By Michael S. Borella –

From a technical standpoint, everything a computer does involves reading, manipulating, and storing information through microcode instructions that move around 0’s and 1’s. Each operation performed by a processor, such as addition, comparison, and data transfer, can be manually simulated as a series of these instructions. Undergraduate students in computer architecture courses routinely perform such exercises with a pencil and paper to understand how a processor works. In principle, one could simulate any software algorithm this way (albeit with the parallel portions serialized). But completing even modest tasks might take years.

U.S. patent law fails to appreciate these facts, and the consequences are absurd.

The law deems certain categories of inventions as ineligible for patenting, among them those drawn to so-called abstract ideas. Mental processes are a sub-category of abstract ideas. The reasoning behind this exclusion seems straightforward and even reasonable at first blush – if something can be performed purely in the human mind, like adding two small numbers or making a basic decision, it should not be patentable. However, the doctrine has expanded far beyond that.

Under decisions such as CyberSource Corp. v. Retail Decisions, Inc., courts have ruled that a mental process also includes tasks that could be performed by a human using pen and paper. The U.S. Patent and Trademark Office (USPTO) adopted the same stance in its Manual of Patent Examining Procedure (M.P.E.P.) § 2106, asserting that there is no distinction between processes performed entirely in the mind and those that require physical aids. Courts have gone even further, declaring that claims performed by a computer can still be considered mental processes. Cases like Mortgage Grader v. First Choice Loan Services and Intellectual Ventures I v. Symantec treat computer-implemented methods as mental acts simply because they can be theoretically replicated by the human mind.[1]

This principle is a legal fiction. A process that requires a machine executing billions of instructions per second is treated as if it were a mere mental exercise even though it is not and never could be. No amount of coffee will help a person think their way through these complex calculations, much less at a rate that would provide a useful result in a reasonable period of time. And the consequences can be dire for patentees, with claims to legitimately innovative technical inventions being rejected by the USPTO or invalidated by courts because their steps could be simplified to appear analogous to human thoughts.

In In re Killian, the Federal Circuit wrote a small treatise defending this nonsense.  The USPTO’s Patent Trial and Appeal Board (PTAB) continues to wield the mental steps doctrine like a sledgehammer against software applicants, affirming examiners’ patent eligibility rejections at a very high rate.

The USPTO’s own guidance attempts to draw a line between processes that “can practically be performed in the human mind” and those that cannot. Examples deemed mental include generic data analysis and DNA sequence comparison, while examples deemed non-mental include GPS position calculations, network packet inspection, and certain encryption methods. Yet the logic behind these distinctions is paper-thin.

For example, comparing BRCA1 and BRCA2 gene sequences – which the USPTO believes to be mental tasks – involves analyzing tens of thousands of DNA base pairs, an undertaking as unsuited for human performance as inspecting hundreds of network packets, which the USPTO classifies as non-mental. The supposed boundary between these categories has no technical coherence.

The doctrine collapses because it confuses what is theoretically imaginable with what is physically possible. While a human could, given enough time and paper, emulate any computer program, that does not make machine-based computation a mental act. Software execution is a physical process of manipulating electrical states in silicon, not a series of thoughts.

By treating computer-implemented algorithms as abstract mental activities, courts ignore the physical nature of software execution and the integral relationship between code and hardware. Software causes a machine to change states and behaviors when its instructions are performed. Separating the idea of the algorithm from the computer that carries it out distorts both technology and law. This is like confusing a mere recipe with the act of cooking using kitchen utensils and appliances.

This confusion has especially damaging effects on emerging fields like machine learning. Despite the USPTO’s own examples showing that certain machine learning claims are not drawn to mental processes, examiners frequently reject similar claims as ineligible. The reasoning is circular: because a neural network follows mathematical logic, it can be performed mentally, even though no human could realistically simulate millions of weighted matrix operations by hand.

The current application of the mental steps doctrine, coupled with the Supreme Court’s overly-broad abstract idea exclusion, has made software patent eligibility unpredictable and irrational. Courts invoke metaphysical distinctions between “thinking” and “computing” while ignoring the physical, technical character of modern digital systems. As a result, patent examiners are left to apply vague rules that depend more on inconsistent precedent and personal interpretation than on consistent reasoning.

A rational approach would confine the doctrine to its original intent – excluding only those inventions truly meant to be performed mentally by humans. A claim to a method that requires a computer should not be treated as a mental process simply because it can be described in terms of steps that have an attenuated relationship to human thought.

If anyone still insists that computers and human minds are analogous for patent eligibility purposes, I invite them to prove it. Go ahead and train a large language model in your head. Memorize billions of parameters, calculate trillions of matrix multiplications, backpropagate the gradients, and fine-tune it across terabytes of text. You have to do it all mentally, of course, or maybe with a handy pen and paper if you need a physical aid. But you can’t do it because no human can. If you disagree, then go ahead and try and show me the results. I’ll wait but probably not long enough, because by the time you finish the sun will have boiled the oceans and swallowed the Earth.

* This article has been adapted from the author’s earlier article, “The Mental Process Exception to Patent Eligibility is Remarkably Brainless,” Patent Docs, Jan. 9, 2023.

---

[1] To be fair, a claim deemed abstract can still be eligible if it also recites an “inventive concept.” But in practice, this can be quite difficult because once a claim is characterized as directed to an abstract idea, courts often view most or all of its features as part of that abstraction. Further, generic hardware or using routine functions cannot provide an inventive concept. The courts have set a high bar for a claim surviving once it has been damned to the Hell of abstraction.