Bilski patent decision: trying to return patents to their technological origins

November 9, 2008

The software industry is abuzz—almost as much as the legal field—with a October 28 court decision that everyone regards as a verdict on business patents, and that some think it will change software patenting as well. But there’s an enormous amount of confusion over the meaning and long-term impact of this decision, known as in re Bilski.

Pamela Jones has issued some wonderfully readable articles on her Groklaw site, one discussing the basic controversy and issues at stake, another laying out the background of the judges, and a third covering attempts to distinguish what is patentable and non-patentable. The highly respected PatentlyO site has a short summary of the core ruling in legal terms, plus some other blogs. Many other legal analyses are trickling out.

The experts are definitely on the case, so what’s the value of this article? I wrote it because I believe that no one has tried to explain how the Bilski decision changes what is patentable (and what is not) using concepts that engineers and computer programmers can understand.

I don’t want to wait for the experts to stop fighting among themselves because that will never happen, at least not till more court cases come up. (The court ruling said so itself.)

I also believe that it’s useful to try to reach enlightenment through a different path. Much of the consternation and confusion can be blamed on the old terminology and precedents that most observers carry around like a ball and chain. We have to get beyond hair-splitting over words such as process, particular machine, transformation, and physical steps—words that the courts have to deal with, because they are part of the legal history of patents, but words that too easily lead to flights of rhetoric that are merely distracting to the layperson.

That’s why I hope, while accepting my limitations as an amateur, that this article can help readers understand what has changed. What I offer here is an inductive exploration based on hypothetical examples. This is a risky approach lawyers don’t tend to take, because law is based on real life and not on speculation. But my approach allows me to make a lot of comparisons in a very small space and to base principles on relatively clean examples.

The core of my concept is that software patents are still allowed, but that a patent on a process or machine has to involve some kind of external agent, with software meeting the bill. I’m using “agent” in its programming sense, not as a legal term.

At the center of this case is an attempt by two patent applicants (Bernard L. Bilski and Rand A. Warsaw) to obtain a patent on a method for manipulating commodity markets. In purely social terms, therefore, the denial of their application seems like a fitting monument to the end of the past decade’s financial orgy, which was based on just such machinations. Few things could be further from the “useful arts” that patents were meant to protect. But this historical observation has no bearing on the legal or technological issues we have to deal with.

Commentators have issued opinions ranging from this marks the end of software patents at one extreme to everybody can just write their applications to work around the ruling on the other. Let’s do a thought experiment and try to slip some patents through the system; then I’ll draw some general observations.

The sections of this article are:

An example: directing traffic through a graph

Let’s use an example to try to distinguish when a process is eligible for a patent. I’ll lay out a series of related processes and give my humbly nonlegal opinion of each. As you can see, the discussion becomes non-linear quickly, because many court cases with different outcomes impinge on each situation.

  1. You develop a new procedure for traversing a graph and determining the fastest way to reach every node on that graph, while allocating a weight to each edge that limits the speed at which edges can be reached.

    Analysis: Everyone, before or after the patent offices started granting software patents, would agree that this is an algorithm or a series of “mental steps,” and therefore unpatentable.

  2. You develop this same procedure (or take it from a published article) but now use it to create a process for delivering water through pipes in a chemical plant or nuclear facility to cool an out-of-control reaction as quickly as possible.

    Analysis: This is clearly a physical (and very useful) process. It would probably be worthy of a patent, even if the algorithm wasn’t yours to start with. By tying the algorithm to a concrete, tangible result, you have gained entry to the palace of patentability.

    But the courts are alert to people gaming the patent system by finding trivial real-world applications of an algorithm. You can’t turn an abstract principle (or “a method of calculating,” as the Supreme Court put it) into a patentable process just by applying it to the flow of water; you must use that principle to change the flow of water in some novel, non-obvious fashion.

    Of course, this distinction is extremely subtle, a spur to recurring arguments and shifting fronts on the patent battlefield, as I’ll show later in comparing in two famous and oft-cited cases, Parker v. Flook and Diamond v. Diehr.

  3. You use this same procedure to create a new routing protocol for Internet routers.

    Analysis: To U.S. courts, Internet routers are just as valid loci for patent activity as chemical plants and nuclear facilities. But there’s a good chance that a patent examiner would turn down this application, even if she approved application number 2—and she’d probably be backed up by a court. Routing is already a well-established practice (after all, what else would a router do?) and the patent system might see that you’re just using an algorithm as input to it.

    This is the kind of application that was supposedly squelched by Flook, where the court refused to allow a patent on a calculation. The patent application used the calculation as input to a chemical process that was already well-known—but the court declared that wasn’t enough to make the calculation patentable.

  4. You use this same procedure to create a new system for disseminating postings automatically across friends’ web pages on a social network.

