By: Andrew J. Costa, Associate
In this month’s post I provide an update on the “Thaler Saga” that’s been testing the legal bounds of AI inventorship in the United States (and the world) since our last post on the subject in May 2022. I will also examine key AI questions for patentability left open in the wake of the recent Federal Circuit decision in Thaler. In August, the Federal Circuit rendered its decision in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022) (read the opinion here) affirmatively closing the door on whether an AI can be a named inventor on a patent application – it cannot. Despite this decision, AI continues to grow as a key driver of what the World Economic Forum calls the “Fourth Industrial Revolution” while simultaneously empowering innovation through AI’s capacity to augment and complement human ingenuity and streamline data processing. Indeed, even after the Thaler decision, interest continues in the implications for patenting of AI inventions. On Valentine’s Day, the USPTO even published questions for public comment that specifically focus on AI’s contributions to inventorship and the creative process (the comment period ends May 15, 2023). To understand some of the current developments, I attended the USPTO’s Artificial Intelligence (AI) and Emerging Technologies (ET) Partnership Series (AI/ET Conference) in Dallas, TX on February 8, 2023 and discuss some of the key takeaways from the conference here.
Status of the Thaler Saga
Dr. Stephen L. Thaler, Ph.D. is the inventor of DABUS (the “Device for Autonomous Bootstrapping of Unified Science”), an AI system that generates inventions. Dr. Thaler’s DABUS created two “inventions” for which he sought patent protection. The first is a “Neural Flame” and the second a “Fractal Container”. Understanding what these inventions do is less important than the fact that (1) Dr. Thaler maintains he did not contribute to the conception of these inventions and (2) he believes any person of ordinary skill in the art could reduce the DABUS inventions to practice. Thaler, 43 F.4th at 1209. The Patent Act 35 U.S.C. § 115 requires that inventors affirm or declare that they are the original inventor, or joint inventor, of a claimed invention in a patent application. Further, 35 U.S.C. § 100(f) defines said inventor as an “individual” or if a “joint invention” the “individuals” who collectively “invented or discovered the subject matter of the invention.” Likewise 35 U.S.C. § 100(g) defines a “joint inventor” or “coinventor” as “any 1 of the individuals who invented or discovered the subject matter of a joint invention”. With these definitions in mind, Dr. Thaler signed the oath and submitted statements to the USPTO listing DABUS as the inventor, and assigning “to himself all of DABUS’ rights as an inventor.” Id. at 1210.
Dr. Thaler’s patent applications naming DABUS as the inventor were rejected at every level. First, the USPTO rejected both applications, maintaining that the term “inventor” applied strictly to only “natural persons” and summarily denied the applications. Thaler then appealed to the United States District Court for the Eastern District of Virginia that likewise affirmed the USPTO’s decision on summary judgment, holding that the Patent Act requires an “inventor” to be a natural person. Id. at 1209. Thaler then appealed to the Federal Circuit.
In its opinion, the Federal Circuit affirmed that an AI software cannot be an inventor on a patent application because a plain language reading of the Patent Act requires inventors to be “natural persons” i.e. “human beings.” Id. at 1210. The Court’s analysis relied on their interpretation of the term “individual” as used in Patent Act §§ 100(f), 100(g) and 115. Id. at 1211. The Act does not define “individual”, but citing to the Supreme Court’s decision in Mohamad v. Palestinian Auth., 566 U.S. 449, 454 (2012), the Court explained that in normal “parlance”, “individual” always means a human being unless otherwise qualified. Id. “Congress understands ‘individual’ to indicate natural persons unless otherwise noted.” Id. The Court further explains that the Patent Act consistently refers to “individuals” internally through the personal pronouns “himself” or “herself” (See e.g.,35 U.S.C. § 115(b)(2)) providing further evidence that “individual” necessarily means a human being. Id. Lastly, the Court flat-out rejected Thaler’s arguments that other sections of the Act, such as 35 U.S.C. §§ 101 and 271 that use the pronoun “whoever” are broad enough to support non-human inventorship. Id. at 1212. The Court dismisses this idea, stating unequivocally that any analysis of those provisions necessarily leads back to an analysis of “inventor” and “individual” and that just because non-humans may infringe patents, it does follow that those provisions would change the clear definitions that require inventors themselves to be human beings. Id.
The Court thus affirmed the Eastern District of Virginia’s decision and on October 20, 2022 rejected Dr. Thaler’s petition for a rehearing en banc. Despite this rejection, Dr. Thaler plans to petition the United States Supreme Court for a writ of certiorari, and the High Court recently granted him an extension to file that petition until March 19, 2023. So while every judicial body has rejected his contentions thus far, the Saga indeed continues, and Dr. Thaler may yet find refuge at the Supreme Court.
