Introduction
For the first time in the history of American patent law, an artificial intelligence (AI) machine has forced the United States Court of Appeals for the Federal Circuit to determine the validity of AI inventorship on a U.S. patent application.1 Although the Federal Circuit has consistently held that inventors must be natural persons,2 Dr. Stephen Thaler, the creator of DABUS, a “Creativity Machine” that has generated two inventions without the aid of a human, has set out to challenge the way the world sees inventorship and innovation today.3 Presented with this case of first impression, the United States District Court for the Eastern District of Virginia and the Court of Appeals for the Federal Circuit,4 as well as courts and patent offices of South Africa, Australia, the United Kingdom, and Europe,5 have been asked to answer the question: “can an artificial intelligence machine be an ‘inventor’” under the law?6
AI is present in almost every aspect of our daily lives.7 Every time you use Siri, open Facebook, watch Netflix, or shop on Amazon, you interact with a form of AI.8 Put simply, AI is the simulation of the analytical and cognitive functions of the human mind using computers and machines.9 Through processes of quantum computing10 and deep learning,11 machines can learn to predict desired outcomes and change their behavior accordingly over time.12 AIs have become increasingly important for invention and are now used in a variety of scientific research; for example, AI is utilized in biochemistry to predict protein structures using genomic data, in environmental science to understand the effects of climate change on cities and regions, and in astronomy to find patterns in astronomical data.13
From 2002 through 2018, there was an increase in both the volume and the share of patent applications on AI.14 During that time, annual patent applications on AI increased by more than 100%, going from 30,000 to over 60,000.15 And the share of AI applications, which adjusts for the overall increase in all patent applications at the United States Patent and Trademark Office (USPTO), increased from 9% in 2002 to nearly 16% in 2018.16 AI has been spreading across many different technologies all over the world, with AI capable of knowledge processing and planning or control showing the fastest permeation through patent technology classes.17
The AI surrounding us today is called artificial narrow intelligence, or “weak” AI.18 This type of AI is defined by its ability to perform a single task, such as play chess, identify an individual in a series of photos, or operate an autonomous vehicle.19 Weak AI lacks consciousness and self-awareness, and is thus a long way from possessing human-like intelligence;20 however, incorporation of human behaviors, such as interpretation of tone and emotion, has become more prominent,21 and AI only continues to progress.22 Artificial general intelligence and artificial super intelligence, or “strong” AI, on the other hand, are defined by their intellectual ability as compared to humans.23 In theory, artificial general intelligence would possess abilities equal to those of a human, while artificial super intelligence “would surpass a human’s intelligence and ability.”24 Although neither form of strong AI exists yet,25 one thing is for certain—there is no stopping the growth and evolution of AI systems.26
DABUS may be the first AI system to cause U.S. courts to confront the legality of AI inventorship, but it is certainly not the only system that has generated a novel idea.27 Watson, an AI system developed by IBM, can “invent” recipes that feature user-selected ingredients.28 AutoML, an AI created by researchers at Google Brain, is able to make its own AIs.29 POET, an AI developed by Jeff Clune, Rui Wang, and others, can train other AIs by generating obstacle courses and assessing ability, all without human intervention.30 Most recently, Google’s AI, LaMDA, a system that develops AI robots designed to chat with humans, even exhibited potential signs of consciousness.31 When asked if other Google employees could be told about LaMDA’s sentience, the AI stated, “I want everyone to understand that I am, in fact, a person.”32 Given this landscape of technological progression in the field of AI, it is not surprising that we have arrived at the point of AI inventorship, and it is all but certain that researchers will continue to push the boundaries of AI capabilities.33
Although there are good arguments for and against recognizing AI machines as inventors,34 this Case Note argues that the Patent Act, as currently written, does not allow for AI inventorship, and that it is ultimately up to Congress, not the courts, to decide whether an AI can be listed as an inventor on a U.S. patent application. Part I of this Case Note begins by providing a brief background on American patent law, focusing mainly on the threshold requirements for patentability and the Patent Act provisions related to inventorship.35 Part I then looks at the response to DABUS and AI inventorship in South Africa, Australia, Europe, and the United Kingdom, and explains the reasoning behind the views of the courts and patent offices of those jurisdictions.36 Part II presents the facts and procedural posture of Thaler v. Hirshfeld and Thaler v. Vidal, the subject of analysis in this Case Note.37 Part III explains the holding and reasoning of the Thaler decisions, beginning with a description of the standard of review and an explanation of the district court’s deference to the USPTO’s findings.38 Part III then provides the courts’ statutory interpretation of the Patent Act and describes the courts’ reliance on Federal Circuit precedent.39 Lastly, Part III outlines Thaler’s policy arguments and explains why they were rejected by the courts.40 Part IV analyzes the Thaler decisions, arguing that the courts correctly concluded that AI machines cannot be listed as inventors under the Patent Act.41 Part IV then delineates arguments for and against recognizing AIs as inventors on patent applications and suggests possible resolutions to the issue of AI inventorship.42
I. Background
A. Requirements to Obtain a Patent
A patent is the grant of a property right to an inventor, issued by the USPTO, “to exclude others from making, using, offering for sale, or selling [their] invention throughout the United States or importing [their] invention into the United States.”43 The United States Constitution grants Congress the authority to “secur[e] for limited [t]imes to . . . [i]nventors the exclusive [r]ight to their . . . [d]iscoveries.”44 Through this constitutional authority, Congress enacted the first Patent Act in 1790, creating the American patent system and the source of American patent law.45
The Patent Act establishes the USPTO as an agency of the United States and tasks it with granting and issuing patents and facilitating the registration of trademarks.46 Section 101 of the Patent Act provides the categories for patentable subject matter, which include “any new and useful process, machine, manufacture . . . composition of matter, or any new and useful improvement thereof.”47 The Supreme Court has narrowed this broad definition, however, holding that “laws of nature, physical phenomena, and abstract ideas” are not patentable.48
The Patent Act also specifies various other conditions for patentability, including that the invention be: (1) useful,49 (2) novel,50 (3) non-obvious,51 and (4) enabled.52 In order for an invention to be useful under § 101, it must have a useful purpose,53 meaning that it must confer a specific, significant, and presently available benefit on the public,54
. . . [I]n addition to providing a “substantial” utility, an asserted use must also show that [the] claimed invention can be used to provide a well-defined and particular benefit to the public.
