Drawing the line: The Rise of theInformation Processing Patent

Ben Klemens∗

January 16, 2007

AbstractNow is the right time to revisit an old but still contentious

question: should software and business methods be patent-able? On the legal front, the debate hinges over whether aclaimed invention consisting of an information processing stepplus a trivial physical step can be claimed as a bona fide phys-ical innovation. The Supreme Court ruled three times that itcan not, but the Court of Appeals for the Federal Circuit ruledthat the combination must be considered as a whole, meaningthat it is to be treated like any other physical invention. Onthe economic front, there is no self-contained information pro-cessing industry: every business in every field uses softwareand business methods. For a multitude of reasons, patents areill-suited for such massively decentralized industries. There-fore, this paper recommends a return to the distinction thatinventions consisting of information processing plus a trivialphysical step be barred from patentability.

Contents

I The legal perspective 31 Exceptions to patentability . . . . . . . . . . . . . . . . 42 Congressional intent . . . . . . . . . . . . . . . . . . . 73 The Church-Turing Thesis . . . . . . . . . . . . . . . . 84 The Supreme Court’s trilogy . . . . . . . . . . . . . . . 105 The Freeman-Walter-Abele test . . . . . . . . . . . . . 146 CAFC rulings . . . . . . . . . . . . . . . . . . . . . . . 157 The logical conclusion . . . . . . . . . . . . . . . . . . 18

∗Thanks to Tahmineh Maloney, who provided many hours’ worth of commen-tary, critique, suggestions, and notes. Thanks also to Abigail Rudman for her deftlibrarianship.

1

II The economic perspective 211 Empirical evidence . . . . . . . . . . . . . . . . . . . . 212 Patent efficiency in massive industries . . . . . . . . . . 243 Patent trolls . . . . . . . . . . . . . . . . . . . . . . . . 284 Contributory infringement . . . . . . . . . . . . . . . . 325 Is it all just obviousness? . . . . . . . . . . . . . . . . . 336 Disclosure . . . . . . . . . . . . . . . . . . . . . . . . . 34

IIIConclusion 361 Politics . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

The maelstrom over software patents and business methods hasraged for a few decades now. It is an example of a more generalquestion: where should one draw the line between what is patentableand what is not?

The Labcorp v. Metabolite case came before the Supreme Courtregarding this very question. Unfortunately, certiorari was withdrawndue to technicalities, but the dissent to the withdrawal made it clearthat the question of what is patentable subject matter is a pressingone. Justice Breyer (with Justices Stevens and Souter) explains:

Patent law seeks to avoid the dangers of overprotectionjust as surely as it seeks to avoid the diminished incentiveto invent that underprotection can threaten. One way inwhich patent law seeks to sail between these opposing andrisky shoals is through rules that bring certain types ofinvention and discovery within the scope of patentabilitywhile excluding others.1

In the case in question, the dissent indicated that the current linebetween the patentable and the unpatentable was not set in the rightplace to correctly strike this balance. Justice Breyer stated that theCourt of Appeals for the Federal Circuit (CAFC), applying logic muchlike that discussed below, was allowing a patent to stand that hindersrather than promotes progress.

With this in mind, the paper here looks at the many limits topatentable subject matter that have been proposed. The weakest, andeffectively the status quo, is that laws of nature may not be patented,but any application thereof may be. For example, an equation isunpatentable, but a machine that evaluates that equation is. In thecase of Labcorp, the claim was not for a correlation but for the actof correlating. The fact that such a simple rewording can bypass the

1548 U. S. 04-607 (2006) (Justice Breyer dissenting), at 3

2

nonpatentability of laws of nature indicates that this line excludesfrom patentability only the purest of natural concepts.

This line, which made software and business methods possible, hascreated a number of unresolved problems in the real world, as docu-mented in Section II. The set of businesses that might independentlyinvent the claims of a patent on a pharmaceutical include any busi-ness with a drug synthesis factory (a few dozen); the set of businessesthat might independently invent the claims of a patent on a web sitedesign include any business with a web site (about a hundred million).With a hundred million potential independent inventors, someone isalmost certain to have written software that matches the claims of anygiven patent—but independent invention is not a valid defense againstpatent infringement claims. Thus, liability risk and opportunistic law-suits are almost certain to appear—and in fact, they have. They arenot due to details of patent procedure, but the fundamental fact thatevery business must process information and use business methods.