The Tip of the AI-Iceberg – Open Patent Questions after Thaler
While the Federal Circuit’s Thaler decision seems to definitively end the question of AI-inventorship (pending a writ of cert), there are a number of AI concerns for which the opinion provides little guidance. I discuss a few of those here:
Joint-Inventorship: While it seems to logically follow from the reasoning in Thaler that an AI cannot further be a joint-inventor on a patent application (because joint-inventors must be inventors themselves by definition), the case does not directly address this issue. Joint inventorship generally requires that a putative joint inventor “(1) contribute in some significant manner to the conception or reduction to practice of the invention, (2) make a contribution to the claimed invention that is not insignificant in quality, when that contribution is measured against the dimension of the full invention, and (3) do more than merely explain to the real inventors well-known concepts and/or the current state of the art.” Plastipak Packaging, Inc. v. Premium Waters, Inc., 55 F.4th 1332 (Fed. Cir. 2022) quoting Pannu v. Iolab Corp., 155 F.3d 1344, 1351 (Fed. Cir. 1998). While facially, the simple solution would be to outright refuse patent applications where an AI is a joint-inventor, there no doubt will be litigation to clarify this dynamic, to say nothing as to what “contributions” by AI to inventorship might mean. But some courts in other countries such as the United Kingdom and New Zealand in addition to the European Patent Office have drawn a distinction between invention and joint-invention, suggesting that AI could be named as a co-inventor alongside at least one human inventor on a patent. Even if the current statutory language does not allow for such a conclusion under US law, perhaps this is something that will change in the future as AI develops.
Application Affirmations: If a human inventor earnestly believes that an AI has contributed in a “significant manner to conception, or reduction to practice”, and does not believe he can honestly affirm under 35 U.S.C. § 115 that he is the sole inventor (without naming the AI), what should that “inventor” do? This is a perplexing question. Should an inventor simply sign the affirmation and entirely discredit the AI’s contribution? Or should he name the AI and risk the USPTO rejecting the application for inventorship deficiencies under Thaler? Previously, Dr. Thaler himself chose not to list an older AI called the “Creativity Machine” as a joint-inventor on a 1995 patent application even though that AI assisted him in conceiving of the invention. Notwithstanding his “omission” that patent issued. See U.S. Patent No. 5,852,815 (“US ‘815”) for the “Neural Network Based Prototyping System and Method”. Similarly in 2005, Dr. John Koza refrained from listing an AI called the “Invention Machine” as a joint-inventor on his patent application despite the machine’s contribution to conception – that patent also issued. (See U.S. Patent No. 6,847,851 (“US ‘851”) for an “Apparatus for Improved General-Purpose PID and non-PID Controllers”). Thus far, there is no binding guidance.
POSITA: Although patent discussions concerning AI regularly focus on AI inventorship, equally as intriguing is whether (and to what extent) AI will impact other patent issues, like who the court determines is a “person of ordinary skill in the art”. Could this one day be an AI or an AI-assisted human? Furthermore, the POSITA is a material part of both the non-obviousness (35 U.S.C. § 103) and enablement (35 U.S.C. § 112(a)) analyses. How AI will affect these key patentability requirements remains unknown, but may prove to be an even more difficult question because it necessarily involves treating AI as a tool rather than simply wondering whether AI satisfies a statutory “inventorship” requirement.
Relatedly, some at the AI/ET Conference expressed concerns that widespread AI adoption in certain industries (or the lack thereof) could create a disparity of access to AI infrastructure, and result in major anticompetitive effects. For example, what may be obvious to an inventor at Google with tremendous access to AI assistance may not be so obvious to an inventor working at a smaller organization without such augmenting resources – even if that inventor works in the same field. In effect, the “state of the art” and level of skill of a POSITA could vary wildly even amongst peers within the same industry resulting in an inequality of “conception” ability.
No clear consensus emerged from these discussions, except that for one – inventors will always use the various tools at their disposal, and that AI is a uniquely distinct tool from those throughout history, in no small part, because it can actually contribute to conceiving an invention. No matter how advanced a hammer has become, no hammer or other tool to date has substantially contributed to conceiving a chair, an armoire or any other invention– but AI can do just that.
Given this uncertainty, would-be inventors working with or leveraging AI technologies (or their owners or assigns) may have justifiable concerns about whether listing AI contributions on a patent applications presents too much risk or that judicial uncertainty may yield unpredictable invalidity defenses for reasons like inventorship, enablement or non-obviousness. In light of this uncertainty, AI inventors could avoid patents all together in favor of alternative protections like copyright or trade secrets.
Trade Secrets: Possibly a One-Size-Fits-All IP Protection for AI Inventions
In lieu of pursuing patent protection, in everchanging fields like software or technology, trade secrets law may provide a simpler – but still effective – solution for IP protection. In general, trade secrets include a wide variety of information that has actual or potential economic value that derives from the fact that the information is secret. The Coca-Cola recipe is a prime example and provides a good analogy to AI inventions. This may not be as strange as it sounds, because just like how the public interacts with Coca-Cola without knowing the trade secret recipe, likewise consumers interact with an AI without knowing how the underlying code operates.
But as AI evolves, will this always be true? Since we do not still know all the potential evolutions and ramifications of this new technology (along with other emerging technologies like quantum computing), can we always rely upon trade secrets for protection of these inventions? Only time will tell. For now, we can only work with the IP protection tools that we have to cobble together as best protections as possible. The law is often one (or several) steps behind emerging technologies and playing catchup. As technologies advance more and more beyond our wildest imaginations, that may be a very tall order indeed – perhaps we should ask an AI how we solve that problem?