Artificial intelligence (“AI”) is everywhere, and it continues to be a featured topic at many industry conferences. Sometimes this relatively new topic can feel a bit stale. However, a very interesting AI paper was recently published:  Artificial Intelligence, Scientific Discovery, and Product Innovation by Aidan Toner-Rodgers (November 6, 2024). While several previous papers have shown that AI may be useful in drug discovery (e.g., Merchant 2023), this is perhaps the first paper to provide “real world,” causal evidence that AI improves outcomes for scientific research and development.

The high-level takeaway from the paper is that “AI-assisted researchers discover 44% more materials, resulting in a 39% increase in patent filings and a 17% rise in downstream product innovation. These [discovered] compounds possess more novel chemical structures and lead to more radical inventions.” Id. at Cover Page. Moreover, highlighting the synergy between human skill and AI, the benefit of AI was more pronounced for top researchers at the firm – whose output nearly doubled – while the bottom third of scientists saw little benefit from AI. Id.

This human-AI synergy may be crucial for companies seeking to patent an AI-discovered drug molecule. Indeed, Courts have held that “only a natural person can be an inventor, so AI cannot be.”  Thaler v. Vidal, 43 F.4th 1207, 1213 (Fed. Cir. 2022), cert denied, 143 S. Ct. 1783 (2023). Moreover, according to The Inventorship Guidance for AI-Assisted Inventions issued by the U.S. Patent and Trademark Office, “[t]he patent system is designed to encourage human ingenuity.” See Federal Register, Vol. 89, No. 30, at 10046, available at (https://www.federalregister.gov/documents/2024/02/13/2024-02623/inventorship-guidance-for-ai-assisted-inventions). Accordingly, “inventorship analysis should focus on human contributions, as patents function to incentivize and reward human ingenuity. Patent protection may be sought for inventions for which a natural person provided a significant contribution to the invention[.]” Id. at 10044.

But what constitutes a “significant contribution”?  The closest analogy is likely the scenario in which multiple natural people are listed as inventors on a single patent application. According to the Guidance, Courts must evaluate several factors articulated in Pannu v. Iolab Corp., 155 F.3d 1344, 1351 (Fed. Cir. 1998), including ‘[did the individual] “(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.’’ Federal Register, Vol. 89, No. 30, at 10047. Failing to meet any one of those factors “precludes that person from being named an inventor”. Id.

Further, according to the Guidance:

[A] natural person must have significantly contributed to each claim in a patent application or patent. In the event of a single person using an AI system to create an invention, that single person must make a significant contribution to every claim in the patent or patent application. Inventorship is improper in any patent or patent application that includes a claim in which at least one natural person did not significantly contribute to the claimed invention.

Id. at 10048.

Accordingly, companies should consider documenting human involvement in all phases of its materials discovery process. The goal, of course, would be to show that a human inventor “substantially” contributed to each claim in a patent application. To determine whether a human’s contribution was significant, the Guidance stated that “a person who takes the output of an AI system and makes a significant contribution to the output to create an invention may be a proper inventor.” Id. at 10048. Just as critically, the Guidance provides some examples of human contributions that are likely not considered significant enough to meet the Pannu test, including: (1) recognizing a problem and having a general goal or research plan (which is then presented to an AI tool); (2) reducing an AI-created output to practice; and (3) “overseeing” an AI system that eventually creates an output. Id. at 10048-49.

The full paper is linked for your reference, and the highlights are presented below. If you have questions about this study or AI more generally, please do not hesitate to reach out.

  • Study Design:  The author randomized the use of an AI tool for materials discovery in a cohort of 1,018 scientists at the R&D lab of a large U.S. firm, which specializes in materials science (specifically healthcare, optics, and industrial manufacturing). Toner-Rodgers at p. 1.
  • AI Tool:  The AI tool selected for the study was a set of graph neural networks (GNNs) trained on the “composition and characteristics of existing materials”, which then “generat[ed] ‘recipes’ for novel compounds predicted to possess specified properties.” Id. at 1, 9. The GNN architecture represents materials as multidimensional graphs of atoms and bonds, enabling it to learn physical laws and encode large-scale properties. Id. at 9. From there, scientists evaluated the outputs and synthesized the most promising options. Id. at 1.
  • Materials Discovery:  AI-assisted scientists discovered 44% more materials. These compounds possessed superior properties, revealing that the model also improves quality. This influx of materials leads to a 39% increase in patent filings and, several months later, a 17% rise in product prototypes incorporating the new compounds. Accounting for input costs, the AI tool boosted R&D efficiency by 13-15%. Id. at 1. The effects on materials discovery and patenting emerge after 5-6 months, while the impact on product innovation lags by more than a year. Id. at 16. No data was presented regarding commercialization of a discovered compound.
  • Quality: AI increased the quality of R&D as opposed to merely building out currently understood, low value outputs. For atomic properties, the tool increases average quality by 13%. Id. at 18. AI led to statistically significant improvements in average quality (9%) and the proportion of high-quality materials. Id.
  • Novelty:  The AI tool increased the “novelty of discoveries, leading to more creative patents and more innovative products.”  Id. at 20. In the absence of AI, scientists focus primarily on improvements to existing products, with only 13% of prototypes representing new lines. But this AI tool caused an increase in the treatment group (22%), engendering a shift toward more radical innovation. Id. at 18-20.
  • Synergy with Humans:  The high-quality scientists saw the most benefit from AI. The bottom third of researchers saw minimal gains but, the output of top-decile scientists increased by 81%, which suggests that AI and human expertise are complements in the innovation production function. Id. at 22. Moreover, top scientists leverage their expertise to identify promising AI suggestions, enabling them to investigate the most viable candidates first. Id at 21-22. Moreover, some highly skilled scientists can observe certain features of the materials design problem not captured by the AI tool. Id. at 33. In contrast, others waste significant resources investigating false positives. Id. at 26.
  • Impact on Human Labor:  AI dramatically changes the discovery process. The tool automates a majority of “idea generation” tasks, reallocating scientists to the new task of evaluating model-suggested candidate compounds. Id. at 38. In the absence of AI, researchers devote nearly half their time to conceptualizing potential materials. This falls to less than 16% after the tool’s introduction. Meanwhile, time spent assessing candidate materials increases by 74%. Id. at 2. Time spent in the experimentation phase also rises. Id. at 27.

In sum, AI is labor-replacing when it comes to identifying new materials, but labor-augmenting in the broader process because of the need for human judgment in evaluating potential compounds. Id. at 28. The author noted a slump in workplace satisfaction among scientists with 82% reporting reduced satisfaction with their work. Id. at 36. Further research is needed on this potential “brain drain”, as this paper highlights the benefits of combining the top human minds at a company with AI.