The U.S. Patent and Trademark Office is making it increasingly difficult to obtain legal protections for inventions related to AI, a field that encompasses autonomous cars, virtual assistants and financial analyses, among countless other uses. The agency, seeing an influx of AI applications, is grappling with how to comply with a law that PTO Director Andrei Iancu has called “anything but clear” concerning what can be patented.
“The U.S. right now is a strong leader in artificial intelligence,” said Kate Gaudry, a patent attorney with Kilpatrick Townsend & Stockton in Washington who analyzed data from LexisNexis PatentAdvisor on pending applications at the agency. She found that, of applications given a primary classification of AI-related, about 90 percent got initial rejection letters saying they were abstract ideas.
“People are going to catch on that they are unlikely to get a patent,” she said.
Patents can be a marketing tool to show that a company is innovative but also are meant as a sort of “keep out” sign to deter competitors from copying the ideas and give their owners exclusive control over who can profit from them. Investors often look to a firm’s patent holdings to help determine how well it could establish a unique niche.
That’s particularly important in a growing industry like artificial intelligence. Venture capitalists and companies are pouring funds into AI in the hopes of riding a new wave of advances in the field to commercial success.
In 2017 and 2018, investments and equity deals in AI-related startups exceeded $32 billion for both years combined, according to research firm CB Insights. Companies like Alphabet Inc.’s Google, Microsoft Corp., Amazon.com Inc., Facebook Inc. and Apple Inc. are hiring more scientists and AI engineers, acquiring companies in the field and racing to release products.
Among International Business Machine Corp.’s record 9,100 patents received in 2018 year were 1,600 related to artificial intelligence.
“AI has been the one that’s been pushing us the hardest,” said Jeff Welser, vice president and lab director of IBM Research - Almaden. The company is working on AI “no matter what the rules are around patenting,” he said.
There’s a host of ethical, regulatory and legal issues related to artificial intelligence, and patents aren’t always at the top of the list. Still, there are growing questions about what can and should be patented.
The patent office, which is integrating AI into its own system to analyze applications, issued new guidance earlier this month outlining how patent-eligibility should be considered. It’s also planning a Jan. 31 hearing on intellectual property policy considerations related to artificial intelligence.
“We are all currently grappling with the eligibility of all sorts of technology, from things like toys that communicate with one another, to computer virus screening; from computer databases, to methods of treating various diseases,” Iancu told a group at the Patent Law & Policy Conference at Georgetown University Law School in November.
It’s common for patent applications to get initial rejections, setting off negotiations between patent lawyers and examiners. For artificial intelligence, though, Gaudry found that nine out of 10 applications were getting an initial rejection saying they weren’t patent eligible -- a difficult argument to overcome. At the same time, the number of applications getting final approval dropped to the 20 percent range.
Among the patent applications getting rejection notices from the U.S. patent office were ones for an artificial memory system, for risk evaluation, for a system to estimate audience interest, and one to predict the chances of an infrastructure failure.
The question goes back to 2014, when the Supreme Court, in a ruling on business methods software, said it wasn’t enough to take an abstract idea and apply it on a general purpose computer. The U.S. Court of Appeals for the Federal Circuit, the nation’s top patent court, has extended that ruling to other areas.
Gaudry links the high rejection rate of AI applications to an August 2016 appeals court ruling that invalidated Electric Power Group LLC’s patents for a system to monitor and analyze data from electric power grids. In another case, the Federal Circuit rejected a patent for a way to encode and decode image data.
Michelle Holoubek, a patent lawyer with Sterne Kessler in Washington, said areas particularly affected include machine learning and bioinformatics, which uses data to find correlations between, for instance, the likelihood of getting a disease or discovering the best course of treatment.
‘It’s Just Math’
“For some of these very valuable patent applications, the examiners are coming back and saying, ‘Yes, but at the end of the day it’s just math,’” Holoubek said. “They say processing data by a computer is just what a computer is used for. It’s not just math -- there’s a lot of processing -- but they say ‘it’s just a computer doing what a computer does.’”
Not everything related to artificial intelligence should qualify as an invention, said Igal Raichelgauz, founder and chief executive officer of Cortica, a Tel Aviv-based firm that uses AI in the areas of autonomous vehicles, radiology and financial transactions.
Many systems are simply software applications that make predictions after parsing data put into the computer, Raichelgauz said.
“If you put in a thousand images of a cat in the machine, the next time it sees a cat, it would know what it is,” he said. “The whole idea of a patent is to enable innovation. Many big companies try to commoditize artificial intelligence.”
Cortica works in the area of deep learning, using artificial neural networks to more closely imitate how the human brain identifies things and makes decisions. Raichelgauz said that’s enabled his company to get more than 200 patents in the world’s patent offices over the past decade, even with the tighter standards.
“The uniqueness of the technology is very important -- the tech has to work, and it has to be unique,” Raichelgauz said. “We want to have patents to protect innovation, but we don’t want to dilute their value.”
That puts added pressure on companies to be clear that what they are creating is worthy of a patent.
“When you look at our economy, our economy is undergirded by data,” Holoubek said. “It’s the software that runs all of our home automation, home appliances, the software that is taking care of our medical industry, our transportation system.”
The patent office remains fully operational during the government shutdown, using prior-year fee collections. The agency hasn’t said when it will run out of money.