Lawmakers are beginning to consider ways in which artificial intelligence systems are involved in inventive processes like discovery of new drugs, and how the use of such technologies affects what is, and is not, patentable.
The issue is fast becoming critical because U.S. law allows patents to be issued only to human inventors, whereas technologists and scientists are demonstrating that in some cases an artificial intelligence system is able to invent a new molecule or design a product that previously eluded humans.
“Make no mistake, AI presents novel questions across a wide range of areas of intellectual property policy,” Sen. Chris Coons, D-Del., chairman of the Senate Judiciary Subcommittee on Intellectual Property, said at a recent hearing.
Coons said the panel plans to hold a series of hearings to ensure that innovation generated by artificial intelligence systems remain in the United States, “instead of incentivizing innovators to turn to other countries with more favorable laws to protect their AI-generated inventions.”
At the June 7 hearing, Coons cited findings last month by scientists at the Massachusetts Institute of Technology who used an artificial intelligence system to identify a new antibiotic. It was the first new antibiotic to be discovered since the 1980s, after scientists had abandoned the quest because of the time and expense of traditional processes, Coons said.
The MIT antibiotic kills a bacteria called Acinetobacter baumannii that causes pneumonia and meningitis among hospital patients and was resistant to the current family of antibiotics. The new drug was identified from “a library of nearly 7,000 drug compounds using a machine-learning model that they trained to evaluate whether a chemical compound will inhibit the growth” of the bacteria that causes the infections, the university said in a statement.
Under current laws it’s not yet clear whether human scientists or AI systems can be considered inventors of the new molecule, and it’s a question the courts haven’t resolved either. In April, the Supreme Court declined to take up a case by computer scientist Stephen Thaler, who had sought a U.S. patent for a beverage holder and an emergency light beacon designed by his artificial intelligence system without any human input.
Thaler’s application was shot down by the U.S. Patent and Trademark Office because only humans can be inventors, and the decision was upheld by lower courts, prompting Thaler to seek intervention from the high court.
South Africa, however, offered a patent for the products with the artificial intelligence system listed as the inventor, according to the trade publication IPWatchdog. That’s the situation lawmakers fear, with innovation moving offshore.
AI vs. human inventor
Several experts who testified before the Senate Judiciary panel said U.S. law should recognize inventions made with the help of artificial intelligence systems, with humans being listed as inventors, but they differed on whether the law should be changed to allow an artificial intelligence system to be considered an inventor.
An artificial intelligence system may find a new molecule or design a product, but it’s not yet capable of recognizing and appreciating the invention, said Corey Salsberg, head of IP affairs at pharmaceutical giant Novartis.
“The key point I think between invention and noninvention is an area of the law called its recognition and appreciation that is part and parcel of conception,” Salsberg said. Only a conscious human who can recognize a new invention has that capability, Salsberg said. “We don’t believe that, at this point in time, even generative AI is recognizing anything that it is doing,” he said.
Generative AI refers to ChatGPT and other systems that ingest a huge quantity of existing text and images and are capable of generating new text and images when prompted.
Although Novartis uses a generative AI tool called JAEGER to assist in generating new virtual molecules for treating malaria, “what it’s doing is not equivalent to human inventorship,” Salsberg said.
The tool had to be trained with about 20,000 examples of compounds and then told about three compounds that have anti-malarial properties “to teach it how to design a molecule,” Salsberg said.
Ryan Abbott, a professor at the University of Surrey in the U.K., and a partner at the law firm Brown Neri Smith & Khan LLP, told lawmakers that U.S. law must be amended to allow AI-generated inventions to be patentable.
Abbott, who represented Thaler in the case that the Supreme Court declined to hear, said allowing such inventions to be patented “will provide encouragement to the use and development of inventive AI that will generate more socially valuable inventions.”
Allowing such AI inventions to be patented would encourage “the owners of these systems to disclose inventions rather than keep them as trade secrets, and it will encourage the investment needed to commercialize inventions,” Abbott said.
Although human computer scientists may be involved in developing an artificial intelligence system, such an individual “who merely develops an AI with problem-solving capabilities without specifically conceiving of a particular output, should not qualify as an inventor under U.S. law,” Abbott said.
Computer scientists and programmers building an AI system are only analogous to “a human inventor’s teachers, mentors and even parents,” he said. And such teachers and mentors “do not qualify as inventors on their patents — at least, not without directly contributing to the conception of a specific invention.”
Between a purely human invention unaided by artificial intelligence and an invention generated by AI lies “a spectrum of inventive behavior” where a human and an AI system collaborate to varying degrees, said Laura Sheridan, head of patent policy at Google.
Google believes that only humans should be named as inventors and AI-generated inventions should not be issued a patent, Sheridan said.
“In our view, current industry uses of AI are well within the zone where humans are properly named as the inventors and where AI is leveraged as a tool in the invention process,” Sheridan said. “We expect to remain in this zone for some time.”
The post Congress ponders whether AI should have the power of the patent appeared first on Roll Call.
