By Michael Borella —

In Liu Cixin's novel The Three Body Problem, the characters create a "computer" from human labor.  Millions of people serve as "bits" and hold up flags to indicate whether they represent 0s or 1s.  These individuals are given instructions to behave like various types of logic gates that would appear in actual digital circuitry.  In combination, these gates are formed into a processor, memory, and other components that would be found on a real computer's motherboard.  Instructions for a human "operating system" are also provided, as well as for a human "application" to execute a solution to the titular problem of Newtonian physics.  The calculations take 16 months.

Liu's novel is speculative but — in theory — such a human computer could exist (for a small-scale, real-world experiment of simulating neural processing with a group of 300 people, see The Stilwell Brain episode of Mind Field).  In reality, though, one might question why anyone would want to undertake such a project.  Even relatively simple calculations, such as evaluating the length of a right triangle's hypotenuse using the Pythagorean Theorem, would likely take several minutes.  We have actual digital computers that can perform such calculations in microseconds.

But the notion of the human computer can be simplified even further.  As anyone who has studied microprocessor design can attest to, everything (and I mean everything) that a computer does involves the reading, manipulation, and storing of 0s or 1s.  Such a processor contains a number of registers and stored microcode that can perform a limited number of arithmetic and logical operations on these bits.  An addition operation might involve, for example, reading one set of bits from memory representing a first number into a first register, reading another set of bits from memory representing a second number into a second register, performing microcode-specified addition of these numbers, and storing the result in a third register.  The result may also be written back to memory.  Even when the processor controls the actions of peripheral components, such as a display screen or network interface unit, it does so by moving bits around.

Thus, a single human can simulate anything a computer can do as a mental exercise or preferably with pen and paper.  Indeed, many undergraduates are subjected to these exercises in order to better understand how computers actually work.  In other words, if you give me a general purpose computer architecture and a problem to solve, I can simulate how that architecture would solve the problem using no more than pen and paper, no matter how complex either the architecture or the problem may be . . . with the caveat that I might not live long enough to provide you with an answer.  Indeed, I might be able to perform a handful of processor instructions per minute as opposed to the trillions of instructions per second that a modern, massively parallel multicore processor can carry out.  If you replace me with another person when I expire, replace that person when they expire, and so on, the sun will likely consume the Earth before many complex calculations can be completed.

Simply put, any software program, algorithm, technique, application, etc. can be — in theory — performed by a human with no more than pen and paper.  And a lot of time.  So much time that mental performance would often be impossible or impractical.  Thus, the contention that there is a meaningful analogy between computer-performance of a task and human performance of a task is just plain silly.

But then comes patent law.

One of the exceptions to patentability is when a claimed invention can be carried out in the human mind – the so-called mental process doctrine.  At first blush, this might not seem like a problem.  If a claim is no more than instructions for a person to think of two relatively small numbers then add them together, or some other task that the vast majority of human beings can perform in their heads, one can understand why such a claim might be excluded from patentability.  After all, how would you go about detecting infringement?

But the mental process doctrine is much more than that.  It incorporates "steps can be performed in the human mind, or by a human using a pen and paper."  CyberSource Corp. v. Retail Decisions, Inc.  Indeed, the USPTO states in M.P.E.P. § 2106 that "courts do not distinguish between mental processes that are performed entirely in the human mind and mental processes that require a human to use a physical aid (e.g., pen and paper or a slide rule) to perform the claim limitation."  Moreover, claims that could be performed in the human mind but are recited as being performed by computer are still "mental processes."  See Mortgage Grader, Inc. v. First Choice Loan Servs. Inc. and Intellectual Ventures I LLC v. Symantec Corp. among many others.

This means that patent law views "mental processes" as encompassing tasks performed by or with the assistance of devices or machines — a legal fiction if there ever was one.

The expansion of mental process doctrine from pure mental processes to processes that are actually performed by or with the assistance of tools or machines was long and slow.  In Gottschalk v. Benson, claims to a method for converting between binary and BCD representations of numbers were found to be ineligible for patenting.  The Supreme Court justified its decision by contending that "mental processes[] and abstract intellectual concepts are not patentable, as they are the basic tools of scientific and technological work" even though one of the claims at issue required a special piece of hardware.  The Court implied that mental processes, even if novel, are analogous to newly discovered laws of nature and thus not patentable in and of themselves.[1]

The claims in Benson were relatively simple to understand and for a human to perform.  If the mental process doctrine stopped there, so would have the controversy.  But a few years later in Parker v. Flook, the Court held a far narrower claim to a more complicated process for calculating alarm thresholds to be ineligible.  In doing so, it recognized but discounted that "the abstract of disclosure makes it clear that the formula is primarily useful for computerized calculations producing automatic adjustments in alarm settings" (emphasis added).

Things were quiet for over three decades until the Court once again opened the door for a more expansive view of mental processes in Bilski v. Kappos and Alice v. CLS Bank.  The Federal Circuit followed this legislating from the bench with a line of cases that led to the current, rather absurd, state of the law.  These include, but certainly are not limited to, claims found ineligible for detecting fraud and/or misuse in a computer environment (FairWarning IP, LLC v. Iatric Sys., Inc.), computerized parsing and comparing of data (Berkheimer v. HP, Inc.), scanning hardcopy documents into a computer program (Content Extraction & Transmission LLC v. Wells Fargo Bank, N.A.), monitoring a power grid (Electric Power Group v. Alstom, S.A.), and translating a description of a logic circuit into a hardware component representation of the logic circuit (Synopsys, Inc. v. Mentor Graphics Corp.).

These cases appear to establish that a claim to a novel and non-obvious computer-implemented algorithm that requires computer implementation to be of any use at all can still be a "mental process."  The Supreme Court's holding in Alice further establishes that computer implementation can often be completely discounted for purposes of patent eligibility.

