LLM (Large Language Model) and AI (artificial intelligence) inventions are patentable in the US. But since AI is no longer new, their “Alice immunity” has worn off.
A brief history of software patentability in the US
To understand LLM and AI systems from the patentability standpoint, let’s quickly review some history. Until the early 1980s, it was unclear if any software inventions were patentable. Then from about the mid-1990s to 2014, the patent system was relatively permissive. During this era, it was often possible to describe a known process and then teach how to computerize it. Software patents were often granted as long as they weren’t obvious or a repeat of previously known methods.
Then two things happened. One is that as the art advanced, the concept of “implementing a known process using a microprocessor” became evident to everyone, even patent examiners and judges. The other is that an Alice corporation inventor obtained some computer-assisted financial patents. They then tried to assert their inventions against CLS, a large bank with deep pockets. Money talks. The bank took their case to the US Supreme Court (SCOTUS) and won. Unfortunately the Supreme Court nearly wrecked the patent system in the process. Here, I refer to the infamous “Alice” patent eligibility decision.
One “Alice” decision result: “Generic computers” are “bad.”
“The mere recitation of a generic computer cannot transform a patent-ineligible abstract idea into a patent-eligible invention.” [Emphasis added]
https://supreme.justia.com/cases/federal/us/573/208/
We will now skip over about ten years of intense litigation and cut to the present. What does the USPTO do these days? Among other things, under MPEP 2106, pure software inventions that run on “generic computers” and lack “something more” (such as an attached gadget) are vulnerable to 35 USC 101 “Alice” rejections. The term “generic computer” in a USPTO examiner’s office action, or a judge’s decision, often signals an upcoming rejection.
LLM and AI systems (were formerly) not considered to be generic computers
However, from the standpoint of AI and LLM systems (such as ChatGPT, Bard, LLaMA, and related systems), the situation used to be easier. The patent legal system previously did not consider these to be “generic computers.” Indeed, from personal experience, AI and LLM claim limitations were formerly helpful.
Patent law tends to evolve with time, so there was never any guarantee that AI and LLM methods would continue to enjoy “Alice immunity.” However, this technology was new, and the US legal system started off as quite open. Before 2024 to 2025, if you had an AI or LLM invention, the USPTO would tend to treat it like any other patent. The good news/bad news is that this is still the case, but AI and LLM no longer have “Alice immunity.”
As an example of how patent law evolves, note that on July 17, 2024, the USPTO updated its AI and LLM patent eligibility guidance. More recent court cases will be covered in a later blog article.