Another line, effectively encapsulated by a common reading ofthe Freeman-Walter-Abele test, is that bona fide physical inventionsshould be patentable, but information processing algorithms with atrivial physical step, such as anything that could be written to a com-puter file or a piece of paper, should not be. In his 1981 ruling inDiamond v. Diehr, Justice Rehnquist stated as much, acknowledgingthat a rubber press with a computational step was a bona fide physicaldevice, but at the same time “. . . [I]nsignificant postsolution activitywill not transform an unpatentable principle into a patentable pro-cess.”2 Section II will show that this distinction is a much better fitto the economy at large, because the problems with applying patentsto pseudo-industries such as the set of computer or business methodusers evaporate, as do a number of political problems.

But the CAFC rejected this line, indicating that it is impossibleto bifurcate a patent into an algorithm and insignificant postsolutionactivity—in all cases, one must take the entire patent as a whole.If some steps are physical, then the whole is physical, and if somesteps are novel and useful, then the whole is. This ruling was primar-ily intended to allow software—abstract instructions loaded onto astock computer—to be patentable, but the logic immediately appliesto other technically physical processes such as many business methodsor the act of correlating.

Knight3 argues that storylines—instructions written down for ac-tors—fall into the class of patentable information designs exactly asdoes software—instructions written down for a computer. It seemsobvious that patent law should not cover storylines, as evidenced by

2450 U.S. 175 (1981).3Andrew F Knight, “A Potentially New IP: Storyline Patents”, The Journal

of the Patent and Trademark Office Society, Vol. 86, No. 11.

3

the fact that the US Patent and Trademark Office (USPTO) has nevergranted a storyline patent, yet the question of codifying that intuitionremains: without the requirement of an innovative physical step, howwould one draw a legal line that allows the patenting of informationdesigns such as software but does not allow the patenting of infor-mation designs such as storylines? Or has the USPTO been remissin not granting patents on storylines, novel musical compositions, orinnovative dance routines?

Another possible line between the patentable and unpatentable isa technological arts test. To the consternation of tax lawyers, methodsof using tax loopholes have also been patented. Would a technologicalarts test allow patents on methods to price shares but exclude patentsto exploit tax loopholes, and is such a test optimal policy?

Section I of this paper will look at the legal context of patentablesubject matter, documenting how patents on information designs wentfrom being consistently rejected by the USPTO to the current regime,where approximately 1 in 6 of the USPTO’s granted patents are ininformation designs. Section II will look at the economic studies thathave measured the effects of over a decade of software patentability,and consider the reasons why no such study has found any positivebenefit.

I The legal perspective

U.S. Patent law is founded on the U.S. Constitution, Article I, Sec-tion 8, Clause 8: “The Congress shall have power to . . . promote theprogress of Science and useful Arts, by securing for limited times toauthors and inventors the exclusive right to their respective writingsand discoveries.”

The first patent law was enacted on April 10, 1790. Thomas Jef-ferson, Secretary of State and gadget inventor, penned the descriptionof what types of invention may be patented. Since then, Congress haschanged only one word of Jefferson’s text, replacing art with process.4

His minimally modified text is now 35 U.S.C. §101:

Whoever invents or discovers any new and useful process,machine, manufacture, or composition of matter, or anynew and useful improvement thereof, may obtain a patenttherefor, subject to the conditions and requirements of thistitle.

These two items, the Constitution and 35 U.S.C. §101, are theentire text of the statutes regarding patentable subject matter. They

4See Diamond v. Chakrabarty, 447 U.S. 303 (1980).

4

already pose two challenges to patents on software and business meth-ods. First, do they “promote the progress of Science and useful Arts?”This question will be discussed in detail in Section II. Second, are theycovered in the list of patentable subject matter described in §101?

1 Exceptions to patentability

Looking at the statute, it may seem obvious that a computing processor business method is a process, and therefore covered. However,Congress, the courts, and even Jefferson himself understood that thereare many restrictions to what should be patented beyond the sparsetext of §101.

