AI Systems May Invent, But Are They Inventors?
Published: August 11, 2022
Previously, the Court of Appeals for the Federal Circuit (“Federal Circuit”) has found that a non-human may infringe patents. Arguably, an AI system, which is a non-human, can also create or invent. But can an AI system be a named inventor on a patent? The Federal Circuit recently addressed this issue in Thaler v. Vidal.
With the advances in AI technologies, AI systems create drugs, treatments, designs, and more. While these systems may have been programmed and/or trained by a human, the human may not have actually invented the apparatus or method claimed in the patent application. That may have been done by the AI system, which raises the question as to who is the inventor of the invention created by that system. The human programmer/trainer, the non-human AI system itself, the owner of the AI system, or someone else. Alternatively, if the Patent Act requires inventor(s) must be human, are AI-created inventions not patentable at all under the current statute?
Stephen L. Thaler, assignee, filed U.S. Patent Appl. No. 16/524,350 (“the ’350 Appl.”) entitled, “Devices and Methods for Attracting Enhanced Attention” with the United States Patent and Trademark Office (“USPTO”). The USPTO requires an application data sheet (“ADS”) that lists the identity of each inventor by his or her legal name. In the matter of the ’350 Appl., the ADS was filed with the sole inventor’s given name as “DABUS,” which stands for Device for the Autonomous Bootstrapping of Unified Sentience. Instead of a last name, Thaler stated “the invention [was] generated by artificial intelligence.”
The USPTO issued a Notice of Missing Parts for failure to properly identify the inventor(s) on the ADS. Thaler petitioned the USPTO to vacate the Notice of Missing Parts, asserting the claimed invention was developed by a creativity machine named DABUS, which was “trained with general information in the field of endeavor to independently create the invention,” and that inventors should not be limited to natural persons. In 2020, the USPTO refused to vacate the Notice of Missing Parts, pointing to the Patent Act’s text as support that the inventor must be a natural person. Therefore, the USPTO found the Patent Act would not allow AI machines to be inventors.
Thaler appealed to the U.S. District Court for the Eastern District of Virginia, which granted summary judgment in favor of the USPTO. Then Thaler appealed the district court’s ruling to the Federal Circuit.
The Federal Circuit recognized the Patent Act defines an “inventor” as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.” While the Patent Act does not define “individual,” the U.S. Supreme Court “has explained, when used ‘[a]s a noun, ‘individual’ ordinarily means a human being, a person’” in the context of a statute unless Congress provides some indication it intended a different meaning.
The Federal Circuit found no indication Congress intended a different meaning for the term “individual” in the Patent Act. Rather the Patent Act supports that an “individual” refers to a human being. For example, the statute refers to “himself” and “herself” rather than “itself” when referring to “individuals.” The Patent Act also requires an oath or declaration from the inventor who “believes himself or herself to be the original inventor or joint inventor of a claimed invention.” The Federal Circuit stated that “[w]hile we do not decide whether an AI system can form beliefs, nothing in our record shows that one can, as reflected in the fact that Thaler submitted the requisite statements himself, purportedly on DABUS’ behalf.”
Thaler made several arguments as to why “‘inventor’ should be broadly read to include AI software.” First, Thaler argued the word “whoever” in 35 U.S.C. §§ 101 and 271 includes references to corporations and other non-human entities as potential infringers. But the Federal Circuit has previously ruled that neither corporations nor sovereigns can be inventors. Consistent with these prior rulings, the Court was not persuaded that the use of “whoever” in §§ 101 and 271 was relevant in determining qualifications for inventors, stating, “[t]hat non-humans may infringe patents does not tell us anything about whether non-humans may also be inventors of patents.” Further, the Patent Act’s definition of “inventor” uses the word “individual” rather than the word “whoever.”
Second, Thaler argued that AI software programs qualify as inventors because “otherwise patentability would depend on ‘the manner in which the invention was made,’ in contravention of 35 U.S.C. §103.” The Federal Circuit disagreed, stating that Section 103 is not about inventorship but rather about obviousness and “that inventions may still be nonobvious even if they are discovered during ‘routine’ testing or experimentation.” Thus, §103 “relates to how an invention is made” and cannot override “a provision that specifically addresses who may be an inventor.”
Thaler also argued that “inventions generated by AI should patentable in order to encourage innovation and public disclosure.” The Federal Circuit found such speculative policy arguments that lack a basis in the text of the Patent Act do not change the result when the text of the statute is unambiguous. The Court is not allowed to “elevate vague invocations of statutory purpose over the words Congress chose.”
Further, Thaler “invokes the canon of constitutional avoidance,” arguing that “permitting AI programs to be inventors” “support[s] the constitutional purpose of patents ‘[t]o promote the progress of science and useful arts.’” Thaler believes “not recognizing AI as an inventor undermines such progress, raising potential constitutional concerns [the Court] should be careful to avoid.” Again, the Federal Circuit disagreed, stating that the constitutional provision cited by Thaler grants legislative power to Congress, and Congress acted on that power by passing the Patent Act. “[T]he canon of constitutional avoidance is simply inapplicable.”
As a result, the Federal Circuit affirmed the lower court ruling, finding Thaler’s argument unpersuasive and that “there is no ambiguity: The Patent Act requires that inventors must be natural persons; that is, human beings.”
While the decision of the Federal Circuit comports with those of most courts around the world that have addressed this issue, there are exceptions. For example, South Africa granted a patent with DABUS as an inventor. But as the Federal Circuit pointed out, South Africa was not interpreting the United States Patent Act, so South Africa’s decision had no bearing on the Federal Circuit’s ruling.
Given the statutory basis for the Federal Circuit decision, it seems unlikely the Supreme Court would come to a different decision if it were to consider this case. But questions still remain. For example, how will the U.S. and other countries handle patent applications that cover inventions directly or indirectly resulting from use of an AI system.
The Federal Circuit left open the possibility that inventions made by human beings with the assistance of AI may be patent eligible, but who or what should be the inventor(s}? In other words, what should the law be? Should the law be that the inventor is the person who developed the algorithm, the programmer/trainer, the person ran the program, and/or the owner of the AI system? Should the AI system be an inventor, alone or in combination with one or more humans? After all, if an AI system can infringe a patent, why can’t an AI system be an inventor?
Updating patent laws to keep up with technology is the job of Congress. But how will Congress deal with science fiction of the past that is rapidly becoming reality of the present?