Artificial intelligence is transforming drug design — but it could also disrupt intellectual property law. To realize AI’s full promise, the US may have to reconsider its approach to issuing patents.
AI is increasingly being used to identify new chemical compounds, predict their properties, and develop new drugs. It can accelerate the drug development process and may increasingly enable us to develop new drugs we could not develop through traditional approaches. The social and economic benefits could be transformational.
However, this use of AI also raises thorny legal and commercial issues, particularly related to intellectual property inventorship. These issues are highlighted in the recent US Supreme Court decision to deny an appeal in Thaler v. Vidal, a case that hinged on whether an AI platform can qualify as an inventor.
In its decision in Thaler, the Court of Appeals for the Federal Circuit determined that the US Patent Act only allows “natural persons” to be designated as inventors. The patent application under consideration listed an AI platform (Device for the Autonomous Bootstrapping of Unified Sentience, or DABUS) as the only inventor. Because AI platforms are not natural persons, the Court found DABUS could not be listed as the sole inventor.
Here is where the real complications arise. The Patent Act requires patent applications to list all inventors who contributed to the invention in question. The fact that no human was listed as an inventor on the application along with DABUS suggests that no natural person qualified.
In such cases, which may arise with increasing frequency in the era of AI, are the inventions in question simply ineligible for copyright? The question remains unanswered under current US law.
Inventor in the machine
Aside from not being a natural person, an AI platform could meet the criteria for being legally designated as the inventor for certain inventions.
The Federal Circuit has indicated that “conception” is the “touchstone of inventorship,”1 and it has defined conception as the “formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention, as it is hereafter to be applied in practice.”2 Specifically for the invention of a chemical compound, conception consists of both (1) an appreciation of the structure of the chemical compound, and (2) an understanding of a method of making it.3
As generative AI platforms become increasingly capable of designing compounds and methods of making, they would seem to qualify as inventors according to this two-pronged test. Indeed, AI models can already design structures of novel drugs from scratch, and they can predict the properties of those drugs with increasing accuracy.4
But because AI platforms are not natural persons, they cannot qualify as inventors under current US law.
When no natural person qualifies as the inventor
A natural person who used AI as part of the process of invention could still qualify as an inventor if the person’s contribution met the two-pronged test — including in cases when the AI model’s contribution also qualified it as an inventor under the test.
There may be cases, however, when AI platforms meet the two-pronged test but no natural person does. If the AI conceives of a new chemical compound, for example, and no natural person made sufficient contributions to qualify as an inventor, it is unclear who, if anyone, may apply for a patent on the compound.
One might argue that a scientist using AI to develop new drugs is always doing more than merely using AI as an analytical instrument or tool to aid invention. For example, the scientist must provide the AI with the goal of designing a compound with a particular property. But the inventor’s contribution cannot be merely suggesting a general goal and asking another to achieve it.5 The Federal Circuit has indicated that “asking someone to produce something without saying just what it is to be or how to do it is not what the patent law recognizes as inventing.”6
Of course, the scientist would do more than provide a goal at the outset. Instructions may have to be iterated to guide the AI over the course of invention. Eventually the scientist must conduct a biological assay to validate the predicted potency of the molecules proposed by the AI. However, none of these steps — setting a goal,7 providing feedback,8 or validating a proposed solution9 — qualifies as conception under current judicial precedent.
Making way for the future
New technologies often require us to rethink the law, and AI is no different. To harness the potential of AI to invent new chemical compounds and design new drugs, we may have to find ways to encourage the use of AI, even as we guard against the risks it may present.
The function of patent law is to encourage inventors, who often take significant financial risks, by allowing them to protect the intellectual property they create so they may earn a return on their investments of time, effort, and capital. If the incentives are not there, AI may not be used to its full potential, and the world could miss out on the full benefits of using AI to accelerate development of new life-improving and life-saving chemical compounds.
[1] Burroughs Wellcome Co. v. Barr Labs Inc., 40 F.3d 1223, 1227–28, 32 USPQ2d 1915, 1919 (Fed.Cir.1994).
[2] Hybritech Inc. v. Monoclonal Antibodies, Inc. (quoting 1 Robinson on Patents 532 (1890)).
[3] Oka v. Youssefyeh, 849 F.2d 581, 583, 7 USPQ2d 1169, 1171 (Fed.Cir.1988).
[4] Xiangxiang Zeng et al., “Deep generative molecular design reshapes drug discovery,” Cell Reports Medicine 3, 100794 (2022).
[5] Nartron Corp. v. Schukra U.S.A., Inc., No. 08-1363 (Fed. Cir. Mar. 5, 2009)
[6] Morgan v. Hirsch, 728 F.2d 1449, 1452 (Fed. Cir. 1984).
[7] Amgen, Inc. v. Chugai Pharm. Co., 927 F.2d 1200, 1206 (Fed. Cir. 1991).
[8] Morgan v. Hirsch, 728 F.2d 1449, 1452 (Fed. Cir. 1984).
[9] Burroughs Wellcome Co. v. Barr Labs Inc., 40 F.3d 1223, 1227–28, 32 USPQ2d 1915, 1919 (Fed.Cir.1994).
[View source.]