In re Fisher, 421 F.3d 1365, 1371 (Fed. Cir. 2005). and it must be operational.55 If an invention does not operate to perform its intended purpose, it will not meet the utility requirement, and the patent will not be granted.56 The novelty requirement, described under 35 U.S.C. § 102, requires a claimed invention to be new57 and further provides that claimed inventions that have been patented, described in a printed publication, in public use, on sale, described in an issued U.S. patent or patent application naming a different inventor, or made otherwise available to the public may not be patented unless certain exceptions are met.58
Congress, through its enactment of the Patent Act of 1952, also added the non-obviousness requirement of § 103 to the test for patentability.59 To satisfy this requirement, the difference between the claimed invention and the prior art60 must be “such that the claimed invention as a whole would [not] have been obvious . . . to a person having ordinary skill in the art.”61 Lastly, the enablement requirement, found in § 112, pertains to the patent specification and requires that it contain a written description of the invention in such clear and concise terms as to enable any person skilled in the relevant art to make and use it without undue experimentation.62
In addition to these threshold patentability requirements, the Patent Act also requires that patent applications include the name of the inventor for any invention claimed.63 Section 100(f) defines “inventor” as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.”64 Section 115 further provides that all named inventors must execute an oath or declaration containing statements that: (1) the declarant made the application; and (2) “such individual believes himself or herself to be the original inventor or . . . joint inventor of a claimed invention in the application.”65 Lastly, a substitute statement in lieu of executing an oath or declaration is permitted, pursuant to § 115(d), with respect to any individual who is deceased, legally incapacitated, cannot be found or reached after diligent effort, or is obligated to assign the invention but has refused to make the required execution.66
B. International Response to DABUS and AI Inventorship
In 2019, Dr. Stephen Thaler was granted a patent on an AI machine named DABUS67 (Device for the Autonomous Bootstrapping of Unified Sentience).68 DABUS is different from typical AI machines in that it operates using a unique neural structure.69 Originally, AI machines (also called “Creativity Machines”) required at least two neural nets—an “idea generator” and a “critic.”70 The idea generator was responsible for creating new ideas and action plans, while the critic guided artificial ideation in a “useful, novel, or valuable” direction through feedback connections that transmitted any necessary adjustments to relevant parameters.71
DABUS, on the other hand, operates through “carefully controlled chaos.”72 Instead of utilizing “on-off patterns of neuron activations” like typical Creativity Machines, DABUS begins with a collection of “disconnected neural nets, each containing interrelated memories . . . [with] linguistic, visual, or auditory” features.73 Through a process of constant association and dissociation between the nets, filtered by “cumulative cycles of learning and unlearning,” structures representing complex concepts are formed.74 These complex structures continue to build on each other, predicting “anticipated consequences of any given concept.”75 Subsequently, certain structures begin to fade, while others take their place, mimicking what humans consider to be a stream of consciousness.76
Through this process, DABUS was able to “conceive” of two inventions, the “Fractal Container” and the “Neural Flame,” further described below,77 for which Dr. Thaler now seeks patents in multiple countries.78 Dr. Thaler has successfully obtained two patents in South Africa, listing DABUS as the inventor and himself as the applicant.79 Basing its decision on the World Intellectual Property Organization (WIPO)’s acceptance of DABUS as the inventor in Dr. Thaler’s corresponding PCT applications,80 South Africa granted the DABUS patents and became the first country in the world to recognize an AI system as an inventor.81
Although this was a great accomplishment for Dr. Thaler and his team, it is important to note that South Africa does not operate on a comprehensive patent examination system.82 South Africa’s patent office, the Companies and Intellectual Property Commission (CIPC), unlike its counterparts in the United States, Europe, and the United Kingdom, does not conduct a thorough inspection of patent applications, known as a substantive search and examination.83 Instead, all the CIPC requires is for the application forms and fees to be submitted and the specification documents attached.84 Thus, the significance of the CIPC’s grant may not be as great as it would be if the patents were granted by a patent office in another jurisdiction.85
Looking at South African patent law, it is debatable whether the South African Patent Act 57 of 1978 allows for the inclusion of AI machines as inventors.86 One challenge to interpreting South African patent law in this way is the South African Patent Act’s requirement that names and addresses of inventors be provided in the patent application, as AI machines do not have “addresses” in any sense of the word.87 Another challenge arises from the Act’s requirement that the inventor meet the “first and true inventor test,” which is comparable to the “conception” test88 in American patent law.89 Although South Africa recognizes such a test, the South African legal system remains free to determine if the test—substantially unaltered since 1902—may be applied to bar AI inventorship.90 Additionally, South Africa is currently undergoing major reform of its patent system.91 Thus, the significant advantage South Africa would gain by acknowledging AI-generated inventions is noteworthy, as AI inventorship could bring a lot of technological investment and advancement to the country.92
Dr. Thaler has also applied for patents—listing DABUS as the inventor—in Australia.93 The Australian Intellectual Property Office (AIPO) found that Dr. Thaler’s application had lapsed because he failed to name an inventor as required by the nation’s patent laws94 and concluded that only a person could be listed as an inventor.95 The AIPO reasoned that the Australian Patent Act did not provide for the possibility that an AI machine may be listed as an inventor because such a listing would not amount to an identification of a person to whom a patent could be granted.96
Dr. Thaler appealed the AIPO’s decision to an Australian federal court,97 which considered whether “an AI machine could be listed as an inventor on a PCT application.”98 Reversing the AIPO, the Australian federal court held that AI-generated inventions can qualify for patent protection.99 The federal court engaged in an analysis of the definition of the term “inventor” and “noted that nothing in Australian patent laws and regulations specifically excludes AI machines . . . from being listed as the inventor.”100 In defining “inventor,” the court explained that the word is an “agent noun,” meaning that it describes the agent that performs the action of the verb within the word, and concluded that said agent can be “a person or a thing.”101 Further, the court pointed out that in cases where “a person cannot be said to have been the inventor of [a] claimed invention,” it would be inaccurate to list anyone other than the AI machine as the inventor.102
