distinguishing AI -inventions is apparently evidence-oriented. See submission by Canada. The evaluation of AI’s involvement will be highly dependent on the invention’s disclosure. See submission by Germany. Frankly, although most submissions agreed on defining AI-inventions as the first priority, it is difficult to delineate a cutting-edge technology like AI, let alone defining inventions made by it. Thus, hurdles and bumps can be expected even before entering the core discussion of AI inventorship.

Incentive and Discrimination

The solution to AI inventorship should hinge upon the goal of incentivizing innovation. See submission by International Trademark Association. If listing AI as an inventor can better stimulate technology development , it seems reasonable to consider that option. See submission by Korean Intellectual Property Office. Nevertheless, it is also evident that the patent system’s reward should be balanced with the efforts invested in development. Since AI seemingly makes innovation much easier and less costly than common “human” inventions, shortening the patent term for AI inventions is one option on the table. Id. However, some suggest AI-generated inventions should not be eligible for patent protection at all. See submission by Computer & Communications Industry Association. Those submissions focus on the argument that patent incentives must be balanced with public interest.

Inventorship of AI-Inventions: Sketching the Contour

With the discussions above in mind, we now turn to possible solutions.

  1. AI as Inventor

The principle that an inventor has to be a natural person is more or less entrenched, but the idea that a computer can “think” autonomously has never been part of the discussion. The DABUS team advocates naming AI as the inventor and owner thereof as the default assignee. See generally, Ryan Abbott, I Think, Therefore I Invent: Creative Computers and the Future of Patent Law, 57 B.C. L. REV. 1079 (2016). This argument focuses on correctness; while AI is unlikely to complain and does not need recognition, it would be unfair to name a person as an inventor if they did not contribute much to the invention process. Id, at 1103. That said, the person operating the AI (“AI operator”) should not be entitled to be named inventor merely because they fed information or asked AI to solve a problem. Id, at 1094.

However, the DABUS team agreed that AI is a machine owned by its owner. Id, at 1114. As the UKIPO pointed out, that inevitably poses some logistical conflicts—since AI is an object, how could it transfer the right to its owner? See UKIPO Decision, at ¶ 21. The “work-for-hire” doctrine doesn’t fit this scenario because a machine cannot be employed. If the owner obtains the right as a successor, can a computer sign an assignment? What about the oath or declaration requested by the Patent Act? And since AI presumably is tremendously knowledgeable, how does AI fulfill the duty to disclose? The DABUS application accurately highlights the possible unfairness rooted in credibility issues with naming an AI owner as inventor, but its implementation would raise many additional issues, thus, solving one problem but generating more.

  1. Excluding AI-Generated Inventions from the Patent System.

This proposal calls for “abolishing patent protection for AI inventions”. See Shlomit Yanisky Ravid & Xiaoqiong (Jackie) Liu, When Artificial Intelligence Systems Produce Inventions: An Alternative Model for Patent Law at the 3a Era, 39 CARDOZO L. REV. 2215, 2222 (2018). The patent system aims to provide “not merely an incentive but the right incentive” See Michael McLaughlin, supra note, at 244. The proposal considers the cost and hurdle for innovation will significantly drop in the wake of AI technology, See Shlomit Yanisky Ravid, supra note, at 2240, questioning whether AI-generated inventions are still worth the reward.  Rewarding AI owners exclusive rights seems to give them “unproportioned reward for their labor.” See Michael McLaughlin, supra note, at 244.

It is extremely important for this model to distinguish the level of human intervention—when the human intervention is only minimal, the invention is likely not patent eligible. Id, at 240. Some suggest evaluating the sufficiency of human intervention through the following consideration: whether the computer is designed—as a tool—for a particular problem; or whether it is used to create an invention with minimal human intervention. Id, at 247.

However, under 35 U.S.C. § 103, “[p]atentability shall not be negated by the manner in which the invention was made.” This article thus questions whether considering the effort invested in development is relevant. There is also concern about whether, under this scheme, the incentive is still sufficient to drive innovation. The proposal suggests that patents are not the only option and suggests tools such as trade secrets. Id, at 248. However, some inventions by their nature are not suitable for trade secret protection, and that will not change whether it is created by a human or computer.

