In an April 18 ruling in a dispute involving Fox Corp., the appeals court upheld a judge's finding that Recentive Analytics Inc. patents on using machine learning to generate event schedules and arrange TV broadcasts are not patent eligible because they "do no more than apply established methods of machine learning to a new data environment."
The court found that the patents are directed only to abstract ideas like creating schedules that predate computers, and do not include a specific improvement in computer technology that would make them patent eligible. The opinion noted, though, that "machine learning is a burgeoning and increasingly important field and may lead to patent-eligible improvements in technology."
The case relies on reasoning from the court's many decisions finding patents ineligible when they only use computers to add efficiency to existing processes, so the ruling "confirms the intuition that a lot of us had, that if you're just claiming a generic machine learning model and applying it to data, that is certainly suspect," said Michael Borella of McDonnell Boehnen Hulbert & Berghoff LLP.
The challenge now will be to find ways to write patents that describe technological improvements in AI that could be eligible under Section 101 of the Patent Act, but the court did not provide insight into how to do that, said Borella, co-chair of the firm's artificial intelligence practice group.
If you can describe a technological improvement in AI and describe why it's an improvement, "the court seems to be opening the door to saying, well, that might be something that will get you over the 101 hurdle," he said. "But we just do not have any examples from the Federal Circuit of what this would look like in practice."
There are several potential routes that applicants seeking patents tied to machine learning or AI could consider to increase their odds of securing a patent, though until a court weighs in, there's no way to be sure if those will be successful.
"Unfortunately, 101 is the area of patent law where it's all amorphous," said Sid Kapoor of Pierson Ferdinand LLP, noting that even individual examiners at the patent office can have their own take on what is sufficient for patent eligibility.
"That makes it a little more challenging in terms of, what should you put in there that would just move the needle just enough," he said.
The Federal Circuit's decision emphasized that Recentive's patents stated that the company's inventions relied on "any suitable machine learning technology," which the court said amounted to a new use for existing technology that cannot be patented.
While the company said its invention represented a patent-eligible technological improvement because the patents require the machine learning model to be trained and adjusted with data, the Federal Circuit said doing that is "incident to the very nature of machine learning."
Kapoor said that applicants might be better served by instead focusing machine learning patents on more specific technical improvements, like new methods of training the AI tool, or framing the invention in terms of how the invention achieves specific technical solutions, like greater computational efficiency or data accuracy.
"I wouldn't say these are all tested and proven solutions, they're sort of suggestions," he said. "I wouldn't say, 'Just take this, plug it in, and you get over 101.' I'm just saying, 'Hey, these are all the things you may want to consider.'"
While the decision suggested that technological improvements in the machine learning algorithm itself could be patent eligible, "it's not certain that's going to work, because algorithms are also a category of unpatentable subject matter under Federal Circuit precedent," said Mark Liang of O'Melveny & Myers LLP.
"The remark by the court at the end of how to get around its own decision I think is still a little up in the air, is the take home," he said. "I think it's just going to have to be tested as more cases come up, and I'm sure that will happen."
Recentive's attorney, Robert Frederickson of Goodwin Procter LLP, raised that issue at oral arguments in the case, when he explained that the company specifically did not seek a patent on an improved form of machine learning.
"This wasn't an invention of a new machine-learning technique, because that would fall into another one of this court's Section 101 traps," he said. "If the claim is improving the mathematical algorithm or making machine learning better, then we're claiming the natural law, the mathematical algorithm itself."
To improve the odds of surviving an eligibility challenge, what those applying for patents involving machine learning "may have to do in the future is literally put down the details of their machine learning. It has to be new and different from something that was used before," Liang said.
That could include specifying the data being used to train the model, or the particular formulas or parameters that are involved, but such an approach comes with risks.
"The problem is that might end up making your claim very narrow and effectively not enforceable against most of your competitors," Borella of McDonnell Boehnen said.
Another approach could be for patent applicants to describe ways that machine learning applies to more tangible or technical environments, such as using AI to improve the operation of machines on an assembly line.
"So in that case, I might not even claim it as a machine-learning invention. That's kind of tangential. I would just claim the machines themselves," Borella said, adding that the Federal Circuit decision didn't address that scenario, "but I think that that's still a valid approach."
Attorneys said that despite the patent eligibility challenges artificial intelligence inventions face, the decision is unlikely to discourage companies from seeking patents on them, amid surging interest in the technology. There has not been much litigation on AI patents to date, but that is expected to change as more patents are issued and the technology becomes more widely used.
Borella emphasized that the Federal Circuit pointedly did not say that AI inventions are not patent eligible, and stated that patents that provide sufficient detail about technological improvements could survive.
"This is one of those cases where I'm just a little nervous that a lower court or someone at the [patent office] is going to look at this and then just take an even stricter, narrower view of what's patent eligible," he said. "And that's not really warranted from this decision, in my opinion at least."
The appeal court's ruling is "just sort of a speed bump" that "we just have to deal with as patent drafters and advise clients accordingly," Kapoor of Pierson Ferdinand said.
"So don't be disheartened. I wouldn't say these patents are dead, that's not the case," he said. "I think we just need to be more clever about it, and be more strategic in advising clients."
The case is Recentive Analytics Inc. v. Fox Corp. et al., case number 23-2437, in the U.S. Court of Appeals for the Federal Circuit.
--Editing by Kelly Duncan and Emily Kokoll.
For a reprint of this article, please contact reprints@law360.com.