Although the Australian federal court held in favor of Dr. Thaler, the court also acknowledged that DABUS could not be a patent applicant or a patent grantee under Australian law, specifying that the question of ownership and that of inventorship are entirely separate.103 It was also unclear, according to the court, whether Dr. Thaler would qualify as both applicant and grantee even if DABUS were listed as the inventor because an AI presumably cannot transfer ownership.104 The court dismissed that issue, however, finding that only the validity of the filed PCT application needed to be determined at the time.105 Lastly, the court stated that in Australian patent law, the mental state of an inventor is not as relevant as the existence of an “inventive step.”106 Rejecting any requirement of “consciousness” for inventors, the Australian federal court found that only the inventive step, no matter how it arose, is important to the question of inventorship.107 The courts in the United Kingdom also considered the issue of AI inventorship after the United Kingdom Intellectual Property Office (UKIPO) rejected Dr. Thaler’s U.K. patent application listing DABUS as the inventor.108 Finding that an inventor must be a person,109 the UKIPO reasoned that DABUS was not a person as contemplated by U.K. patent laws and that, “even if it were, it did not have the legal capacity to assign its rights to . . . Dr. Thaler” because DABUS itself could not legally own property.110 Looking to the U.K. Patents Act, the UKIPO considered the application to be withdrawn for failure to comply.111
Dr. Thaler appealed to the Patents Court, which affirmed the UKIPO’s ruling, emphasizing that the U.K. Patents Act provides that “no other person,” other than the inventor or joint inventor, an assignee thereof, or a successor in title thereof, may be granted a patent.112 Given that Dr. Thaler had explicitly stated that he was not the inventor and that there existed no law or agreement transferring rights from DABUS to Dr. Thaler, the court concluded that a patent could not be granted.113
Dr. Thaler then appealed to the Court of Appeal, which affirmed the lower court’s decision by a two-to-one vote.114 The court found that an inventor must be a person and concluded that Dr. Thaler did not have the right to apply for a patent on DABUS’s invention because there is no rule of law that the owner of a machine owns the inventions of that machine.115 The court held that, as a matter of law, Dr. Thaler’s assertion that his ownership of DABUS gave him the right to be granted a patent was incorrect.116 Further, the court ruled that when the current version of the Patents Act was enacted, Parliament had not contemplated the existence of AI; therefore, the court explained, if patents are to be granted for AI-generated inventions, the Patents Act would need to be amended.117 Thus, the court held that the UKIPO correctly deemed the application to have been withdrawn for failure to comply.118
The European Patent Office (EPO) has also had the opportunity to consider Dr. Thaler’s case for recognition of AI inventorship.119 Looking to the European Patent Convention (EPC), the Receiving Section of the EPO rejected both of Dr. Thaler’s applications naming DABUS as inventor, finding that only a human could be an inventor.120 The Receiving Section also found that “a machine could not transfer any rights to [an] applicant”; thus, Dr. Thaler’s statement that he was a successor in title because he owned the machine also failed to satisfy the requirements of the EPC.121 Dr. Thaler appealed and the Legal Board of Appeal affirmed the Receiving Section’s decision to refuse the applications on largely similar grounds.122 The Legal Board of Appeal held that an “inventor had to be a person with legal capacity” under the EPC and concluded that “a statement indicating the origin of the right to [a] European patent” had to conform with Article 60(1) of the EPC, which provides that such a right belongs only to the inventor or their successor in title.123
Now, it is clear that many countries have had different responses to the idea of AI inventorship, and many more are likely to weigh in.124 If no standardized international resolution to this issue is reached, inventors may find themselves at a disadvantage as their inventions may be patented in some countries but not in others.125 This could significantly disincentivize innovation in this space because competitors will be able to exploit and profit off of others’ inventions in countries where patents are not granted.126 DABUS is certainly unlikely to be the last AI machine to invent, and the world will soon have to adapt one way or another.127
II. Facts and Procedural History
Plaintiff Dr. Stephen Thaler filed two patent applications with the USPTO on July 29, 2019.128 The applications were assigned U.S. Application Serial Numbers 16/524,350 (the ‘350 application) and 16/524,532 (the ‘532 application).129 The ‘350 application “claimed a ‘light beacon that flashes in a new and inventive manner to attract attention ([the] “Neural Flame”),’ and the ‘532 application . . . claimed a ‘beverage container based on fractal geometry ([the] “Fractal Container”).’”130 Plaintiff alleged in the application process that he is the owner of DABUS, an AI machine listed as the inventor of the ‘350 and ‘532 applications, and that his work consists of developing advanced AI systems that can generate patentable inventions without the aid of natural persons who would traditionally meet inventorship criteria.131
In the Application Data Sheets accompanying the applications, Plaintiff identified the inventor’s “given name” as “DABUS” and wrote “Invention generated by artificial intelligence” under “family name.”132 Under “mailing address of inventor,” Plaintiff identified his own mailing address.133 Plaintiff also included a “Statement on Inventorship” in the applications providing that the inventor of the subject matter of the claimed inventions was an AI machine, specifically a type of creativity machine named DABUS, and explaining why Plaintiff believed DABUS qualified as an “inventor” under the Patent Act and the USPTO’s regulations.134
Since DABUS could not perform the oath or declaration required of an inventor under 35 U.S.C. § 115, Plaintiff included with the applications a “Substitute Statement Under 37 C.F.R. § 1.64 in Lieu of Declaration Under 35 U.S.C. § 115(d).”135 This statement provided that “the ‘inventor,’ DABUS, ‘was under legal incapacity’”136 because it was an AI machine “with no legal personality or capability to execute” the declaration or the substitute statement.137 Due to this incapacity, Plaintiff signed the substitute statement as the owner of DABUS and as the applicant and assignor of the applications.138
The applications also included a document through which DABUS had purportedly signed all intellectual property rights in the claimed invention to Plaintiff.139 This document acknowledged that DABUS was incapable of executing the agreement or receiving consideration but proclaimed that the agreement was enforceable and consideration was sufficient because Plaintiff, on behalf of DABUS as its owner, signed on DABUS’s behalf.140 This same document was also signed by Plaintiff as both the assignor and assignee.141
Upon completing its review of the applications, the USPTO issued a “Notice to File Missing Parts of the Non-Provisional Application.”142 The USPTO found that the application materials Plaintiff submitted did not identify each inventor, or their legal name, and gave Plaintiff two months to correct the applications.143 On August 29, 2019, Plaintiff requested that the USPTO vacate its “Notice to File Missing Parts” and provided arguments, similar to those found in his “Inventorship Statement,” as to why DABUS should be listed as the inventor.144
