December 13, 2022
In the Federal Circuit decision in Amgen Inc. v. Sanofi, while invalidating two of Amgen’s antibody patents, the CAFC made the following statement regarding the enablement requirement of 35 U.S.C. §112.
What emerges from our case law is that the enablement inquiry for claims that include functional requirements can be particularly focused on the breadth of those requirements, especially where predictability and guidance fall short. In particular, it is important to consider the quantity of experimentation that would be required to make and use, not only the limited number of embodiments that the patent discloses, but also the full scope of the claim.
On November 4, 2022, the U.S. Supreme Court granted Amgen’s petition for certiorari regarding the Federal Circuit’s enablement review. The second question of the petition, which was taken up by the Court, asks the following:
Whether enablement is governed by the statutory requirement that the specification teach those skilled in the art to "make and use" the claimed invention, 35 U.S.C. §112, or whether it must instead enable those skilled in the art "to reach the full scope of claimed embodiments" without undue experimentation—i.e., to cumulatively identify and make all or nearly all embodiments of the invention without substantial time and effort.
While this case concerns pharmaceutical and biotechnology patents, we know from Alice that when the Court weighs in on the interpretation of a key patent statute, the ramifications will be felt across many technology areas. Artificial Intelligence is no exception.
We have written multiple blog posts on challenges of the written description and enablement requirement with regard to AI. In our previous post of June 28, 2021, we explored the challenges of broad AI claims running into problems with 35 U.S.C. §112(a). In particular, in the AI realm, recognition of a broad number of objects in a particular field is the ultimate goal of a many AI-based inventions. It should also be noted that functional limitations are often contained in software-related patents such as AI patents.
Further, hawse have written in-depth on describing and enabling AI inventions. In our post of April 30, 2021 we point out that “[t]he trained model is at the center of applied machine learning inventions and presents the greatest challenges to patent disclosure because the inner workings of the model are not fully understood or easily explained.” The ambiguity or the black-box nature of the trained model in an AI invention can sometimes put a lot of weight on what is understood to those skilled in the art.
Consider the following scenario: an object-recognition system for a self-driving car has a patent claim which includes the following hypothetical functional limitation: “…send a signal to the car to take action X in response to detecting a non-vehicle moving object within Y feet of the car.” The specification for this application may explicitly take into account certain non-vehicle objects that may be recognized in video images, such as pedestrians, pedestrians on bicycles, and others. The specification may describe how the object recognition model is trained to recognize such objects. However, since the full scope of the claim covers detection of any non-vehicle moving object, the Amgen decision from the Court could significantly affect whether this type of claim meets the requirements under 35 U.S.C. §112(a).
In other words, (i) will it be enough that the specification discloses to those skilled in the art how to "make and use" the claimed invention so that the AI model can be trained to detect any potential non-vehicle object different than the ones considered by the inventors? Or (ii) would the specification have to cumulatively identify and make all or nearly all possible recognition scenarios without substantial "'time and effort”?
While the difference between the impact of these two standards is murky, it is clear that the Court will take a position somewhere along the spectrum that will influence the manner in which AI patent applications will be drafted and that will cause some change in the scrutiny applied to existing AI patents.
For these reasons, patent practitioners in all technological fields, and especially AI inventions, should pay close attention to the ruling (and the language) that comes down from the Court when this case is decided.