How generative AI is forcing a rethink on obviousness

Generative AI burst into public consciousness at the end of 2022, and by January 2023, ChatGPT—one of the most popular large language models—amassed over 100 million users, becoming the fastest-growing app ever. Since then, much of the focus for intellectual property lawyers regarding genAI has been on novel questions in copyright law, like whether AI can be the author of a copyrightable work (no) or whether using copyrighted works as training data for LLMs is infringement (to be determined).

On the patent law side, the Federal Circuit has weighed in on whether AI can be an inventor of a patent (no) but hasn’t had the opportunity recently to opine much on AI-related inventions. That changed last month when the Federal Circuit issued its opinion in AliveCor, Inc. v. Apple Inc.

AliveCor’s patents

AliveCor’s three patents at issue relate to measuring and analyzing a body’s data to detect cardiac arrhythmias. Two of the patents describe “a method for monitoring a subject to determine when to record an electrocardiogram (ECG),” and the third describes a wearable device designed to predict the occurrence of arrhythmias.

One of the described embodiments is a smartwatch with an optical sensor that senses changes in blood flow. The sensor transmits the data to a smartphone, which analyzes it for irregularities. When an irregularity occurs, the smartphone notifies the user that an ECG should be recorded. When the user does so, the system uses this data to determine if the user has an arrhythmia.

The patents describe using machine learning to detect arrhythmias from the ECG data, ranging from using simple operations like “ranking,” “classifying,” “labeling,” “predicting,” and/or “clustering” data to complex operations like “random forest, association rule learning, artificial neural network inductive logic programming, [and] support vector machines.” However, the claims themselves only require the use of machine learning at a very high level of generality. For example, one claim is as follows:

The method of claim 1 further comprises determining the presence of said arrhythmia using a machine learning algorithm.

Apple’s arguments

In a trio of IPRs seeking to invalidate the patents, Apple argued that the patent claims relating to machine learning were obvious to a person of ordinary skill in the art at the time of their invention in light of prior art that disclosed confirming the presence of arrhythmias based on ECG data and (separately) that disclosed using machine learning to assess ECG data. The Patent Trial and Appeal Board agreed, invalidating the machine learning claims as obvious. AliveCor appealed to the Federal Circuit.

The Federal Circuit’s opinion

AliveCor did not dispute that the prior art taught how to use machine learning on ECG data, but it argued that they did not teach how to implement that machine learning to detect arrhythmias in particular. This, the Federal Circuit said, was an improper restriction of what the art taught, especially because AliveCor’s claims did not demand a particular type of machine learning algorithm. Perhaps more importantly, AliveCor ignored a fundamental tenet of obviousness the Supreme Court declared nearly two decades ago in KSR International Co. v. Teleflex Inc.: “A person of ordinary skill is also a person of ordinary creativity, not an automaton.” Thus, even if the prior art did not disclose using machine learning to detect arrhythmias, specifically, a person of ordinary skill (and creativity) reading the prior art would have found it obvious to do so.

Obviousness in the age of genAI

This raises the question: What does “ordinary creativity” mean today when ordinary people (let alone ordinarily skilled artisans) can ask ChatGPT to design machine learning algorithms—and much more?

As genAI becomes integrated into the standard toolkit for a growing number of industries, courts will increasingly be faced with this question and have to police the boundary between inspired invention and assisted iteration.

For now, AliveCor reinforces the notion that merely invoking “using machine learning” in a claim—without more detail—is unlikely to save it from obviousness. In the long term, as “ordinary creativity” begins to include the use of these powerful models, and as the process of invention gets outsourced to AI, the bar for what is patentable is likely to increase.

← Back to News