The uncertainty in ascertaining human intervention is also worrisome. Uncertainty is a bad neighbor of business. Stakeholders need to know how to handle this problem effectively and to what extent the threshold must be met to preserve the ticket to patent rights. Thus, this option does not seem feasible.

  1. Truffle Hogs Compromise and Put the Dispute on Hold

This article proposes that, to incentivize innovation, embracing the patent eligibility of AI-inventions is key. Thus, we suggest viewing AI as a tool, or a pet—to indicate autonomy to some extent. See Elizabeth Wasserman quoting Oren Etzioni, head of the Allen Institute for Artificial Intelligence in Seattle. Pets, though they have intelligence and a certain level of autonomy, do not have legal personhood. The owner should own the truffles even if they have no idea how their truffle hogs located them. Likewise, the AI operator earns the title of inventor because they legitimately control, confine, and possess the AI during conception. See Barrow v. Holland, 125 So. 2d 749, 751 (Fla. 1960). The ownership of the AI-invention can be held by the AI operator, their employer (work-of-hire), or successor. In this way, the origin of the invention and true entity entitled to the exclusive right can be clearer and more easily identified.

Naming an AI operator as the inventor can efficiently fit this cutting-edge technology into the current patent system without upending it, and in fact is reasonable in terms of analogizing with some existing technology. For instance, antibodies used to be generated by injecting antigens into experimental animals and collecting suitable B cells that secrete the desired antibodies for producing hybridoma. Researchers in fact cannot control how the antibodies are generated—the animals’ immune system should take the credit. Does that make them less qualified as inventors?

This article is also mindful of the concern of wrongful credibility, but considers that as a question of disclosure instead of inventorship. The key is to list the AI operator and show truthfully if any AI was involved. Analogizing inventions involving bacterial materials, patent law requests “reference to a deposit of such biological material.” Thus, imposing an obligation to disclose the AI used in developing the invention could be a tiebreaker that can properly show the contribution of the listed human inventor. The public can conclude the true nature of the invention from the full disclosure of the invention; it doesn’t have to be from the inventors listed. In this way, the thorny questions of whether AI is a human and how to evaluate the level of human intervention can be temporarily set aside.

This model is not flawless, because we surely will encounter inventions in which the human intervention is so minimal that is seems unacceptable. Therefore, measures to properly restrain the exclusive right of AI-inventions are open for discussion. For instance, a framework similar to FRAND in the realm of standard essential patents (SEPs) can be introduced to curb the monopoly and ensure fluidity and accessibility of the technology. Beyond that, we might need to compromise and put the dispute on hold until a day the world is ready to see AI as a human.

A Critical Intersection

While the world of The Terminator is not yet approaching, we are close to the critical intersection where these questions must be decided. AI cannot be listed as an inventor without being caught up in the question of whether AI is a human. On the other hand, abolishing AI-inventions from patent protection is undesirable because it sacrifices the major incentives driving innovation. This article suggests that, instead of being stuck on this question of AI inventorship, requiring the disclosure of AI involvement is a better way to notify the public and manage AI-inventions under the current patent system.


Image Source: Deposit Photos
Copyright: Syda_Productions
Image ID: 179468030 


The Author

Zong Wei (Wade) Lin

Zong Wei (Wade) Lin is a Taiwan patent attorney and a qualified China patent agent. He is currently an LLM candidate in intellectual property law at the University of Washington.

Warning & Disclaimer: The pages, articles and comments on do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author and should not be attributed to the author’s employer, clients or the sponsors of Read more.

Discuss this

There are currently 11 Comments comments. Join the discussion.

  1. Joe Williams June 20, 2020 9:51 pm

    This is a thoughtful post; while if it has a reward mechanism, then, get back to alice.

  2. MaxDrei June 21, 2020 5:44 am

    An enjoyable read. Ownership of the AI-made invention seems to be easier to agree upon than what to declare to the PTO as the inventorship entity.

    I note the peremptory dismissal of the notion that we can now “employ” a machine to make an invention but isn’t the word “employ” entirely apt to describe what we humans do with machines?

    Since 1978, Section 39 of the UK Patents Act 1977 has decreed that an invention made by an employee is, ab initio, the property of the employer. No instrument of assignment needed, under the law. Why can’t I declare Dabus to be my employee, in the sense of S. 39 of the 1977 UK Act? Then a Dabus signature is of no interest to anybody.