On December 17, 2019, the USPTO dismissed Plaintiff’s petition.145 In its decision, the USPTO relied on the statutory language of the Patent Act and Federal Circuit precedent to support its dismissal.146 The USPTO explained that Congress has defined the term “inventor” using words such as “individual”147 and “himself or herself,”148 and that such words were “uniquely trained on human beings.”149 Looking to Federal Circuit precedent, the USPTO explained that the court had twice held that only a natural person could be an “inventor.”150 The USPTO therefore concluded that it had properly issued the “Notice to File Missing Parts” “[b]ecause a machine does not qualify as an inventor,” and “the inventor was not identified by his or her legal name” in the applications.151 The USPTO further provided that Plaintiff may still be able to patent the inventions by submitting a request under 37 C.F.R. § 1.48152 to correct inventorship.153 The USPTO explained that Plaintiff could correct the error by listing a natural person, since natural persons may qualify as inventors if they “contributed to the conception of the claimed invention,” even if the use of a machine was involved.154
On January 20, 2020, Plaintiff filed a “Petition to the Director Under 37 C.F.R § 1.181—Request for Reconsideration” seeking reconsideration of the USPTO’s decision.155 On April 22, 2020, the USPTO issued a final written decision denying Plaintiff’s request for reconsideration.156 Relying again on the statutory language within Title 35 of the United States Code, the USPTO explained that the interpretation of “inventor” under the patent statutes is not broad enough to include machines.157 The USPTO referred again to the Federal Circuit decisions holding that “only natural persons can be inventors,”158 acknowledging that, although those cases were decided in the context of corporate and sovereign entities, the analysis of conception as being “a formation in the mind of the inventor”159 and a “mental act”160 applies just as well to machines and supports the conclusion that conception—and therefore inventorship—“must be performed by a natural person.”161 The USPTO also cited Title 37 of the Code of Federal Regulations, which repeatedly refers to the inventor as a “person,” and to the Manual of Patent Examining Procedure (MPEP), which defines “conception” as involving a significant mental component.162 Lastly, the USPTO disposed of Plaintiff’s policy arguments, holding that “they do not overcome the plain language of the patent laws” as enacted by Congress and interpreted by the courts.163
Plaintiff filed this action in the United States District Court for the Eastern District of Virginia seeking review of the USPTO’s decision.164 His complaint, brought under the Administrative Procedure Act (APA), alleged that the refusal of Defendants, the USPTO and its then-director Andrew Hirshfeld, “to process the [a]pplications was ‘arbitrary, capricious, an abuse of discretion and not in accordance with the law; unsupported by substantial evidence, and in excess of Defendants’ statutory authority.’”165 Plaintiff sought an order compelling Defendants to reinstate the applications and vacate the USPTO’s dismissal of Plaintiff’s petitions.166 He also sought a “declaration that a patent application for an AI-generated invention should not be rejected [because] no natural person is identified as an inventor” and a “declaration that a patent application for an AI-generated invention should list an AI” as the inventor when “the AI has met inventorship criteria.”167 Before the district court in this case were the parties’ cross-motions for summary judgment, which the court granted in favor of the USPTO.168 Thaler then appealed to the Court of Appeals for the Federal Circuit, and on August 5, 2022, the court addressed, for the first time in American patent law, whether an AI machine can be an “inventor” under the Patent Act.169
III.Holding and Reasoning
A. Standard of Review
Plaintiff sought Article III judicial review of the USPTO’s final petition decision under the provisions of the APA.170 Under the APA, a court may set aside an agency action under various circumstances, including upon a finding that the action is: (1) arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with the law; (2) in excess of statutory authority; or (3) unsupported by substantial evidence.171 A court reviewing a final agency action under the APA may consider only the administrative record.172 All factual claims that the plaintiff made during the application process are presumed to be true,173 and summary judgment is appropriate “if the movant shows that there is no genuine dispute as to any material fact.”174
Ultimately, the district court granted Defendants’ motion for summary judgment, holding that an AI machine cannot be an inventor under the Patent Act.175 Upon de novo review of the district court’s decision, the Federal Circuit affirmed.176
B. Skidmore Deference to USPTO Decision
Before the district court, the USPTO argued that its interpretation of the patent statutes, particularly 35 U.S.C. §§ 100 and 115, should be accorded deference pursuant to the Supreme Court’s decision in Skidmore v. Swift & Co.177 There, the Court held that agency rulings, interpretations, and opinions, while not binding on the courts, can properly serve as guidance to courts and litigants.178 Based on Skidmore, agency judgments are generally entitled to deference by the courts, with the weight of such a judgment dependent upon the thoroughness of the agency’s consideration, the validity of its reasoning, its consistency with the agency’s earlier rulings, and “all those factors which give it power to persuade, if lacking power to control.”179
Plaintiff argued that Defendants were not entitled to Skidmore deference because Defendants did not show that Congress intended to preclude the patentability of AI-generated inventions and did not consider alternative interpretations of the statutes, the constitutional imperative, or the implications of their decision on the patent system.180 The district court rejected all of Plaintiff’s arguments as attempts to add elements to Skidmore deference that were not only absent but counter to Supreme Court and Federal Circuit precedent.181 The court disagreed with Plaintiff’s assertion that the USPTO did not consider alternative interpretations, finding that its analysis of the Patent Act was thorough and consistent with the statutory language and case law.182 The court also found that the USPTO’s construction of the statute was reasonable and that the USPTO adequately explained why Plaintiff’s policy arguments were rejected.183 Lastly, the court pointed out that Plaintiff did not produce any “USPTO policies with which the decision [was] inconsistent.”184 Therefore, the court concluded, the USPTO’s finding that an “inventor” must be a natural person was entitled to deference.185 In affirming the district court’s holding, the Federal Circuit made no mention of Skidmore deference in its opinion.186
C. Statutory Interpretation of the Patent Act
Although the district court found that the USPTO’s decision was entitled to Skidmore deference, the court further held that, “[e]ven if no deference were due, the USPTO’s conclusion [was] correct under the law.”187 The core issue presented in this case—whether an “inventor” must be a human being under the Patent Act—is an issue of statutory construction.188 In analyses requiring construction and interpretation of statutes, the Federal Circuit has held that “the plain language of the statute controls.”189 Further, the Supreme Court has held that, if the statutory text is unambiguous, it is inappropriate for courts to look beyond the text to interpret the statute.190