For example, the Supreme Court’s ruling in Diamond v. Diehr5

stated: “This Court has undoubtedly recognized limits to 101 andevery discovery is not embraced within the statutory terms. Excludedfrom such patent protection are laws of nature, natural phenomena,and abstract ideas.” Included in the natural law exception are suchthings as “the heat of the sun, electricity, or the qualities of metals,”6

while the abstract idea exception stipulates that one can not patent“a novel and useful mathematical formula.”7

Jefferson wrote on the abstract idea exception in a letter to acolleague:8

It would be curious then, if an idea, the fugitive fermen-tation of an individual brain, could, of natural right, beclaimed in exclusive and stable property. . . . Its peculiarcharacter, too, is that no one possesses the less, becauseevery other possesses the whole of it. He who receives anidea from me, receives instruction himself without lessen-ing mine; as he who lights his taper at mine, receives lightwithout darkening me. That ideas should freely spreadfrom one to another over the globe, for the moral and mu-tual instruction of man, and improvement of his condition,seems to have been peculiarly and benevolently designedby nature, when she made them, like fire, expansible overall space, without lessening their density in any point, andlike the air in which we breathe, move, and have our phys-ical being, incapable of confinement or exclusive appropri-ation.

These exceptions have been repeatedly reaffirmed by the courts.9

5450 U. S. 175, 185 (1981).6Funk Brothers Seed Co. v. Kalo Inoculant Co., 333 U. S. 127, 130 (1948).7Parker v. Flook, 437 U. S. 584, 585 (1978).8Thomas Jefferson, letter to Isaac MacPherson (August 13, 1813).9 “An idea of itself is not patentable.” Rubber-Tip Pencil Co. v. Howard (87

5

Next in the list of exceptions is the mathematical algorithm excep-tion. From In re Walter:10 “. . . a principle of nature or a scientifictruth (including any mathematical algorithm which expresses such aprinciple or truth) is not the kind of discovery which the patent lawswere designed to protect.”

This ruling indicates that mathematical algorithms express princi-ples of nature, and thus are not man-made technologies, but whetherthis is so is a vehemently open question, which philosophers sincePlato have been unable to resolve.

On one side of the debate are the mathematical realists.11 Plato(taking a page from Pythagoras) saw reality as a mere reflection ofpure mathematical objects, meaning that Nature is a subset of Math-ematics, not the other way around. Paul Erdos, one of the founders ofmodern network theory, was famous for stating that his results weresimply copied from “The Book,” a divine catalog of all mathematicalresults.

On the other side are the formalists, who see mathematics as themanipulation of human-developed symbols, that happen to sometimesreflect reality. This belief first gained a following in the mid-1700s,when Gauss and others introduced the concept of non-Euclidian ge-ometry. In the Elements, Euclid asserted five axioms, and then de-rived other statements therefrom using logical manipulations. Non-Euclidian mathematicians changed one of Euclid’s axioms, and thenderived a new set of results that have equal internal consistency.Mathematics suddenly went from a unified characterization of oneworld to being a description of many, some of which did not seemto correspond to reality.12 Mathematics was now a game of symbolmanipulation. In the words of Ludwig Kronecker: “The integers werecreated by God; all else is the work of man.” The natural philosopherDonald S Chisum13 typifies the modern formalist viewpoint, asserting

U.S. 498 (1874)). “. . . [A] scientific truth, or the mathematical expression of it, isnot a patentable invention. . . ” Mackay Radio & Tel. Co. v. Radio Corp. of Am.(306 U.S. 86, 94 (1939)). “He who discovers a hitherto unknown phenomenon ofnature has no claim to a monopoly of it which the law recognizes.” Funk Bros.Seed Co. v. Kalo Inoculant Co. (333 U.S. 127, 130 (1948)). “A principle, inthe abstract, is a fundamental truth; an original cause; a motive; these cannotbe patented, as no one can claim in either of them an exclusive right.” Le Royv. Tatham, (14 How. 55 U. S. 156, 175 (1953)). “[T]he discovery of a novel anduseful mathematical formula may not be patented.” Gottschalk v. Benson (409U.S. 63 (1972)).

10618 F.2d 758 (1980).11Ruben Hersh, 1997: What is Mathematics, Really?, Oxford University Press.12The non-Euclidians were partially vindicated when Einstein showed that un-

der some circumstances, there are non-Euclidian geometries that do a better jobof describing space than Euclidian geometry does.

13Donald S Chisum, “The future of Software protection: The Patentability ofAlgorithms,” 47 U Pitt Law Rev, 959, 961 (1986).