But software inventions as actually claimed are the combination of the algorithm and the computer implementation, thus requiring the computer hardware.  Courts and USPTO representatives often miss this point or ignore it altogether.  To make things even more confusing, there is another (albeit much shorter) line of cases from the Federal Circuit holding that it is improper to bifurcate a claim in this manner in the eligibility inquiry.  See Gree, Inc. v. Supercell Oy as a recent example.  The Federal Circuit wrote, in relation to a PTAB decision, "[t]o the extent that the Board meant that a proper § 101 analysis may consider some claim limitations only at Alice step one and others only at Alice step two, we do not agree with its reading of Supreme Court precedent."  So on one hand, the Federal Circuit embraces the separation of claim elements in the eligibility analysis, and on the other tells us that doing so is improper.

So where do we stand today?  The Federal Circuit continues to apply mental process doctrine to invalidate patents.  Recently, that Court wrote a small treatise on the viability of mental process doctrine in In re Killian.  The PTAB has been making heavy use of the doctrine to affirm examiner's patent eligibility rejections.  Just in November and December of 2022, 63% of such affirmances relied on the doctrine.

Why are contentions of ineligibility based on alleged mental processes so prevalent?  Perhaps it goes back to what we discussed above.  Any software program, at its core, is driven by sequences of arithmetic and logical operations that could be performed with pen and paper, if not mentally.  Indeed, if the mental process doctrine were applied consistently, either very few software claims would pass muster or all would be eligible.

Despite this bad news for patentees, some software claims are surviving eligibility challenges.  This begs the question of what is so fundamentally different between these claims and their ineligible brethren?

In M.P.E.P. § 2106, the USPTO has attempted to draw a line between claims with limitations that can practically be performed in the human mind with or without the use of physical aids and those that cannot.  The USPTO provides several examples of claims falling into each category.

Those that can be practically performed in the human mind include "collecting information, analyzing it, and displaying certain results of the collection and analysis, where the data analysis steps are recited at a high level of generality," "comparing BRCA sequences and determining the existence of alterations, where the claims cover any way of comparing BRCA sequences," "collecting and comparing known information," and "identifying head shape and applying hair designs."  Contrasting examples of claims that cannot be practically performed in the human mind (according to the USPTO) include "calculating an absolute position of a GPS receiver and an absolute time of reception of satellite signals [using pseudoranges]," "detecting suspicious activity by using network monitors and analyzing network packets," "a specific data encryption method for computer communication involving a several-step manipulation of data," and "a method for rendering a halftone image of a digital image by comparing, pixel by pixel, the digital image against a blue noise mask."

The USPTO provides very little reasoning to support these categorizations, and the reasoning that it does provide is largely conclusory.  Looking to the Supreme Court and Federal Circuit cases on which these examples are based provides scant further insight.

For instance, the USPTO, quoting the Federal Circuit, writes that "the human mind is not equipped to detect suspicious activity by using network monitors and analyzing network packets as recited by the claims."  However, as argued above, a human could perform any computer-implemented task with pen and paper.  And in this case, one could certainly examine the bits of network packets (most such packets have at most 1500 bytes of header and payload) in search of pre-defined patterns that indicate suspicious activity.  Of course, doing so would be slow, cumbersome, and impractical.  But then again, so would "comparing BRCA sequences and determining the existence of alterations."  The BRCA1 gene consists of 81,188 base pairs and the BRCA2 gene contains 10,254 base pairs.  Therefore, performing such comparison of these DNA segments manually would likely be of a similar scale to that of the network packet analysis, resulting in a similarly slow, cumbersome, and impractical procedure.  If the courts or USPTO are trying to draw a line between these cases, they have not done so in a clear or convincing fashion.

This is just one example of how mental process doctrine is inconsistently applied.  A technology area in which we are seeing an uptick in eligibility rejections from the USPTO of late is machine learning.  Despite the USPTO's subject matter eligibility Example 39, which purports to provide a machine learning claim that cannot be practically performed in the human mind, similarly situated claims are often rejected as mental processes.  As they are computer-implemented, the vast majority of these claims can — in theory — be performed with pen and paper, but involve so much complexity that there would be no reason to do so.

In re-reading many of these cases and the M.P.E.P., our conclusion is that the courts and the USPTO are struggling with the doctrine.  As a consequence, decisions and policies are based on a false equivalence between computers and human minds, are riddled with post-hoc rationalizations, and rely too heavily on claim specificity to influence the use of doctrine.[2]  In the USPTO, eligibility rejections for software claims appear to rely at least as heavily on the examiner or art unit assigned as it does on claim language and the underlying invention.[3]

In short, the law as it is currently applied makes little to no sense, is disconnected from real-world concerns, and the deck is unjustly stacked against patentees.

It would be beneficial if the courts and/or the USPTO could develop a reasonable, cogent, and technically sound version of the mental process doctrine — for example, limiting it to claimed inventions that are intended to actually be performed in the human mind.  That way, patentees would have a better idea of what is or is not patent-eligible before investing in preparing and prosecuting applications to these technologies.

[1] The Court's reasoning here does not stand the test of time.  For instance, machine learning algorithms involving artificial neural networks are often viewed as mental processes even though they do not appear in nature.  While loosely modeled on biological structures appearing in brain tissue, artificial neural networks are quite different. 

[2] This should not be surprising given our previous work on analyzing how Federal Circuit eligibility decisions are largely based on whether claims recite a specific, technical improvement over the prior art.

[3] Different examiners and different art units are given a great deal of latitude on how to interpret the eligibility case law and M.P.E.P. § 2106.