New APPS

No, the quote isn’t a new marketing slogan for OpenAI.  I’m actually referring to a budding issue in patent law.  The Patent Act says that “whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title (35 U.S.C. §101).  Although this is very broad, Supreme Court precedent says that it exempts abstract ideas, laws of nature, and natural phenomena. 

As I argued in my IP book (from which I’m lifting some of the discussion below) the rise of the information economy has made understanding these exemptions quite difficult.  In an industrial setting, all of these patentable things tended to occur in certain objects that could then be claimed as patentable.  As Dan Burk notes, “products, at least to the extent that they constitute objects, are inherent in the concept of process …. Making and using entail some type of object: some thing is made, and some thing is used.  In classic industrial setting, the substrates of the process were fairly apparent, and extant in what is now §101; machines and materials visibly interacted as inputs generating outputs” (527).

With the rise of “immaterial” goods and a post-Fordist economy, however, it is increasingly difficult to point to discrete things either at the level of product or process, and the ability to characterize immaterial goods informatically suggests that they could be understood as either thing or process.  Burk argues that the Supreme Court cases on §101 are therefore more about drawing judicial limits on what patents can cover.  As he puts it, “excluding conceptual inventions from patent eligibility pushes exclusivity further downstream to the stage of finished products, requiring narrower claims on concrete implementations, rather than allowing conceptual patents early in the development of a technology” (535).  Still, the devil lies in the details of how to make this work.

In Mayo v. Prometheus (2012), the court unanimously ruled that a drug dosing procedure failed §101 patentability.  In the case the claimed patents “purport to apply natural laws describing the relationships between the concentration in the blood of certain thiopurine metabolites and the likelihood that the drug dosage will be ineffective or induce harmful side-effects” (1294).  In other words, different patients respond differently to the drugs in question, and the patents assisted doctors in titrating the drugs based on metabolite levels present in a patient’s blood after an initial dose.  The Court ruled that the claimed processes did little more than state the laws of nature and direct doctors to apply them; it then expressed a worry about the excessive enclosure of nature and the potential effects that would have on innovation.  Justice Breyer writes for the Court:

“Beyond picking out the relevant audience, namely those who administer doses of thiopurine drugs, the claim simply tells doctors to: (1) measure (somehow) the current level of the relevant metabolite, (2) use particular (unpatentable) laws of nature (which the claim sets forth) to calculate the current toxicity/inefficacy limits, and (3) reconsider the drug dosage in light of the law. These instructions add nothing specific to the laws of nature other than what is well-understood, routine, conventional activity, previously engaged in by those in the field. And since they are steps that must be taken in order to apply the laws in question, the effect is simply to tell doctors to apply the law somehow when treating their patients” (1299-1300).

The 2014 Alice v. CLS Bank decision (again unanimously) affirmed this line of reasoning, arguing that “the mere recitation of a generic computer cannot transform a patent-ineligible abstract idea into a patent-eligible invention.”  Alice is important, both because it has been highly influential and because it shows that the problem of §101 patent eligibility is a larger problem for the information economy than just biotechnology.  In Alice, the Court invalidated as an abstract idea a scheme for mitigating “settlement risk, i.e., the risk that only one party to an agreed-upon financial exchange will satisfy its obligation.  In particular, the patent claims are designed to facilitate the exchange of financial obligations between two parties by using a computer system as a third-party intermediary.”  The court reasoned that the claims were directed to “the abstract idea of intermediated settlement” and that they “merely require generic computer implementation,” thereby “fail[ing] to transform that abstract idea into a patent-eligible invention.”  They thus failed to meet the standards for subject-matter eligibility.  Citing Mayo’s rejection of “apply it,” the Court here rejects stating an abstract idea while adding the words “apply it with a computer” (2350).  The Court here searched for and failed to find an “inventive concept” that is “sufficient to assure that the patent in practice amounts to significantly more than a patent upon the [ineligible concept] itself” (2355). This is all motivated by a concern with avoiding pre-emption, the enclosure of abstract ideas, which are a “basic tool” of scientific work.  In other words, the assumption is that monopoly in such cases is bad for innovation, whether monopolies can be efficient is not a concern the court emphasizes.

