New Federal Circuit Decision - Expect Getting AI/Machine Learning Patents Past 101 to Get Tougher

Newsletter

The Federal Circuit recently issued a decision in Recentive Analytics, Inc. v. Fox Corp., invalidating the patent claims at issue as directed to ineligible subject matter under 35 U.S.C. § 101. In what it noted was a case of first impression, the court held that “claims that do no more than apply established methods of machine learning to a new data environment” are not eligible for patent protection.

The four patents at issue fell into two categories – a first group that uses a machine learning model to generate an optimized scheduling of live events, and a second that uses a machine learning model to generate an optimized network map indicating what live events should be broadcast by different television stations. Recentive argued that the machine learning aspects of the claims rendered them eligible because the models generated “customized algorithms” that could be used to automatically create and update the event schedules and network maps.

In rejecting Recentive’s argument, the Federal Circuit first held that the claims “rely on the use of generic machine learning technology in carrying out the claimed methods” and that the machine learning technology being used was “conventional.” In particular, the claims did not require any particular type of machine learning technology but, instead, could be implemented using “any suitable” machine learning technology or technique. Similarly, the court noted that the patents did not require any particular computing machines or processors but, rather, only required “generic” computer components.

Next, the court held that Recentive did not identify any features of the claims that would transform this abstract idea into eligible subject matter. Rather than identifying something “significantly more” than the abstract idea, the court held that Recentive’s purported inventive concept – “using machine learning to dynamically generate optimized maps and schedules based on real-time data and update them based on changing conditions” – merely restated the abstract idea itself.

The decision is arguably not surprising, given the facts of the case. In addition to only claiming a generic machine learning model that could be run on a generic, conventional computer, the court noted that neither Recentive’s claims nor its specification provided any details describing how any purported improvement was accomplished. Recentive also admitted that the claimed processes historically had been performed by human beings. Further, it conceded that machine learning necessarily involves training and iteratively updating the underlying algorithms, undercutting its argument that those algorithms conveyed eligibility on the claims.

This case does not present a blanket ban on patent eligibility for machine learning-based inventions. Indeed, the court held that “[m]achine learning is a burgeoning and increasingly important field and may lead to patent-eligible improvements in technology.” It remains to be seen if different facts would yield different results. For example, would claims with no “human being” analog fare better? Similarly, the court suggests that claims reciting a specific machine learning implementation for a specific purpose or improvements to existing machine learning models may fare better. In this case, however, the courts have long held that claims that recite abstract ideas and then effectively say “apply it” using a computer are not eligible for patent protection, and this decision essentially says that “apply it” using a machine learning model yields the same results.

Practical Takeaways

While it remains to be seen how the USPTO will adopt the holding from Recentive, a few themes seemed to emerge from the court’s holding. First, claims that use machine learning to generate an output that is then used in a downstream task (e.g., controlling a device, improving secondary processes) should still be viewed as integrating abstract idea into practical applications. Second, claims that recite a more specific (though not necessarily unduly narrow) machine learning model may allow for arguing that the hardware running the model is transformed into a special purpose device. Third, this case highlights the continuing theme from the Federal Circuit that the “devil is in the details” – the more technical features that appear in both the specification and the claims, the better.

To join our AI email list, sign up here. If you have additional questions related to this decision reach out to your Quarles attorney or:

Follow Quarles

Subscribe Media Contact
Back to Main Content

We use cookies to provide you with the best user experience on our website and to analyze statistics related to our website. To understand more about how we use cookies, or for instructions to change your preference and browser settings, please see our Privacy Notice. Please note that if you choose to reject cookies, doing so may impair some of our website's functionality.