“We’re getting closer and closer to where…we want to use these tools, and society wants us to use these tools, because it makes the process faster, it makes it more effective and makes better drugs…so we’re putting the U.S. at a disadvantage if we’re going to throw things out on conception when other countries don’t even ask those questions.” – Corey Salsberg
The U.S. Patent and Trademark Office (USPTO) held its East Coast Listening Session on AI Inventorship today, featuring both USPTO staff and patent stakeholder speakers contemplating possible approaches to patenting in a world in which generative artificial intelligence (AI) almost seems to have snuck up on everyone. While many speakers cautioned against moving too quickly to change the rules for AI-generated inventions, others warned that doing nothing could result in chaos for the USPTO and grave economic and innovation losses for the country.
The listening session comes on the heels of the U.S. Supreme Court’s denial yesterday of a petition by Dr. Stephen Thaler asking the Court to consider the question: “Does the Patent Act categorically restrict the statutory term ‘inventor’ to human beings alone?” With the High Court dodging the question, and Congress being notoriously slow to act on changes to the law to accommodate new technologies, the ball is squarely in the USPTO’s court when it comes to any imminent guidance on the topic.
USPTO Director Kathi Vidal, who said recently that this will be her “year of action,” told attendees today in a brief pre-recorded message that the Office is experiencing a “surge” in more than half of its technology centers of patent applications that utilize AI. She also said the Office feels its February Request for Comments on AI inventorship “gets to the heart of the matter.” Comments are due May 15.
Salsberg: Clarify Conception
The listening session ran from 10:30am to 3:00pm and featured a range of speakers. While a representative of IBM said things should stay the same for now, Corey Salsberg of Novartis explained that AI is being used in the biopharmaceutical space in ways that already test the bounds of the current system.
“Using AI machine learning, we can cut the time to screen and analyze compounds for desirable properties down to minutes or hours instead of months or years, or quickly screen thousands of images to determine if a candidate is having the desired effect,” Salsberg said.
But even thornier is the use of an AI known as JAEGER in the field of generative chemistry, he explained. Researchers trained JAEGER on 21,000 molecules in the Novartis library that had been previously tested for anti-malarial properties and the AI came up with 282 “realistic, entirely new virtual molecules with anti-malarial properties,” Salsberg said. Humans then used other AI tools to test the two most promising ones and ultimately proved them to be effective. This demonstrates the need for the right policies to be in place now, Salsberg added, but the question has thus far been framed too narrowly to focus only on when AI can be an inventor. “The better question for now is whether and how a human’s use of AI impacts the human’s status as inventor,” according to Salsberg.
This analysis implicates the unique-to-the-U.S. requirement of “conception” in the invention process that several of the speakers raised. Conception has to do with the formation in the mind of the inventor of “the definite and permanent idea of the complete and operative invention.” But that standard doesn’t mesh with the use of AI to comb through millions of compounds and narrow them down to a few promising structures, for example, or to use AI to propose entirely new molecules that didn’t exist before. “When these activities and connections happen in the mind, we would certainly call it part of conception, so the key question is whether conception is still satisfied when the actual connections, correlations or designs are happening ‘in silico,’ outside of the human mind, in an AI,” said Salsberg.
For its part, Novartis believes the present law is flexible enough to account for this, “but we also think there’s sufficient uncertainty and ambiguity that the Office should consider clarifying its guidance,” Salsberg added. For example, the MPEP has an entire section devoted to conception being in the mind of the inventor, even though an older version said an inventor can consider external sources in the invention process. And another section includes case law on the conception of chemical compounds that could be interpreted to suggest that a human directing an AI to define or design a compound isn’t enough to make them an inventor or co-inventor. Salsberg summed up:
“The main point is we’re getting closer and closer to where to be more efficient in our R&D, we want to use these tools and society wants us to use these tools, because it makes the process faster, it makes it more effective and makes better drugs, particularly when other countries do not have a conception requirement the way we do, so we’re putting the U.S. at a disadvantage if we’re going to throw things out on conception when other countries don’t even ask those questions.”
Ultimately, while Salsberg said he would probably opt for eliminating the conception requirement, the Office doesn’t need to go that far yet. “We just need to construe it properly to account for the advances in technology that allow us to do things outside the mind that used to happen inside the mind.”
Feldman: Fools Rush In
Professor Robin Feldman of UC College of the Law, San Francisco, on the other hand, said “only fools rush in” on issues like these. Feldman said that human inventors are incentivized by having their names on patents, while AI aren’t. “It’s neither socially desirability nor coherent to list AI in patents,” Feldman said. Plus, there are still too many questions to answer, such as whether AI constitutes a POSITA or when an AI infringes. Feldman recommended potentially using the model of pharmaceutical data rights for AI, with shorter periods of protection. But for now, AI is still a tool, she said, and we should “go slowly.”
Crouch: Act Swiftly
Professor Dennis Crouch of Patently O also weighed in, and called for the USPTO to come out with clear guidance, quickly, due to the “unprecedented pace” of AI. There are already some cases where, if the AI was human, it would be recognized as sole inventor, so the issue is squarely before us, Crouch said. Generative AI will reduce the cost of inventing and is creating valuable technological improvements, so “it is imperative the USPTO acts swiftly,” he added.
Crouch’s solution is a special category of patent for inventions without any direct human contribution. This could include a reduced patent term or limited claims, for example. Crouch likened generative AI to the situation in Amgen v. Sanofi, which is awaiting a decision from the Supreme Court. In Amgen, the “humanized mouse” is equivalent to the AI generating innovative outputs in that the mouse was given a prompt in the form of a PCSK9 injection and, in response, created antibodies that it stored in its spleen for later recovery.
But analogies aside, it’s time for guidance, Crouch said:
“Patent attorneys are facing substantial uncertainty regarding the proper course of action. Innovators, their clients, are seeking to protect valuable inventions but there is a lack of clear guidance and it’s creating ethical dilemmas for patent attorneys. The USPTO should promptly offer guidance stating that patent applications can appropriately list the human contributor to conception as the sole inventor even where AI or other tool provided key elements of the discovery.”
Crouch also warned the USPTO not to fall into the trap the Copyright Office recently did with its guidance on AI, which essentially excludes AI-generated works from copyright protection. In fact, Crouch called for a revival of the idea of forming a U.S. Intellectual Property Office so that IP policymakers can consider the problems holistically, across patents, copyrights and trade secrets. “We don’t want to miss out on a new medical treatment because of some rule we have in place,” Crouch said. “In my mind, it’s imperative the USPTO act swiftly to provide guidance on how to handle generative AI’s role in the invention process.”
The USPTO will hold a West Coast Listening Session on AI Inventorship on May 8.