You can see where this is going.  What if I use AI to do something better or faster?  The case in question is Recentive Analytics v Fox (Fed. Cir, 2025).  As Judge Dyk writes for the Court:

“The patents purport to solve problems confronting the entertainment industry and television broadcasters: how to optimize the scheduling of live events and how to optimize “network maps,” which determine the programs or content displayed by a broadcaster’s channels within certain geographic markets at particular times” (1-2).

There’s two groups of them – network mapping ones and machine learning training ones.

“Claim 1 of the ’367 patent is representative of the Machine Learning Training patents and recites a method containing: (i) a collecting step (receiving event parameters and target features); (ii) an iterative training step for the machine learning model (identifying relationships within the data); (iii) an output step (generating an optimized schedule); and (iv) an updating step (detecting changes to the data inputs and iteratively generating new, further optimized schedules)” (3).

The specification indicates that previous event data become the training data, and target variables like attendance are given to the machine learning model, which then uses “any suitable machine learning technique” (5) to figure out how to optimize for that variable.  The network mapping procedures were analogous.

This is how a lot of ML and AI systems work, so the precedential implications here are enormous.  Is this patentable?  During the litigation, Recentive conceded that network mapping was not a new thing, and they also didn’t claim patents over the ML techniques themselves (after all, the claim recited a list of suitable, off the shelf techniques).  Rather, they “claim[ed] the application of the machine learning technique to the specific context[s]” of event scheduling and network map creation” (8). That is, “Recentive characterized its patents as introducing ‘the application of machine learning models to the unsophisticated, and equally niche, prior art field of generating network maps for broadcasting live events and live event schedules.’” (8).

The district court didn’t buy it, and neither did the Federal Circuit.  Following Mayo and Alice, the court applies a two-step test:

“Under Alice, courts perform a two-step analysis to determine patent eligibility under § 101. “First, we determine whether the claims at issue are directed to one of those patent-ineligible concepts.” If the claims are directed to a patent-ineligible concept, we assess the “elements of each claim both individually and ‘as an ordered combination’” to determine whether they possess an “inventive concept” that is “sufficient to ensure that the patent in practice amounts to significantly more than a patent upon the [ineligible concept] itself.”” (10, internal citations omitted).

In software patent cases, the question focuses on whether “‘the specific asserted improvement in computer capabilities . . . or, instead, on a process that qualifies as an abstract idea for which computers are invoked merely as a tool.” (10, internal citation omitted).  Given the setup, one can see how this isn’t going to go well for Recentive.  They don’t claim to improve the computer.  They claim to improve the “unsophisticated” process of the application of the abstract idea of network mapping by being able to do it faster, or with more data.  Humans worked iteratively with the same data before, and lots of caselaw says that just because the computer is efficient doesn’t turn a patent-ineligible idea into a patent eligible one.  “Do it on a computer” isn’t enough to get a patent.

The “inventive concept” part doesn’t go any better for Recentive:

“Recentive claims that the inventive concept in its patents is “using machine learning to dynamically generate optimized maps and schedules based on real-time data and update them based on changing conditions.” As the district court correctly recognized, this is no more than claiming the abstract idea itself. Such a position plainly fails to identify anything in the claims that would “‘transform’ the claimed abstract idea into a patent-eligible application.” (16-17, internal citations omitted).

Again, the computer might be a lot more efficient at updating the maps (“dynamically generate … real-time data” and so on), but that doesn’t mean that it’s doing anything inventively different from the people who tried to keep the maps updated by hand.

So “do it with ML” or (presumably) “do it with AI” isn’t patent-eligible.  This seems right to me, and not the occasion for reforming §101 to expand patent eligibility, as some patent owners perennially desire.  Briefly, it would be anti-innovative to give the first person to think to run something on a generic AI model the right to lock everybody else out of that idea, no matter how much it would help the first-mover’s bottom line.  It would be like if the first person to think to use a power saw to cut table legs in their furniture shop could prevent everybody else from doing so.

Recentive has appealed to the Supreme Court, so this may not be over.