Drugs designed with the help of artificial intelligence (AI) are now entering clinical trials. Jane Wakefield, Artificial intelligence-created medicine to be used on humans for first time, BBC (Jan. 30, 2020). That important milestone has initiated widespread discussion of a brave new world in which computers will "invent" medicines. Such dramatic discussions, however, misapprehend both the U.S. legal framework and the nature of drug discovery: HAL9000 cannot be an inventor under current U.S. patent law. These discussions also ignore a more pressing problem: There are at least three patent law doctrines—obviousness, written description, and enablement—that AI actually can change. This article separates the myth from the reality of how AI will impact life sciences patent law, and offers practical tips to practitioners seeking to protect drug patents against future AI-related challenges.

Computers Cannot Currently Be Inventors as Either a Legal or Practical Matter. Current U.S. patent law assumes that an inventor must be a human being. The U.S. Constitution, for example, grants Congress the power to "promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries." U.S. Const. Art. I, §8, Cl. 8 (emphasis added). The Patent Act also refers repeatedly to "persons". See, e.g., 35 U.S.C. §116. These assumptions are not unique to American law. The University of Surrey recently submitted two patent applications to the European and United Kingdom Patent Offices naming an AI system as an inventor on two product patents. Those applications were rejected on the grounds that European law requires an inventor to be a natural person. EPO refuses DABUS patent applications designating a machine inventor, EPO (Dec. 20, 2019).

Even if U.S. law permitted AI as inventors, current AI systems simply cannot replace the human intervention inherent to the drug discovery process. Inventing a drug is iterative: Researchers run tests to identify lead compounds that change in focus and number as additional data and hypotheses are considered. Even after a lead compound is identified, it is further revised using input from experts in pharmacology, biology, and chemistry to determine formulation, safety, and dosing. There is a reason that drug patents name multiple inventors.

No doubt AI will have a role in helping scientists refine their data. But that role, properly understood, renders AI merely one of the sophisticated tools already commonplace in the drug design and discovery process. Consol. Aluminum v. Foseco Int'l Ltd., No. 82 C 2792, 1988 WL 391250, at *48 (N.D. Ill. Oct. 31, 1988), aff'd, 716 F. Supp. 316 (N.D. Ill. 1989), aff'd, 910 F.2d 804 (Fed. Cir. 1990) (finding that individual who had very little subject matter of patents-in-suit and lacked requisite technical experience "was but a 'pair of hands,' [and] could not be an inventor of the [patents-in-suit]"). By way of example: High-throughput screening is routinely used in the lab as an automated tool to efficiently screen millions of preliminary compounds. Sandra Fox et al., High-Throughput Screening Update on Practices and Successes, 11 J. Biomolecular Screening 864 (2006). Although that technology has helped bring drugs to market, no one has ever considered naming a screening machine as an "inventor". AI should be treated no differently.

The fact that AI will not become a named inventor in the foreseeable future does not mean, however, that practitioners defending life science patents should ignore issues raised by the use of AI in drug discovery. The growth and adoption of AI technology in drug discovery poses potential challenges to drug patents in three different doctrinal areas.

AI Can Change What It Means for Drugs to be Obvious. Under current law, a patent is invalid for obviousness if a "person of ordinary skill in the art" would have been motivated to combine the existing knowledge in the field—known as "the prior art"—to arrive at the claimed invention with a reasonable expectation of success. Acorda Therapeutics v. Roxane Labs., 903 F.3d 1310, 1350 (Fed. Cir. 2018). AI systems are capable of reviewing far more data than a human being could in a lifetime. As the use of AI becomes increasingly routine, the scope of what a "person of ordinary skill in the art" knows may expand to include an alternative, broader AI-assisted standard. As a practical matter, this will mean that patent challengers could more easily draw on references across multiple fields (and computer programs available to drug company innovators) to argue that patented drug inventions are obvious. Indeed, one can imagine a world in which litigants employ dueling algorithms to analyze the knowledge in the field and determine what AI recommendations would have been known and used at the time of the invention.

