USPTO Leadership Reins in PTAB on AI Eligibility 

What Happened 

The USPTO’s decision in Ex parte Desjardins is good news for applicants in the AI and software spaces. The Appeals Review Panel (ARP), including newly appointed Director John Squires, overturned a Patent Trial and Appeal Board (PTAB) decision that had raised a new § 101 eligibility rejection.  

The PTAB had treated the claims as directed to a mathematical algorithm without integration into a practical application, but the ARP found they reflected concrete improvements in how an AI model functions. This matters beyond one case: the ARP emphasized that eligibility should not be denied merely because claims involve mathematical concepts, algorithms, or machine learning. 

The Technology at Issue 

The application, assigned to DeepMind, addresses “catastrophic forgetting” in machine learning. In simple terms, the invention improves the ability of an AI model to learn new tasks without erasing prior knowledge. 

The claims cover how the technical solution is achieved. Claim 1 recites features that include: “adjust the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task.”  The specification describes that the claimed technique can allow a machine learning model to “effectively learn new tasks in succession whilst protecting knowledge about previous tasks,” and mentions the benefits of reduced storage capacity and system complexity.  

Key Takeaways from the Decision 

• Integration matters. Even if claims include mathematical concepts, they may still be eligible if the claim as a whole applies those concepts in a way that improves how a computer system functions. The ARP found that even if the claims recite an abstract idea, the claims are not directed to an abstract idea because they integrate the concept into a practical application that improves how the model operates. 

• Improving the system itself is enough. The ARP relied on the key Federal Circuit decision in Enfish to reaffirm that software-based improvements to computer technology can be patent-eligible. The ARP confirmed the Federal Circuit’s indication that the eligibility determination should turn on whether claims are “directed to an improvement to computer functionality.” Here, the ARP found that the improvements described in the specification and reflected in the claim language, including how to facilitate model training, were indeed improvements to computer functionality. 

• Signal to examiners. The ARP sent a clear signal to examiners: eligibility should not be denied simply because claims involve mathematical concepts or machine learning, such as an algorithm that can be performed on generic computer components. As the decision put it, “Examiners and panels should not evaluate claims at such a high level of generality.” Instead, the focus should be on whether the invention solves a technical problem and reflects that solution in the claim language.  

• Traditional statutes remain central. Patentability questions should primarily turn on §§ 102 (novelty), 103 (obviousness), and 112 (written description/enablement). Eligibility should not be used as a shortcut rejection when claims show a real technical advance. 

Practical Guidance for Patent Drafters 

• Draft with technical improvements front and center. Describe clearly how the invention makes a computer system or model perform better. Include specific, concrete examples in the specification showing how the claimed invention improves computer functionality or solves a technical problem. 

• Tie benefits to the claims. Make sure the features that drive the improvement are expressed both in the claim language and in the specification. Avoid purely functional or result-oriented language. 

• Expect § 102/103 scrutiny. Clearing § 101 does not end the process; examiners will still focus on novelty and obviousness. 

Bottom Line 

The Desjardins decision reinforces a consistent theme: AI and software innovations can clear the § 101 hurdle when they show real, technical improvements in how computer systems operate. Examples include reducing memory use, improving processing speed, increasing accuracy, or enabling new system capabilities. 

Contacts  

If you would like more information on the issues discussed in this article, please contact Hugo Biermann at HBiermann@slwip.com.