6

that “a mathematical or other algorithm is neither a phenomenon ofnature nor an abstract concept. [A mathematical] algorithm is verymuch a construction of the human mind. One cannot perceive an algo-rithm in nature. The algorithm does not describe natural phenomena(or natural relationships).”

It would be futile for a brief law review article to attempt to resolvea question that has been debated for centuries, if not millennia. Butwhen evaluating the proposed limits of patentable subject matter, thereader should note that there is such ambiguity. If a proposed limitrequires for its application the resolution of an unresolvable philosoph-ical debate, then it is probably not very practical.

Nonetheless, as a matter of simple fact, mathematical algorithmswere not patentable until recently. Before the rulings below, theUSPTO granted only a handful of patents that could be read as math-ematical algorithms, and those few could be attributed more to erroror creative wording than to policy. Whether mathematical algorithmpatents were barred because of a natural law exception, an abstractidea exception, or a mathematical algorithm exception is immaterial.

Does it make sense to have an exception for mathematical prin-ciples? It seems that many side with Plato and Erdos in contendingthat mathematics falls into the peculiar and benevolent design thatJefferson discusses above. For those who see mathematics as a puretechnology, I will discuss the economic motivations for the exceptionin Section II.

The business method exception There is also the question of thebusiness method exception, typically cited as originating from HotelSecurity Checking Co. v. Lorraine Co.14 Examples of the use of thisexception seem to abound in the literature but, in its ruling in StateStreet Bank & Trust Co. v. Signature Financial Group, Inc.,15 theCAFC explained that a wide array of business method cases have allbeen rejected based either on the mathematical algorithm exceptionor on non-novelty grounds, and therefore a business method exceptionhas never existed or been needed.

To give one especially salient example, the CAFC’s ruling in Inre Alappat picked out two seeming business methods for ridicule asunpatentable:16

Maucorps dealt with a business methodology [sic] for de-ciding how salesmen should best handle respective cus-

14160 F. 467 (2d Cir. 1908).15149 F.3d 1368 (1998).1633 F.3d 1526 (1994). Scare quotes in original. See also In re Maucorps, 609

F.2d 481, 484; 203 USPQ 812, 816 (CCPA 1979); In re Meyer, 688 F.2d 789, 796;215 USPQ 193, 199 (CCPA 1982).

7

tomers and Meyer involved a “system” for aiding a neurol-ogist in diagnosing patients. Clearly, neither of the alleged“inventions” in those cases falls within any §101 category.

But in State Street, the court clarified that “closer scrutiny of thesecases reveals that the claimed inventions in both Maucorps and Meyerwere rejected as abstract ideas under the mathematical algorithm ex-ception, not the business method exception.”

The evidence thus seems to indicate that many judges understoodthere to be a business method exception, but may have erred in sodoing. The State Street ruling clarified that most business methods,such as methods for calculating a share price, could easily be invali-dated via the mathematical algorithm exception without recourse toan additional business method exception. I will not attempt to resolvethe question of whether such an exception has existed in the past, andwill follow this ruling’s lead by speaking only of the abstract idea andmathematical algorithm exceptions in the sequel.

2 Congressional intent

The last major patent reform effort was in 1952, before software wasa significant consideration, so there is scant evidence in the congres-sional record as to whether software should be patentable.

However, there is also scant evidence that Congress intended any-thing imaginable to be patented. For example, one Senate report re-garding the 1952 reforms stated that: “[a] person may have ‘invented’a machine or a manufacture, which may include anything under thesun that is made by man, but it is not necessarily patentable un-der section 101 unless the conditions of the title are fulfilled.”17 Thescare quotes around ‘invention’ and the statement that such an al-leged invention is not necessarily patentable suggest that the reportfinds limits to what creative acts may receive patent protection.

Oddly enough, the ruling in Diamond v. Chakrabarty18 used acarefully-cropped version of this statement which gives the oppositeimpression: “[t]he Committee Reports accompanying the 1952 Act in-form us that Congress intended statutory subject matter to ‘includeanything under the sun that is made by man.’” But even with thisreading of the act, Chakrabarty fails to dispute many of the aboveexceptions to patentability.