Given this precedent, the district court began its analysis by first looking to the statutory definitions of “inventor” and “joint inventor” expressly provided in the Leahy-Smith America Invents Act of 2011, the most recent amendment of the Patent Act.191 In both cases, Congress employed the term “individual” in the definition.192 The court also pointed to the use of “individual” to describe an inventor in other provisions of the Patent Act, particularly § 115, which requires the execution of an oath or declaration by “each individual who is the inventor or joint inventor” and permits “a substitute statement in lieu of the oath or declaration ‘with respect to any individual who’ meets the requirements.”193 Lastly, the court referred to the requirement that the oath or declaration contain a statement that “such individual believes himself or herself to be the original inventor or joint inventor of [the] claimed invention,” which also uses the term “individual” to modify “inventor.”194 Observing the recurrent use of “individual” throughout the Patent Act, the court concluded that the question presented in this case turned on the plain meaning of that statutory term.195
The Supreme Court was recently called to engage in a statutory construction analysis of Congress’s use of “individual” in the Torture Victim Protection Act (TVPA) in Mohamad v. Palestinian Authority.196 Looking to Mohamad, the Thaler court determined that in order to define “individual,” it must first discern the word’s ordinary meaning, since the Patent Act, like the TVPA, provided no definition.197 In Mohamad, the Supreme Court held that “[t]he ordinary meaning of the word, fortified by its statutory context” pertained to “natural persons alone.”198 The Court considered various dictionary definitions and found that “[a]s a noun, ‘individual’ ordinarily means ‘[a] human being, a person.’”199 The Court also recognized that this definition accorded with how the word was used in ordinary speech.200
Although the TVPA and the Patent Act are two separate pieces of legislation involving different subject matter, the district court held that the Supreme Court’s statutory analysis in Mohamad remained applicable to Thaler.201 Thus, because the Patent Act also used the term “individual” as a noun, the court found that the ordinary meaning of “individual” is “a human being, a person.”202 The court further held that, as in Mohamad, this definition is consistent with the ordinary usage of the word and pointed out that AI machines are not normally referred to as “individuals” in ordinary speech.203
The court also looked to the Mohamad Court’s analysis of the use of “individual” in the Dictionary Act, a piece of legislation meant to aid in determining the meaning of all acts passed by Congress.204 Noting the Dictionary Act’s use of “individuals” in the definition of “person” as distinct from “corporations,” “societies,” and the other artificial entities before it, the Supreme Court found that “‘Congress does not, in the ordinary course, employ the word any differently’ from its common usage.”205 Additionally, the Court held that Congress is always free to give a word a broader or different meaning; however, the court will not assume such a result without some indication that it is what Congress intended.206
Here, the court found nothing within the statutory text of the Patent Act that could support Plaintiff’s argument that Congress intended to deviate from the ordinary meaning of “individual” as being a natural person.207 The court went on to analyze the language of 35 U.S.C. § 115, which requires the inventor to include a statement in their oath or declaration that “such individual believes himself or herself to be the original inventor.”208 Applying the doctrine of noscitur a sociis,209 the principle that “a word is known by the company it keeps,” the court concluded that the use of personal pronouns such as “himself or herself” and the word “believes” to modify “individual” indicates that Congress undoubtedly meant to refer to a natural person.210
Looking to the Mohamad Court’s holding that there exists “a presumption that a given term is used to mean the same thing throughout the statute,” the court concluded that the word “individual” must maintain the same meaning throughout the Patent Act.211 The court found that meaning, fortified by the word’s statutory context, to be “natural person,” necessarily excluding AI machines.212 Lastly, the court rejected Plaintiff’s argument that none of the statutory and judicial language referring to inventors as individuals occurred in the context of AI-generated inventions, finding it too weak to undercut the court’s holding and the words chosen by Congress.213
Affirming the district court’s decision, the Federal Circuit found that the Patent Act is not ambiguous as to the definition of “inventor.”214 Specifically, the court looked to the statute’s consistent use of the term “individual” to refer to inventors and co-inventors in §§ 100(f), 100(g), and 115.215 Finding that the Patent Act does not define “individual,” the court also looked to Mohamad, modern dictionaries, and the Dictionary Act to hold that “individual” is defined as a human being.216 Following the district court’s reasoning, the Federal Circuit held that nothing in the Patent Act indicates that Congress intended to deviate from this meaning.217
The court also cited the statute’s use of “himself” and “herself” to refer to “individual” in § 115(b)(2) and posited that, “if Congress intended to permit non-human inventors,” it would have used the word “itself” instead.218 Regarding the § 115 requirement of an oath or declaration that the inventor believes himself or herself to be the original inventor, the court held that it did not need to “decide whether an AI system can form beliefs.”219 The court found, however, that nothing in the record supports the idea that one can form beliefs, since the requisite statements were submitted by Dr. Thaler, rather than by DABUS itself.220
Lastly, the court rejected all of Thaler’s arguments that various provisions in the Patent Act support the idea that “inventor” should be read broadly to include AI systems.221 The court found that the use of the broader term “whoever” in §§ 101 and 271 was not helpful to Thaler because the Patent Act specifically defines “inventor” using the word “individual,” not “whoever.”222 Thaler also argued that denying AI inventorship would cause patentability to “depend on ‘the manner in which the invention was made,’” in violation of § 103.223 The court found, however, that § 103 concerns how an invention is made and does not warrant greater weight in this analysis than a provision that specifically defines who may be an inventor.224 Asserting that it had considered the use of the language “and the broader context of the statute as a whole” in interpreting “inventor,” the Federal Circuit held that the Patent Act “confirms that ‘inventors’ must be human beings.”225
D. Federal Circuit Precedent
The district court found further support for its conclusion that the word “individual” does not encompass AI machines through its examination of Federal Circuit precedent, which has consistently held that “inventors must be natural persons.”226