    Analysis: Here we open the curtain on the conflict between proponents and opponents of software patents. The process is executed entirely in software and has no physical effect except for changes in electrical signals on routers and web sites. In jurisdictions that don’t allow software patents, this would probably not qualify for a patent. In jurisdictions that do allow software patents, this could well qualify, and I think the ruling in Bilski would still allow such a patent. This is because, instead of choosing how to disseminate postings to individuals, the friends on the social network are turning the job over to an external agent: the software implementing the graph algorithm.

  5. You use this same procedure to explain to users of a social network to disseminate news more quickly.

    Analysis: This marks the boundary, in my opinion, where the Bilski ruling makes a difference. The process could well qualify for a patent in the U.S. before the ruling, but I think it would be denied now. This is because the process merely advises people what to do; it does not set up an agent to do it for them.

    Everybody’s asking whether a patent of this type could trivially be turned into a patent of type 4, just by declaring that the process could run on a computer. Courts promise to recognize and reject this simple trick (which, in their ever-endearing legal terminology, they call “insignificant extra-solution activity”) but it happens all the time, and Bilski has the patent lawyers whittling new nibs on their writing pens.

  6. You use a novel form of data analysis to produce a plan for disseminating news more quickly through the social network, and present the plan to users.

    Analysis: This, I’m sorry to say, is probably within striking area of gaining a patent. It doesn’t use an agent, but it adds another patentable feature: a transformation that confers patentability even though only incoming data is transformed. This transformation passes muster because it works on an external set of real-life data (not simply mathematical symbols submitted by a user). It is responding to phenomena in the real world and providing a useful result by presenting users with a plan.

    To my mind, this procedure is like one where a patent was granted by the Supreme Court in their ruling In re Abele and Marshall. You are not simply calculating an algorithm, but creatively selecting inputs and choosing the manner they pass through the calculation. In the words of the court, the process “defines the variables and places the algorithm in a particular relationship to a series of steps in a particular type of process, permitting the algorithm to be applied as a further process step.”

These are all hypothetical cases, and real cases always carry details that can sway a patent examiner and a court one way or another. So while you may argue about my interpretation of the hypothetical cases, there’s not much point in splitting hairs until real courts rule on real cases. What I’m trying to do is put computer engineers in a mode of thinking about the current state of software and business patents.

I think that the concept of an agent, in the context of computer science, may help us understand a central passage from the ruling that the lawyers have been arguing over non-stop:

A claimed process is surely patent-eligible…if:
(1) it is tied to a particular machine or apparatus, or
(2) it transforms a particular article into a different state or thing.

It’s easy to understand this terminology so long as a “machine” is made of valves and cranks. When courts rule that loading a program onto a computer creates a novel “machine,” they enter a Sargasso Sea from which they have not yet freed themselves. Of course, “particular machine or apparatus” doesn’t mean the process must be loaded on an Opteron chip-driven server in my data center running Linux 2.6.25 and Apache 2.0. If we think of either the “machine” or the “transformation” as an agent that does some work for us in the background, we can see how Bilski tries to distinguish useful technological advances (including software) from social and business relationships.

Just as copyright doctrine distinguishes between ideas and the expression of ideas, covering only expression, so patent law distinguishes between ideas and the incorporation of those ideas into automated agents—and once again, only the agents are patentable.

In this regard, it is interesting to read a brief by law professor Kevin Collins laying out why patents on mental processes would restrict thought itself, and an ACLU brief on how patent applications like Bilski would be an impediment to free speech. The dissenting judges in the Bilski case basically tossed off all these objections and said that if something provides a new benefit to somebody, it should get a patent.

But my examples show just how wobbly the boundary remains between the patentable and the unpatentable. Offering a system that people can follow to improve their lives is not patentable, but the moment that system is incorporated into an autonomous agent—some widget or Facebook application—it would be patentable. I don’t quite agree with critics of the decision who claim that the boundary between technical processes and unpatentable business processes is arbitrary. But I think a boundary that can be crossed by amateur script writers is impossible to patrol. I’ll pick up on this problem at the end of this article.

The Bilski ruling doesn’t explicitly rule out business methods (although it’s hard to think of one that could survive the ruling), nor does it make much of a dent in software patents. But it does try to restore patents to the role of promoting technological progress. As I described a year ago, many things performed over previous centuries by mechanical, chemical, or electrical devices are now being performed over software. Opponents of software patents will not, and should not, be mollified by the October 28 Bilski ruling.

About the court decision

Among the dozens of court rulings about software and the hundreds of thousands of patents implicated in them, Bilski doesn’t provide a solid rock where ships can dock; it’s barely an anchor in the choppy seas. There have been too many contradictory rulings over the years for things to shake out anytime soon.

As I mentioned earlier, the Supreme Court heard two very similar cases, Parker v. Flook and Diamond v. Diehr, and came up with opposite conclusions less than three years apart. Flook denied a patent application based on a calculation, whereas Diehr granted one.