It should be noted that as of the 1952 reform process, there wasa clear understanding in the courts that there were exceptions topatentability beyond Jefferson’s terse text in 35 U.S.C. §101. Allof the above varieties of exception originate in pre-1952 rulings. How-ever, the Congress failed to elucidate that the exceptions were invalid.

17Senate Report 1979, 82 Cong. 2 sess. (1952). Scare quotes in original.18447 U.S. 303 (1980)

8

3 The Church-Turing Thesis

Much of the debate regarding software hinges around the definitionof a mathematical algorithm, because the definition will advise howone applies the above exceptions.

The general understanding is that an algorithm is a precisely-defined procedure, but that does us no good here, because it doesnot accommodate our intuition that the procedure for evaluating thequadratic equation that we learned in high school (1. Write downthe coefficients a, b, and c. 2. Find b2. 3. Subtract 4ac...) differsfrom the algorithm for producing soap (1. Obtain sodium hydroxide,purified water, and any of a variety of fats. 2. Calculate the correctproportion of sodium hydroxide to fats, using either a saponificationchart or direct calculation. 3. Gradually add the sodium hydroxideto the water, taking care to ensure that lumps do not form. . . . ).

By this definition, it is no surprise that analysts such as Chisum19

found that “The Court erred . . . in implying that algorithms relateonly to mathematical problems. . . ,” because he and others were usinga basically vacuous definition of the term. What patent doesn’t useprecisely-defined procedures?

A better definition of a mathematical algorithm would be that it isthe evaluation of a purely mathematical expression. By this definition,the first procedure above is clearly a mathematical algorithm, becauseit enumerates the steps required to evaluate

−b±√

b2 − 4ac

2a.

Conversely, it would require a great deal of effort and sophistry toexpress the combination of materials of various types in precise con-figurations as the evaluation of a mathematical expression. One couldwrite down the chemical expression for step three above,

H2O + NaOH → Na2O + H2O + H ↑,

but the real-world process of making lye water is at best an approxi-mation of the idea presented by this expression—one must take careto ensure that lumps of lye do not form. Simply put, the quadraticequation evaluation algorithm processes information itself, while thesoapmaking algorithm processes physical ingredients.

An important feature of mathematical algorithms, not shared byphysical algorithms, is that any algorithm can be expressed in a lit-erally infinite number of ways, many of which look very different tothe naked eye. However, the process of running software is indeed theprocess of evaluating a complex expression. Donald Knuth, consid-ered by many to be one of the founders of modern Computer Science,explains:

19Donald Chisum, op cit.

9

[It is not] possible to distinguish between “numerical” and“nonnumerical” algorithms, as if numbers were somehowdifferent from other kinds of precise information. All dataare numbers, and all numbers are data. Mathematicianswork much more with symbolic entities than with num-bers.20

Different programming languages make the mathematical natureof software more or less clear. Some languages, for example theSTELLA modeling language, use pictures and flowcharts to describethe process, while other languages, such as Lisp, use expressions innotation recognizable to any mathematician.

Further, a basic result of Computer Science known as the Church-Turing Thesis indicates that an algorithm written in any language ina very large class of languages can be translated to any other—or toa pure mathematical expression.21

That is, there is a mechanical means of translating any mathemat-ical expression into a computable program, and a means of translatingany computable program into a mathematical expression.

From the perspective of a mathematician, then, all mathemati-cal algorithms, regardless of whether they are expressed in English,mathematical symbols, or FORTRAN,22 are equivalent to the math-ematical expression and should therefore not be patentable under themathematical algorithm exception. It is a direct result of the Church-Turing thesis that the patenting of pure software directly contradicts

20http://lpf.ai.mit.edu/Patents/knuth-to-pto.txt21Alan M Turing, “On Computable Numbers, with an Application to the

Entscheidungsproblem.” Proceedings of the London Mathematical Society, ser. 21936–37, (42): 230–65. Alonzo Church, “A Note on the Entscheidungsproblem.”Journal of Symbolic Logic 1936, 1: 40–41. Alonzo Church, “An Unsolvable Prob-lem of Elementary Number Theory.” American Journal of Mathematics 1936, 58:345–63.

22FORTRAN is generally recognized as the first general-purpose programminglanguage. Its name stands for Formula Translation, providing further indicationthat the engineers who wrote the language recognize the equivalence betweencomputer code and mathematical formulæ.