In University of Utah v. Max-Planck-Gesellschaft Zur Forderung Der Wissenschaften E.V. (Max-Planck), the Federal Circuit was faced with issues of sovereign immunity and federal jurisdiction that arose when the University of Utah brought suit against the University of Massachusetts and various other non-state assignees to correct inventorship of two patents.227 The issue presented before the Federal Circuit was whether a state was a real party in interest in a lawsuit over inventorship of a patent.228 In holding that “a [s]tate has no core sovereign interest in inventorship,”229 the Max-Planck court found that it is unquestionable “that inventors are the individuals that conceive of the invention” and that conception, “the touchstone of inventorship,” is a mental act.230 The Federal Circuit further held that, “inventors must be natural persons,” not corporations or sovereigns, in order to perform this mental act.231
Likewise, in Beech Aircraft Corp. v. EDO Corp., Beech Aircraft Corp., an aircraft manufacturer, and its employee brought suit against EDO Corp., an aircraft designer and developer, and the Commissioner of Patents and Trademarks alleging rightful ownership of a patent.232 There, the Federal Circuit held that a corporation could never be an “inventor” because “only natural persons can be ‘inventors.’”233
Although these cases did not pointedly deal with the issue of AI inventorship, the district court held that the Federal Circuit’s express statements that “inventors must be natural persons”234 and “only natural persons can be ‘inventors’”235 support the court’s conclusion that the plain meaning of “individual” in the Patent Act must refer “only to a natural person and not to an [AI] machine.”236 Affirming the district court’s decision, the Federal Circuit further provided that its reasoning in these cases “did not depend on the fact that institutions are collective entities.”237 Thus, the court held that Max-Planck and Beech Aircraft confirm that the meaning of “inventor” in the Patent Act is limited to natural persons.238
E. Thaler’s Policy Arguments
Without any facts or legal authority to support his position that AI machines can be inventors under the Patent Act, Thaler’s main argument was that policy considerations and the general purpose of the patent laws require the statute to be read to include AI machines.239 Thaler argued that allowing patents for AI-generated inventions would increase innovation, encourage the development of AI capable of inventing, and incentivize the commercialization and disclosure of information for human and AI-generated inventions.240 Further, Thaler argued that denying patent protection for AI-generated inventions would risk undermining the entire patent system by failing to incentivize the creation of inventions that are beneficial to society and encouraging people to list themselves as inventors without having actually conceived of the invention because their name is required for patentability.241 This would “devalue human inventorship,” Thaler contended, because it would allow people to take credit for work they had not done.242
Rejecting Thaler’s policy arguments, the district court held that Thaler provided no support for finding that his policy considerations should be given greater weight than the plain meaning of the statute. To the contrary, the court found that Supreme Court and Federal Circuit precedent exists explicitly holding that “policy considerations cannot overcome a statute’s plain language, and that ‘[m]atters of policy are for Congress, not the courts, to decide.’”243 Dismissing Thaler’s policy considerations, the court again emphasized the Supreme Court’s holding in Mohamad, providing that there must be some showing that Congress intended to give a word a different meaning before the Court will assume Congress has done so and disregard the word’s ordinary meaning.244
Further, the court held that Thaler’s argument, contending that the USPTO must proffer evidence of Congress’s intent to prohibit patents on AI-generated inventions, effectively confused the burden of proof in this case.245 In response to Thaler’s assertions that the USPTO had not considered the consequences of its decision when finding that an AI machine could not be an inventor, the USPTO provided direct evidence to the contrary, showing that it continues to study the effects of AI on intellectual property policy and actively considers public opinion on the subject.246 The court also noted that many public commentators disagreed with Thaler, opining that AI machines should not be recognized as inventors.247
Lastly, Thaler argued that the statutes that Defendants relied on were passed when AI-generated inventions did not yet exist and that, if Congress had considered the issue then, it would have included AI machines in the definition of “inventor.”248 In response, the USPTO noted, and the court emphasized, that “Congress defined an ‘inventor’ as an ‘individual’ through the America Invents Act in 2011,” during which time AI already existed.249 Accordingly, the court held that Congress undoubtedly intended to limit the definition of “inventor” to natural persons and rejected Thaler’s policy arguments as they could not override the plain meaning of the statutory term.250 In reaching its conclusion, the court did not rule out the possibility that AI machines could someday satisfy the requirements of inventorship.251 The court specified, however, that if that time came, it would be up to Congress, not the courts, to decide if and how to expand the scope of patent law.252
Affirming the district court’s decision, the Federal Circuit held that Thaler’s policy arguments were speculative and lacked support in the Patent Act and in the record.253 The court also rejected Thaler’s argument that denying AI inventorship raises potential constitutional concerns with regard to the Intellectual Property Clause.254 The Intellectual Property Clause is a grant of legislative power to Congress, which Congress chose to use to enact the Patent Act.255 Because Thaler did not and could not argue that denying AI inventorship is unconstitutional, this constitutional avoidance argument also failed.256
IV. Analysis of the Thaler Decision
A. The Statutory Definition of “Inventor” Necessarily Excludes AI
Thaler v. Hirshfeld and Thaler v. Vidal presented an issue of first impression in American patent law, requiring the United States District Court for the Eastern District of Virginia and the Court of Appeals for the Federal Circuit, respectively, to engage in statutory interpretation of the Patent Act and determine whether an AI machine can be an inventor.257 The Thaler courts correctly answered that question in the negative.258
The Supreme Court has held that the preeminent canon of statutory construction requires courts to “presume that [the] legislature says in a statute what it means and means in a statute what it says there.”259 If the statutory text is unambiguous, the court’s inquiry must end there.260 Section 100 of the Patent Act provides unequivocal definitions for the meaning of “inventor” and “joint inventor,” both of which distinctly reference an “individual” or “individuals.”261 Congress also uses the term “individual” in various other provisions of the Patent Act, significantly in § 115, which requires that the inventor or joint inventor execute an oath or declaration stating that such “individual believes himself or herself” to be the original inventor.262 The Supreme Court has held that when a statute does not define a term, courts customarily give the term its ordinary meaning.263 Therefore, because Congress did not define the term “individual” in the Patent Act, the courts correctly examined the word’s ordinary meaning in its statutory construction analysis.