Patent challengers may also argue that drug discoveries are obvious because the availability of AI renders steps in the drug discovery process "routine". The Federal Circuit has held that routine optimization of variables is not patentable, even if it is time-intensive and requires effort; optimization merely reflects a desire to improve upon what is already known. Pfizer v. Apotex, 480 F.3d 1348, 1367-68 (Fed. Cir. 2007). Patentees may see future obviousness challenges premised on the argument that—although the process of discovering a particular drug was expensive and time-consuming—the use of AI in that process made the discovery optimization rather than an invention.

To prepare against obviousness challenges relating to AI, life science patent lawyers should consider the following recommendations for their pharmaceutical and biotechnology clients:

• Use caution when describing the invention story after the fact. Specifically, take care to explain the human intervention at each applicable step using the assistance of AI.
• Document that human input directed AI activities in drug discovery, including in internal lab notebooks and meeting minutes.
• Avoid policies that prevent researchers from pursuing paths not recommended by AI-assisted technologies. Inventions that deviate from lockstep AI recommendations are likely to fare better against routine optimization arguments and, more importantly, may lead to creative solutions that benefit patients.

AI Will Also Impact the Enablement and Written Description Doctrines. In patent law, a drug genus is sufficiently described in writing when either "a representative number of species" or common "structural features" are disclosed "so that one of skill in the art can visualize or recognize the members of the genus." Ariad Pharm. v. Eli Lilly & Co., 598 F.3d 1336, 1350 (Fed. Cir. 2010). A genus is enabled when one of ordinary skill can make and use the full scope of the claimed genus of chemical compounds without undue experimentation. Idenix Pharm. v. Gilead Scis., 941 F.3d 1149, 1163 (Fed. Cir. 2019) (finding lack of enablement when a person of ordinary skill in the art would not know, without undue experimentation, which compounds would be effective for treating HCV due to the need to synthesize and screen all the possible compounds). Not all compounds within the claimed genus need to work, but the specification must convey enough information to allow one of ordinary skill to understand how to make the compounds and use them for their claimed purpose. Warner Lambert Co. v. Teva Pharmaceuticals USA, 2007 WL 4233015 (D.N.J. 2007) (rejecting accused infringer's contention that the asserted claims lacked an enabling disclosure based on the alleged existence of a small number of inoperative embodiments).

According to some estimates, there are more small molecule drugs in the chemical field than there are atoms in the solar system. Asher Mullard, The Drug-Maker's Guide to the Galaxy, Nature News (Sept. 26, 2017). Given the combination of vast chemical space and ever-increasing computing power, the output of an AI-assisted drug design and discovery program could create extremely large classes of small molecules. Patent lawyers will need to exercise caution in drafting patent claims resulting from AI-assisted efforts; a large or insufficiently characterized genus could be challenged as insufficiently enabled or described. Conversely, the routine availability of powerful AI might make it easier for large geniuses to be disclosed and enabled.

To prepare against written description and enablement challenges relating to AI, life science patent lawyers should consider the following recommendations for their pharmaceutical and biotechnology clients:

• Ensure any claimed genus is small enough, or well-characterized enough, such that a skilled practitioner can "visualize" it, and will know how to use it without "undue experimentation."
• Include activity data or other characterizations for representative members of the class.

Conclusion. Fears that AI will replace human inventors contradict the current U.S. patent law framework and ignore the way that drugs are actually discovered. That is not to say, however, that life science patent practitioners should expect to see all concerns about the impact of AI to fade away. Practitioners would do well to prepare—and prepare their clients—for how AI may change the legal landscape with respect to patent obviousness, written description, and enablement.

Lisa Pensabene is a partner at O'Melveny & Myers and head of the firm's life science litigation practice. Hassen Sayeed is a partner, Carolyn Wall is counsel and James Yi Li is an associate at the firm.