10

the mathematical algorithm exception.23

4 The Supreme Court’s trilogy

But what if software is used on a physical device to produce real-worldoutcomes? Is a computer on which is programmed a mathematicalalgorithm a patentable device? This is the form of the question thathas been debated in the courts for decades.

Another way to cast the question is in regards to a patent withtwo parts:

23The ruling in Gottschalk v Benson,24 a case regarding an algorithm for con-verting one type of number to another, was very much along the lines of theabove discussion: mathematical expressions can not be patented; the respondentis claiming an algorithm that is equivalent to a mathematical expression; thereforethe patent should not be held as valid. In the ruling’s own words:

It is conceded that one may not patent an idea. But in practicaleffect that would be the result if the formula for converting BCDnumerals to pure binary numerals were patented in this case. . . . ifthe judgment below is affirmed, the patent would wholly pre-emptthe mathematical formula and in practical effect would be a patenton the algorithm itself.

It is worth considering the phrase “wholly pre-empt the mathematical formula”in detail. There are an infinite number of ways to evaluate an expression. Statis-ticians frequently take the log of a set of probabilities to ease calculation;25 takingthe log of the probabilities can be proven to be mathematically irrelevant, andyet it facilitates pragmatic issues of computation. That in mind, there are twoapproaches one could take toward the idea of an algorithm pre-empting the un-derlying formula:

1. The mathematical approach: transformations like taking logs or otherwiseusing new approaches to express the same expression can be proven to be equiv-alent. Therefore, any one method for evaluating an expression always whollypre-empts the formula expressed, because all alternatives are equivalent.

2. The legal pragmatist approach: two algorithms may be mathematicallyequivalent, but there are differences in execution for even the most subtly differentalgorithms. Therefore, a method for evaluating an expression never wholly pre-empts the formula expressed, because all alternatives are pragmatically different.

These perspectives are simultaneously true and correct, and arrive at oppositeconclusions. Of course, patent law tends to be written by legal pragmatists, ratherthan by persons having ordinary skill in the art of computing, and therefore thesecond approach tends to prevail in court. But this formulation is a rehashing ofthe realism-versus-formalism debate: the first approach takes the algorithm as aPlatonic ideal equivalent to all of its manipulations; the second takes it merely asa set of symbols that by luck represent something. The main text offers a meansof framing the math-machine dichotomy in a more productive manner, that doesnot require resolving unresolvable debates.

24409 U.S. 63 (1972).25Ben Klemens, Modeling with Data: Modern Statistical Computing with C,

Princeton University Press (forthcoming).

11

Patent NClaim 1: a useful computing machine, comprising(a) a mathematical algorithm, which may be creatively and painstak-ingly derived, but which is clearly unpatentable by the mathematicalalgorithm exception, and(b) an obvious physical step such as loading the algorithm onto astock computer, which meets the requirements for patentable subjectmatter but is unpatentable because it is not novel.

Can Patent N combine these two steps to form a single device thatis both novel and patentable subject matter?26

The Supreme Court handed down three rulings that present arather clear perspective that an invention that includes a mathemati-cal algorithm is not automatically excluded, but it must do somethinginnovative beyond the algorithm to merit patentability. Simply trans-lating the algorithm into a programming language and loading it ontoa stock computer is not sufficient.

In Gottschalk v. Benson, the respondent sought to patent a general-purpose computer on which is loaded a method for converting fromone type of binary code to another. The Supreme Court held that thepresented algorithm fell soundly within the abstract idea exceptiondiscussed above.

The Benson ruling went on to quote (in concurrence) the 1966President’s Commission on the Patent System:27

Uncertainty now exists as to whether the statute permitsa valid patent to be granted on programs. Direct attemptsto patent programs have been rejected on the ground ofnonstatutory subject matter. Indirect attempts to obtainpatents and avoid the rejection, by drafting claims as aprocess, or a machine or components thereof programmedin a given manner, rather than as a program itself, haveconfused the issue further and should not be permitted.

This ruling is therefore clear on reading Patent N as a mathemat-ical algorithm with a trivial extension that only “confuse[s] the issuefurther,” and therefore unpatentable.

26The reader will note that this formulation of the question, although difficult toresolve, is preferable to the formulations above regarding the relationship betweenmathematics and nature, because it does not hinge on unanswerable philosophicalquestions.

