In Recentive Analytics vs Fox, the Federal Circuit determined that absent more, machine learning methods are now considered “generic.”
Patentability is always a moving target. Patentability criteria change because formerly novel ideas quickly become dismissed as commonplace. These days, some early 20th-century inventions now look like, “Let’s do this old process using a motor!” Similarly, some late 20th-century inventions now look like, “Let’s do this old process using a computer processor!” After a while, patent examiners and courts become jaded and essentially turn off because they have seen this all before.
About “Alice”, “Abstract”, 35 USC 101, and “Patent Eligibility”
Before the 2014 “Alice” ruling, the “we’ve seen this all before” argument was essentially one of lack of novelty or obviousness. Everyone now recognizes that motors and processors can be used for many things, so establishing that your idea is not obvious becomes ever harder.
Patent law went sideways in the 2014 “Alice decision,” however, when the Supreme Court introduced a new way to reject patents as “abstract” (without defining the term). This decision sent the US patent legal system off on a never-ending snipe hunt that continues today. At this point, the rules are long and inconsistent, with the USPTO going one way and the courts often going other ways.
Other countries have generally avoided this “side quest“, and the US should probably take a hint. But here we are, and if you want a US patent, we must deal with some Medieval logic.
Although everyone is bored with “generic” motors and processors, “improved” motors and processors are another matter. So, assuming that an examiner or judge has deemed other parts of your invention “abstract,” you can avoid rejection if you show that it uses an “improved” component. Machine learning methods were considered an improved processor for a while, so this strategy worked until recently. But no longer.
In Recentive Analytics vs. Fox, as of April 18, 2025, the courts are now officially telling us that just saying that you are using “machine learning” systems or methods no longer cuts it.
About the Recentive Analytics vs. Fox case:
Recentive Analytics used machine learning methods to optimize television broadcast schedules. They received four patents and sued a division of Fox Television for infringement. Fox appealed, arguing that the patents were invalid under “Alice” as “Abstract.” The lower courts agreed, and this went up to the Federal Circuit.
Since SCOTUS has pretty much now washed its hands of this area (they did such a good job in 2014, after all), the Federal Circuit is now the highest court hearing this sort of thing. So this decision looks like it will stick.
The Recentive Analytics attorneys conceded that workers had been manually making network schedules for a long time. They also acknowledged that they were not claiming any new machine learning methods. The court interpreted this as an admission that their machine-learning methods were conventional. Further, the court concluded that since Recentive did not teach any improved machine learning algorithms or methods, Recentive was merely applying standard machine learning to a new field.
The Federal Circuit determined that as of 2025, this doesn’t cut it. That is, a patent that claims “do a previously known activity” with “generic machine learning” is no longer valid. To get a valid patent, you must teach a new activity, an improved machine learning system, or ideally both.
The good news, however, is that the Federal Circuit is not shutting down the field. Instead, they concluded on an encouraging note:
Machine learning is a burgeoning and increasingly important field and may lead to patent-eligible improvements in technology. Today, we hold only that patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under § 101.”