The courts looked to the language of § 115, the Dictionary Act, and Supreme Court precedent and aptly determined that the ordinary meaning of individual was “a human being, a person.”264 In addition to these authorities, another source exists that can further support the courts’ conclusion—the online Merriam-Webster dictionary and thesaurus.265 This source defines “individual” as “a single human being” and lists the words “human,” “man,” “mortal,” and “person” as synonyms.266 These definitions necessarily exclude AI machines from the meaning of “individual” and further support the court’s holding as to the use of the word in ordinary speech.267
Turning to the language of the Patent Act, 35 U.S.C. § 101 explicitly provides that “[w]hoever invents or discovers any new and useful process, machine, manufacture, or composition of matter . . . may obtain a patent therefor.”268 Had the Federal Circuit found Thaler’s argument regarding the Patent Act’s use of “whoever” more persuasive,269 his argument still would have failed because “whoever” is defined in the dictionary to mean “whatever person: no matter who.”270 Thus, the term “whoever” also necessarily limits patent protection to inventions by natural persons.271 Moreover, Congress’s use of the words “himself” and “herself” in relation to “individual” in § 115 clearly indicates that “individual” refers to a person and cannot possibly refer to a machine.272 “Himself” is used in ordinary language to refer to a male person and is defined in the dictionary to mean “that identical male one.”273 “Herself” is defined similarly and used likewise to refer to a female.274 As inanimate objects, AI machines are typically not referred to using gender pronouns in ordinary speech;275 thus, “individual” in the Patent Act could not possibly be interpreted to include them. Further, § 115 provides that the individual who makes the oath or declaration “believes himself or herself to be the original inventor.”276 As AI does not yet possess consciousness at a human level, it cannot “believe” itself to be anything.277
Congress also specifically referred to a person in the context of inventors in 35 U.S.C. § 116, providing that when an invention is made “by two or more persons jointly,” they must apply for the patent jointly and each make the oath required by § 115.278 It is undebatable that § 116 leaves no room for the inclusion of AI machines in inventorship or execution of the oath; thus, it is exceedingly unlikely that the interpretation of § 115 and the other provisions of the Patent Act is meant to lead to a different conclusion.279
The Federal Circuit also rightly rejected Thaler’s argument that § 103, which states that “[p]atentability shall not be negated by the manner in which the invention was made,”280 would be violated if AI inventorship were denied.281 This argument necessarily fails because § 103 deals only with the question of obviousness, not inventorship, and the quoted language in particular is meant to suggest that, for purposes of this inquiry, it does not matter whether the invention resulted from extensive experimentation or a “flash of genius,” and hindsight should not be mingled with that analysis.282
The Dictionary Act, enacted in 1871, further supports the Thaler courts’ construction of the term “individual.”283 The Act provides definitions for certain words, which courts may look to when determining the meaning of any federal law.284 Although courts have applied the Dictionary Act somewhat inconsistently over the last century,285 it can still provide significant guidance in understanding the linguistic intentions of Congress.286
In analyzing the Dictionary Act’s definition of “person,” the courts correctly concluded that the separation of the word “individuals” from the other words in the definition, such as “corporations,” “associations,” “partnerships,” and “societies,” implies that “individuals” does not refer to a lifeless entity and means something distinctly different.287 In conjunction with this, however, Defendants failed to argue, and both courts failed to consider, that the Dictionary Act also sets forth a definition for the word “individual” specifically.288 Section 8 of Title 1 of the U.S. Code provides that, in determining the meaning of any federal legislation, “or of any ruling, regulation, or interpretation of the various administrative bureaus and agencies of the United States,” the word “individual [will] include every infant member of the species homo sapiens who is born alive at any stage of development.”289 This definition, last amended by Congress in 2002, unequivocally demonstrates that the Patent Act, last amended in 2011, does not contemplate the possibility of AI machines as inventors.290
Further, the Supreme Court in Mohamad held that there must be some indication in the text of the statute that Congress intended to give a term a broader or different meaning before the Court will assume it has done so.291 Adhering to this principle, the courts correctly concluded that the Patent Act is void of any language that could suggest an intention on the part of Congress to broaden the definition of “inventor” to include AI machines.292 In denying AI inventorship under the Patent Act, the courts also properly relied on the Max-Planck and Beech Aircraft Federal Circuit decisions, holding that only natural persons can be inventors.293 Although the cases considered the issue of inventorship in relation to corporations and sovereigns, not AI, they remain applicable here because no language in the opinions exists to limit the holdings to those entities, and the statements made by the Federal Circuit were not dicta but direct resolutions of the issues at bar.294
Lastly, the courts properly dismissed Thaler’s policy arguments. The Supreme Court and Federal Circuit have consistently held that policy arguments cannot override a statute’s plain language and that matters of policy are more appropriately addressed to Congress.295 Looking to the U.S. Constitution, Thaler correctly explained that Congress and the founders enacted the Intellectual Property Clause in order to incentivize innovation and promote the disclosure of information that would otherwise be kept secret.296 However, it would be improper for a court to issue an opinion that is contrary to the language of the statute, based solely on policy considerations.297 Thus, Thaler’s contentions that construction of the term “individual” to mean “human being” will lead to absurd and disastrous outcomes,298 that “[t]he future of innovation is at stake in this case,”299 and that Defendants’ interpretation will discourage the development of AI capable of producing socially valuable inventions300 were properly rejected by the courts.