27“To Promote the Progress of ... Useful Arts,” Report of the President’s Com-mission on the Patent System (1966).

12

The position was extended in Parker v. Flook,28 another case abouta mathematical algorithm loaded onto a stock computer, this timeabout a computer that took a few measurements, did calculationsbased on the measurements, and then rang an alarm if a certain vari-able exceeded a certain limit. There was no novelty in the inventionexcept in the mathematical algorithm: “The only difference betweenthe conventional methods of changing alarm limits and that describedin respondent’s application rests in the second step—the mathemati-cal algorithm or formula.”

The question in the case was thus “whether the identification of alimited category of useful, though conventional, post-solution applica-tions of such a formula makes respondent’s method eligible for patentprotection.” The court ruled that it does not:

The notion that post-solution activity, no matter how con-ventional or obvious in itself, can transform an unpatenta-ble principle into a patentable process exalts form oversubstance. A competent draftsman could attach someform of post-solution activity to almost any mathemat-ical formula; the Pythagorean Theorem would not havebeen patentable, or partially patentable, because a patentapplication contained a final step indicating that the for-mula, when solved, could be usefully applied to existingsurveying techniques.

Again, the Court comes out strongly opposed to granting PatentN .

The Supreme Court repeated its position a third time in Diamondv Diehr.29 In this case, the claimed machine was not a general-purposecomputer, but a rubber-curing device that was found to be inventiveby itself. The steps of that patent “include installing rubber in a press,closing the mold, constantly determining the temperature of the mold,constantly recalculating the appropriate cure time through the use ofthe formula and a digital computer, and automatically opening thepress at the proper time.”30 Because the patent covered more thanjust an algorithm plus a trivial physical step, it did not take the formof Patent N , and the Court allowed the patent to stand. However,the ruling reiterated the above warnings regarding insignificant post-solution activity:

. . . [I]nsignificant postsolution activity will not transforman unpatentable principle into a patentable process. Tohold otherwise would allow a competent draftsman to evade

28437 U.S. 584 (1978).29450 U.S. 175 (1981).30ibid. at 187.

13

the recognized limitations on the type of subject mattereligible for patent protection. . . . Because we do not viewrespondents’ claims as an attempt to patent a mathemat-ical formula, but rather to be drawn to an industrial pro-cess . . . for the molding of rubber products, we affirm thejudgment of the Court of Customs and Patent Appeals.

All three of these rulings take pains to make clear that “insignifi-cant”, “conventional or obvious” postsolution activity is not sufficientto transform a formula into a patentable device.31

There are two types of insignificant postsolution activity enumer-ated in the discussion above. The first, in Flook, is the simple appli-cation of a formula to a real-world subject. The question of the casewas clearly answered in the negative: a mathematical expression cannot receive a patent for a limited application.

The second, in Benson and Diehr, is that the shrouding of a math-ematical algorithm in the language of a mechanical invention will nottransform it into a patentable process.

A consistent reading of the Supreme Court trilogy thus arrives at amoderate stance, that still excludes from patentability both pure soft-ware and a stock computer on which is loaded pure software. However,an invention that is novel and statutory subject matter but happensto rely heavily on a computer or mathematical processing is not au-tomatically excluded from patentability.

31Many, including the CAFC rulings below, seem to take Diehr as a reversalof the prior two rulings. The case explicitly addresses both rulings, so if it werea true reversal, one would expect the ruling to state as much. Instead, it takespains to explain why Diehr does not contradict the past rulings. The discussionsof both Benson and Flook were similar; regarding Flook, the ruling stated: “Theclaims were drawn to a method for computing an ‘alarm limit.’ An ‘alarm limit’ issimply a number and the Court concluded that the application sought to protecta formula for computing this number. . . . In contrast, the respondents here do notseek to patent a mathematical formula.”

One can interpret this two ways. The first is to assert that Justice Rehnquistwas ignorant of the fact that the Benson and Flook patents were of the form ofPatent N , thinking that there were no physical claims of any sort. The second isto presume that Justice Rehnquist was aware that the Flook invention includedphysical steps, but they were so insignificant that they did not merit discussion,and the claims could indeed be read as attempts to patent a formula dressed upin the language of physical registers. Those who see Diehr as a reversal seemto read the discussion of Benson and Flook via the first interpretation, but thesecond interpretation is more consistent with the distinction between significantand insignificant postsolution activity in the rest of the ruling, does not imply anunannounced reversal, and does not presume that Justice Rehnquist committeda major factual error in reading two key cases.

