A Patent Eligibility Paradox: Why Broader Claims Can Win When Narrower Claims Fail
"I was told there'd be no math"
A recent United States Patent and Trademark Office (USPTO) memo on Subject Matter Eligibility highlights one of the most counterintuitive aspects of patenting artificial intelligence: a claim that seems broader can be deemed patent-eligible, while a more specific, seemingly narrower version of it is rejected as an abstract idea. This "weirdness," as some practitioners call it, isn't a flaw in logic but a critical distinction based on the legal principle of preemption.
The memo uses two examples to illustrate this very point. One claim passes the eligibility test at the first step, while the other immediately gets flagged as an abstract idea.
The claim limitation "training the neural network in a first stage using the first training set" of example 39 does not recite a judicial exception. ... Contrast this with the limitation "training, by the computer, the ANN based on the input data and a selected training algorithm to generate a trained ANN, wherein the selected training algorithm includes a backpropagation algorithm and a gradient descent algorithm" of claim 2 of example 47. This limitation requires specific mathematical calculations ... and therefore recites a judicial exception, namely an abstract idea1.
At first glance, this is baffling. The second claim adds specific details—the exact algorithms being used. How can adding detail make a claim less patentable?
The Fear of Preemption: Don't Patent the Building Blocks
The core reason for this paradox is the legal doctrine against preemption. Courts have long held that fundamental principles—like laws of nature, natural phenomena, and abstract ideas (including mathematical formulas)—are the basic tools of scientific and technological work. Granting a patent on one of these tools would "preempt" its use and improperly inhibit, rather than promote, innovation.
Think of it this way:
You can patent a novel type of engine that uses the principles of thermodynamics in a new way.
You cannot patent the laws of thermodynamics themselves.
When a patent claim explicitly recites a specific mathematical formula by name, like "a backpropagation algorithm," the USPTO sees this as an attempt to claim the fundamental tool itself. The focus shifts from the practical machine or process to the underlying math.
The claim "training the neural network," on the other hand, is viewed as a practical, functional step. It uses math, but it claims the application, not the formula.
A Tale of Two Claims
The USPTO's own analysis of its examples elucidate this distinction:
Example 39 (Eligible): The claim recites steps like "training the neural network in a first stage using the first training set." The USPTO's analysis concludes this is not a judicial exception because, "While some of the limitations may be based on mathematical concepts, the mathematical concepts are not recited in the claims." The claim is focused on the functional process of training a model for facial detection. It avoids the eligibility test's second step altogether.
Example 47, Claim 2 (Ineligible): This claim recites "training...wherein the selected training algorithm includes a backpropagation algorithm and a gradient descent algorithm." Here, the USPTO analysis finds the claim does recite a judicial exception. Why? Because it explicitly names the mathematical concepts. This forces the claim into a much tougher analysis, where it ultimately fails because the other elements are seen as generic computer functions that merely "apply" the abstract idea without adding an inventive concept. The claim recites the generic steps of receiving data, training, detecting anomalies, and outputting data, which are not enough to save it.
It's Not a Death Sentence, It's a Different Path
Crucially, reciting a mathematical concept isn't an automatic loss. It just pushes the claim onto a more difficult path.
This is demonstrated by Claim 3 in the same Example 47. Like the failed Claim 2, it explicitly recites the backpropagation and gradient descent algorithms. So why is it eligible?
Because it integrates those algorithms into a specific, concrete, and technical application that improves computer functionality. The claim goes on to recite specific steps for improving network security:
Detecting malicious network packets
Detecting the source address in real time
Dropping the malicious packets in real time
Blocking future traffic from the source
The USPTO found that these additional steps were not generic but described an improvement in the technical field of network intrusion detection. This specific implementation saved the claim, showing that the abstract idea was integrated into a practical—and patentable—application.
Strategic Drafting for AI Inventions
This apparent “paradox” provides some insight for patent professionals as to how the USPTO examiners may look at claims. Here are some strategies to consider:
Focus on Function in Independent Claims: Draft broad claims that describe the practical, functional application of the AI. Frame the invention as a technological solution to a technological problem.
Avoid Naming the Math (If You Can): Keep the names of specific algorithms out of the broadest claims to avoid triggering the abstract idea analysis in the first place. Math will not be seen as a technical detail.
Use the Specification: The detailed description is the proper place to disclose the specific algorithms used. This is essential for satisfying enablement and written description requirements.
Integrate and Improve: If you must recite a mathematical concept in a claim, ensure the claim as a whole recites a specific, inventive application that improves a technology, rather than merely stating to apply the math on a generic computer.
Ultimately, while framing an invention as a "practical application" may be sufficient to navigate the examination process at the USPTO, a more robust and forward-looking strategy is to demonstrate that the claims recite a technical solution to a technical problem.
This approach does more than satisfy the language of the MPEP; it builds a stronger foundation for the patent's long-term validity.
An argument rooted in a specific, tangible technological improvement tends to resonate more powerfully with district courts and the Federal Circuit, creating a more durable and defensible intellectual property asset that is better equipped to survive the rigors of litigation.
Even then, this potential paradox may lead to a roll of the dice.
Disclaimer: This is provided for informational purposes only and does not constitute legal or financial advice. To the extent there are any opinions in this article, they are the author’s alone and do not represent the beliefs of his firm or clients. The strategies expressed are purely speculation based on publicly available information. The information expressed is subject to change at any time and should be checked for completeness, accuracy and current applicability. For advice, consult a suitably licensed attorney and/or patent professional.