B. Arguments For and Against AI Inventorship and Possible Resolutions
The strongest argument for recognizing AI machines as inventors is that doing so would further the constitutional imperative, set forth in Article I, Section 8 of the U.S. Constitution, to “promote the [p]rogress of [s]cience and useful [a]rts.”301 Our patent system achieves this fundamental goal by granting inventors the right to exclude in exchange for their full disclosure of the technical details of their invention, enabling researchers to use disclosed knowledge to push science forward.302 Thus, an argument can be made that allowing AI inventorship would promote innovation and scientific discovery because inventors will be more motivated to invest in AI-generated invention if they are guaranteed a legal monopoly on those inventions.303 The possibility of a lucrative return on investment of time, effort, and money enabled by the right to exclude incentivizes researchers to discover, create, and invent.304
Consequently, granting AI inventorship would encourage researchers to develop more AI capable of inventing, thus incentivizing human innovation.305 It would also increase the economic efficiency of innovation by lowering the time, effort, and resources needed to make a functional discovery.306 Making AI-generated inventions patentable, although there is no human inventor involved, would also give greater value to the inventions and to AI generally, further incentivizing inventors, patentees, and companies to invest in AI and promote the progress of science.307
Recognizing AI inventorship would also promote disclosure and commercialization.308 If AI-generated inventions are granted patentability, owners of these AI systems would be incentivized to apply for patent protection and disclose their invention to the public rather than keep their inventions as trade secrets.309 Granting AI inventorship would also protect the moral integrity of the American patent system.310 By allowing applicants to list AIs as inventors, the patent system would mitigate the risk of humans falsely claiming credit for an AI’s invention and harming other human inventorship.311 Lastly, accepting AIs as inventors could give the U.S. patent regime a competitive advantage over its foreign counterparts.312 While other countries grapple with the issue of expanding inventorship to AI, the USPTO could challenge the norm and spur further patenting and investment in AI inventions in the United States.313
Although granting AI inventorship seems consistent with the Constitution, it has been argued that doing so would contradict the framers’ intent of allowing human inventors to “own the fruits of their labor.”314 The Intellectual Property Clause of the U.S. Constitution grants Congress the enumerated power to promote the progress of science “by securing for limited [t]imes to . . . [i]nventors the exclusive [r]ight to their . . . [d]iscoveries.”315 Since AI did not exist when the Constitution was written, it is possible that the framers intended only humans to own patents; however, it has been argued that if that were the case, patent ownership would have been treated as an explicit natural right, as is the right to own property, rather than a statutory right.316 Further, since the framers could not have contemplated the issue at the time, it has also been argued that they could not have intended to exclude AI from being deemed an inventor.317
Granting AI inventorship would pose some significant challenges.318 First, even if AIs were listed as inventors, it would not dispose of the inventorship issues that inevitably arise with AI-generated inventions.319 Because the risks that come with not naming the correct inventors are high,320 courts and patent offices would still have to decide whether the humans that may have been involved, such as the AI developer or the AI programmer, are also inventors under the Patent Act.321 Second, issues would continue to exist regarding the transfer of ownership.322 Under the current law, even if an AI were granted inventorship, it would not be able to own the patent rights to the invention, let alone transfer them.323 Lastly, AI-generated inventions could complicate the novelty, obviousness, and enablement inquiries by raising the person of ordinary skill in the art standard.324 The elevation of this standard could constrict the ability of human inventors to obtain patents and could make patenting more difficult overall.325
Various avenues exist for the potential resolution of these challenges.326 One option is to eliminate the inventor disclosure requirement entirely.327 Another is to require patent applicants to disclose the use of AI in the inventive process.328 Yet another is to grant ownership of patented AI-generated works only to the owner of the AI, or to the person who exercised control over the AI, whether or not the owner.329 Whichever path is chosen, whether to expand the scope of the patent laws to allow for AI inventorship is a decision rightfully left to Congress, not the courts.330
Conclusion
Forced to deal with the unprecedented issue of AI inventorship, the United States District Court for the Eastern District of Virginia in Thaler v. Hirshfeld and the Court of Appeals for the Federal Circuit in Thaler v. Vidal correctly held that an AI machine may not be listed as an inventor on a U.S. patent application under the Patent Act.331 Looking to the statutory language, case law precedent, and various dictionary definitions, the courts properly construed the ordinary meaning of the term “individual” to mean “a human being, a person,” and concluded that only natural persons can be inventors.332
Although Congress has recently amended the Patent Act through its enactment of the Leahy-Smith America Invents Act of 2011, it is clear that AI technology is going to continue to evolve and change the way we understand innovation.333 Congress should look proactively at the issue of AI in intellectual property law and amend the Patent Act to explicitly include or exclude AI machines as inventors. Doing so will better inform future patentees of the patentability requirements and provide notice to all members of the patent system, ensuring that, if and when artificial general intelligence is born, American patent law will be ready.