    As we get used to AI making inventions, it’s going to be interesting to debate the attributes of the (already imaginary) PHOSITA. Not so much for novelty or enablement but for the debate on obviousness.

  3. Anon June 21, 2020 6:21 am

    As one who has written extensively on the thrust of this article, I am both heartened that someone is (finally) coalescing some of the important aspects, and rather disappointed with the ‘final’ suggestions of this article.

    By attempting to “compromise” between very opposed viewpoints, the muddle if the result is simply not acceptable — not to me, and likely not to the advocates of the portions of viewpoints presented here.

    This is not a roadmap.

    This is a wishlist.

    I do give credit to the identification and explication of several key issues, but the hard questions were not answered here, and there is NO answer to be had until those hard questions ARE answered.

  4. Anon June 21, 2020 8:40 am

    I note the peremptory dismissal of the notion that we can now “employ” a machine to make an invention but isn’t the word “employ” entirely apt to describe what we humans do with machines?

    Entirely different legal and street vernacular contexts MaxDrei.

    It is decidedly unhelpful to blur those lines.

  5. Anon June 21, 2020 8:45 am

    Since 1978, Section 39 of the UK Patents Act 1977 has decreed that an invention made by an employee is, ab initio, the property of the employer.

    Another item that shows your lack of respect for the US Sovereign.

    It is not a new point — to the board and particularly to you personally — that the US Sovereign has chosen differently with its “ab initio” requirement.

    Also, I have personally put to you that the Dabus case in your Sovereign — under your musings — does NOT resolve the issue, but only kicks the can down the road just a little. You cannot declare Debus to be an employee because you merely displace the critical question (employees – as humans – carry certain rights as humans).

    You are NOT moving the conversation forward by retreading your past points, MaxDrei.

  6. Anthony Claiborne June 21, 2020 2:08 pm

    The proposal is interesting. I note that a deposit of the AI “material” may pose major challenges. Many deep learning and other AI systems rely on highly distributed data processing for operation. In many cases, you can’t just package up the code. Still, this seems to be a nice compromise, before we reach the age where it becomes increasingly difficult to deny personhood to artificial intelligences.

  7. Stephen Potter June 22, 2020 5:13 am

    Iprova’s invention methodology uses AI methods to help its human invention creators and seems to be a general advance: see the recent joint press release with BIC who are the first licensees of the Iprova techniques –

  8. Stephen Potter June 22, 2020 5:21 am

    A personal comment from my side:
    – the concept of a patent is to make available to the public inventions that are novel, not obvious and capable of industrial application
    – none of the 3 patent offices to which the DABUS application was offered denied that the Claims were, in themselves, patentable and, in that sense, had value to the public
    – put crudely, the objections to them were essentially procedural and I can’t help feeling that, over the next 3 or 4 years ways will be found to be able to justify them…

  9. Anon June 22, 2020 12:20 pm

    Mr. Potter,

    Putting aside for the moment any differences in opinion that I may have as to “concept of patent” (perhaps in particular to the non-US view of ‘industrial application’), I can grasp the ‘procedural’ aspect that you would want to highlight, but would caution that — at least here in the US — the very foundation of the legal right is tied to the Lockean philosophy and the evolution of a real person’s inchoate right into a fully functionable legal property right.

    In this sense, only looking (as you do) as to the portion of the patent property right dealing with the subject of the matter properly being patentable is simply not enough to resolve the underlying issue.

    I do not agree that this issue is ONLY “essentially” procedural.

    Certainly, the issue does have its procedural elements, but at least for the US Sovereign patent system, the issue is much more than merely procedural.

  10. Evan Taaved Pert June 22, 2020 5:27 pm

    How could AI properly execute an Oath or Declaration as required by 35 USC 115?

    The AI would have to sign an oath that it “believes” something. “Believes”?

  11. Anon June 23, 2020 10:53 am

    AI’s inability to complete human tasks is only half the picture.

    The other half (as put forth in the Dabus applications and elsewhere) is that if no man may rightfully claim the conception portion of an innovation, then the innovation may fail to be able to obtain patent coverage (but NOT for lack of the innovation’s patent eligibility or other aspects of patentability).

Post a Comment

Respectfully add to the discussion.

Name *
Email *