In Ex parte Desjardins (Google’s DeepMind Technologies), the USPTO’s Appeals Review Panel (ARP) vacated a §101 rejection of claims directed to a machine-learning training method, offering timely guidance for AI-related inventions.
The application claimed a system that trains a model on new tasks while preserving performance on earlier ones, addressing the problem of “catastrophic forgetting.” A PTAB panel introduced a new §101 rejection, finding the claims directed to abstract mathematical concepts. The ARP disagreed, holding that the claims, viewed as a whole, improve the functioning of the machine-learning model itself by reducing storage needs and improving performance—making them patent-eligible under Alice Step 2A.
The decision cautions examiners against treating all AI inventions as abstract ideas and emphasizes evaluating technical improvements to computer functionality or another technology, consistent with the USPTO’s August 2024 Subject Matter Eligibility (SME) Memorandum.
That memo reminded examiners to:
- Analyze claims as a whole, without oversimplifying;
- Distinguish between claims that recite an abstract idea and those that merely involve one; and
- Avoid §101 rejections when eligibility is a “close call.”
Takeaways for innovators:
- AI and machine-learning claims remain eligible when they improve system performance or efficiency.
- Clearly identify technical benefits, such as reduced storage, faster processing, or better accuracy, in the specification and claims.
- Expect examiners to focus more on §§ 102, 103, and 112, using §101 only where ineligibility is clear.
This decision underscores the USPTO’s commitment to a balanced, predictable approach to subject-matter eligibility, particularly in fast-moving fields like AI.
