IP Monday Law Blog 

In my last post, I wrote about the importance of asking the right question because the answer you get is only as good as the question you ask. That idea is practically universal, it also shows up in how patent eligibility under 35 U.S.C. §101 is being applied, especially for AI and software.

A recent precedential decision from the United States Patent Trademark Office's (USPTO) Appeals Review Panel, Ex parte Desjardins, is a good example of what happens when the wrong question gets asked and then corrected.

The Wrong Question

For a while now, many §101 rejections, particularly in AI, have effectively started from the same place: Does this involve an algorithm, math, or data processing? If the answer is yes, the rest tends to follow:

1. Abstract idea;
2. Generic computer components;
3. Rejection.

That framing is efficient, it’s also incomplete and it’s what happened in Desjardins. The claims were reduced to a mathematical concept implemented on a generic computer, and the analysis largely stopped there.

Asking Differently

On rehearing, the Appeals Review Panel didn’t ignore the presence of mathematical operations. It acknowledged them but it stepped back and looked at the claims as a whole.

When viewed that way, the system wasn’t just performing calculations. It was doing something more specific; it was training a machine learning model in a way that:

• Reduces storage requirements,
• Reduces system complexity, and
• Preserves prior learning when training on new tasks (addressing catastrophic forgetting).

That is a change in how the system operates and once the question is framed that way, the answer starts to look different.

Familiar Ground

None of this is new. In Enfish, LLC v. Microsoft Corp. (Fed. Cir. 2016), the Federal Circuit made the same point in a different context, recognizing that many advances in computer technology come from improvements to software itself, not just hardware.

We see that play out every day. The shift from physical media to digital platforms didn’t happen just because the hardware changed, it happened because the software did, too. The focus, the court explained, is whether the claims are directed to an improvement in computer functionality, rather than an abstract idea. Desjardins didn’t move that line, it reinforces it.

A More Direct Observation

The panel was also candid about the risk of getting this wrong. It noted that the earlier analysis effectively treated machine learning as just an “algorithm,” with everything else dismissed as “generic computer components,” without really engaging with what the system was doing.

Taken far enough, it creates a broader issue: if everything gets described at a high enough level of generality, it starts to look abstract and when that happens, the analysis stops reflecting the underlying technology.

The panel didn’t frame this as a change in doctrine...it didn’t need to. The point was simpler: the framework only works if it’s applied with some discipline and a willingness to actually understand what the system is doing. In short, be curious!

Will this be a win across the board?

Perhaps...for companies working in AI and software, it at least brings the focus back to how the system actually operates: memory usage, processing behavior, model training, system structure.

For practitioners, it creates a more predictable path: identify the technical improvement, and make sure the claims reflect it. For the USPTO, it shifts the analysis away from high-level labels and back toward the underlying mechanics of the technology.

One Point That Hasn’t Changed

Even here, the claims in Desjardins remained rejected under §103. The panel made a point of that. Eligibility is not supposed to do all the work. Novelty, obviousness, and written description are still where most of the real filtering happens.

Closing Thoughts

If there’s a common thread between the previous post and this one, it’s this: The outcome often turns on the question you start with. In the §101 context, that question isn’t:

• “Is there math?”
• “Is there an algorithm?”

It’s: “What is this system actually doing, and does that change how the technology operates?”

That has been the standard for a while. Desjardins is a reminder that it still is and that how you ask the question, matters just as much as how you answer it.

Want stronger AI and software patents? Connect with our IP team to ensure your claims highlight real technical improvements and withstand §101 scrutiny.