To muddy the situation further, Diehr was a 5-4 decision, and it turned not on patent law but on different readings of the patent claims. The majority of 5 saw a difference between what Flook was doing with the physical apparatus and what Diehr was doing, but the minority of 4 said they were doing the same thing and that Diehr had simply written the claims more cleverly to make them seem more integrated with the physical result. Incidentally, this dissent contains a useful summary of the history of software patents and the forces bearing upon them.

Thus, unsatisfying as it is, we have to regard patent law in the light of what the Supreme Court majority thought the Diehr patent contained, regardless of what it really contained.

The protection of “mental steps” from patents will only get more precarious unless someone (probably Congress) steps in to make a strong and unambiguous statement. As life—especially social interactions and global trade— gets more complex, we find ourselves turning some of our activities over to the kind of autonomous agents I discussed earlier. A long tradition exists (notably in the philosophy of Jacque Ellul) of opposition to the creation of such routines and formal systems for mediating between ourselves, other people, and life itself. But we’ll continue to create them because they free us to do more creative things with our time. Every agile programmer knows that.

It’s not just the ease of writing programs that guarantees we’ll involved external agents more and more in our lives. It’s also the development of cheap sensors that will become ubiquitous and that will provide raw data for agents to work on. And of course, the availability of Internet access everywhere, all the time, which allows agents to communicate with both the sources of data and the people who want the agents’ output.

Therefore, because patents have crossed the boundary into software, they will increasingly govern those relationships between us and our world. In fact, even outlawing software patents may not stop the trend, because the agents can be burned right into hardware as the costs of ASICs and FPGAs come down and they become easier to configure. (I owe this insight to Van Lindberg, author of the O’Reilly’ book Intellectual Property and Open Source: A Practical Guide to Protecting Code.) But the hardware requirement will establish some threshold of commitment that a process has to cross—an ability to manufacture and distribute a physical item—before it becomes patentable.

What the courts need to do

The law is a system, like an organism or a social grouping. Systems try to maintain their equilibrium, and they incorporate mechanisms to restore equilibrium when it is disturbed by external phenomena (such as the environment heating up) or internal phenomena (such as when the potassium level drops in a red blood cell).

Patent law got thrown off of it equilibrium by the tremendous wealth promised by software patents, and later the real or imagined wealth promised by business patents. The courts felt they must respond. Each case that changes the interpretation of the law, however, disrupts the law system’s internal equilibrium.

Each case like Bilski needs to be harmonized with dozens of earlier cases that go back decades, with the laws passed by Congress, and with the foundations in the U.S. Constitution (which establishes the patent regime). Although lawyers may say I’m wrong, I also believe the judges have to restore equilibrium with things that are even more fundmental than the Constitution: humanity’s need for a stable society, our search for security, and the sense of fairness deeply seated in primates’ brains.

The public calls on the patent system to serve the stability of society and the security of our most talented inventors by ensuring a predictable environment for technological change. Patenting also invoke our sense of fairness, which dictates that those who make serious contributions to our welfare be rewarded, but that people with no real ideas to offer should not hold others hostage.

Law professors will produce their guidelines for restoring equilibrium, and court cases will confirm or deny each conclusion. Many words that have appeared in various court decisions will be interpreted and reinterpreted. But each patent applicant has to try to extract the essence—even if the essence is illusory—in order to decide whether to submit a patent.

Still, I am convinced that in re Bilski did set a standard, and that it is a wise one. It was also wise to warn that this standard may also fail us as technological developments race ahead. For instance, what happens when technology invades our brains and we start to have combined brain/machine processes?

Perhaps at that time someone will find a method for resolving conflicting interpretations of words that have been used in different ways in legal rulings over the past three hundred years. And hopefully, that method will be in the public domain.

Update: the Supreme Court ruling, June 28, 2010

Today the Supreme Court issued its ruling (PDF) on the Bilski case. They weaken its already wavering impact by explicitly opening the door to business patents and software patents, additionally claiming that processes need not be tied to machines or other physical events in order to be patentable. The majority opinion rules, nevertheless, that Bilski’s patent application was correctly rejected because he was trying to patent abstract ideas such as hedging risk. The resulting ruling is profoundly perplexing, leaving one wondering what business method would not be an abstract idea. In his separate opinion, Judge Stevens brings some much-needed common sense to the question and calls the majority to task for their vacillation.

The majority seemed to me aloof from the battles of the past few decades in the patent field. Such detachment can formally be a good thing, but the majority ends up fussing over the meaning of language with too little concern for the actual complexity of daily activity and the threat that can come from fencing it off through patenting.

Recognizing with perhaps a touch of guilt or concern that allowing business patents has produced a flood of overly broad patents, the majority try to assure the reader that this wouldn’t happen because patent applications could turned down on some other basis such as being obvious or not being novel; the majority feels this sets the bar high enough. I think they’ve lost sight of the question of whether any areas of human activity deserve to be free for all to pursue without looking over their shoulders to see whether someone will claim a piece of their work or just say no.


Andy Oram is an editor at O’Reilly Media. This article represents his views only.

Author’s home page
Other articles in chronological order
Index to other articles