14

5 The Freeman-Walter-Abele test

Those in favor of the patentability of a stock computer on which isloaded a novel algorithm also found succor in Diehr, because it stated:

In determining the eligibility of respondents’ claimed pro-cess for patent protection under 101, their claims must beconsidered as a whole. It is inappropriate to dissect theclaims into old and new elements and then to ignore thepresence of the old elements in the analysis.

Without the offsetting discussion above, one could interpret this tomean that the dissection of Patent N into (a) an unpatentable algo-rithm and (b) an unpatentable stock computer is irrelevant, so longas the whole is both novel and somehow physical.

But taken with the caveats and prior rulings, this passage seeks amiddle ground: in some cases the ‘load algorithm onto a computer’step should be detached as insignificant post-solution activity, and inothers it must be considered as an integral part of the whole.

The Freeman-Walter-Abele test was the norm for deciding whenan invention is of the form of Patent N and when it is an integralunit, but was eventually superseded by In re Alappat.

The thinking in In re Freeman,32 In re Walter,33 and In re Abele34

are typified by this passage from In re Walter, a CCPA ruling just ayear before Diehr:

If . . . the mathematical algorithm is presented and solvedby the claimed invention, as was in the case in Bensonand Flook, and is not applied in any manner to physicalelements or process steps, no amount of post-solution ac-tivity will render the claim statutory; nor is it saved by apreamble merely reciting the field of use of the mathemat-ical algorithm.

The rule is consistent with all three Supreme Court rulings, becausethe application to physical elements or process steps in Benson andFlook was basically trivial, while in Diehr, the additional design of therubber-curing equipment was extensive. The rule attempts to addressboth types of insignificant postsolution activity above: the processmust be non-trivially physical, and requires more than “a preamblemerely reciting the field of use.”

32573 F.2d 1237 (1978)33618 F.2d 758 (1980)34684 F.2d 902 (1982)

15

6 CAFC rulings

The Freeman-Walter-Abele test, bolstered by its consistency with theSupreme Court rulings, was the norm in patent law for the remainderof the 1980s, but many complained that it was sometimes difficult toapply. After all, any line between what is patentable and what is notpatentable will have grey areas.

The CAFC responded to the existence of grey areas in the Freeman-Walter-Abele test via a series of additional rulings, key among themArrhythmia Research Technology v. Corazonix Corp.,35 In re Alap-pat,36 In re Lowry,37 State Street Bank and Trust v. Signature Finan-cial,38 and AT & T Corp. v. Excel Communications Inc.39

Simply put, this series of rulings purged the Freeman-Walter-Abeletest of the caveats regarding insignificant postsolution activity.

The first type of insignificant postsolution activity is prefacing analgorithm description with the phrase “a general-purpose computeron which is loaded an algorithm to. . . .” In In re Alappat, the CAFCruled that

We have held that such programming creates a new ma-chine, because a general purpose computer in effect be-comes a special purpose computer once it is programmedto perform particular functions pursuant to instructionsfrom program software.

Because programming a stock computer creates a “new machine,”Patent N must not be dissected into an algorithm plus potentially in-significant postsolution activity; it is therefore fully patentable subjectmatter.40

This seems to be a reversal of the Supreme Court’s rulings andthe Freeman-Walter-Abele test. Applying this logic to Benson andFlook would certainly allow those novel algorithms on stock hardwareto stand as new machines.

The Supreme Court took the generalist approach to the question ofexceptions to patentable subject matter, focusing on finding rules for

35958 F.2d 1053 (1992).3633 F.3d 1526 (1994).3732 F.3d 1579 (1994).38149 F.3d 1368 (1998).39172 F.3d 1352 (1999).40This ruling also means that there is no such thing as a software patent, only

what the European Patent Office calls a Computer Implemented Invention. Thisis a fiction which blurs the fact that a vast number of CIIs can be expressed in theform of Patent N , while a vast number can not. Also, it is rather linguisticallydifficult to maintain, so I will continue to use the term software patent to refer topatents of the form of Patent N , involving a general purpose computer on whichis loaded an algorithm. Ironically, we will see below that this fiction is blockingattempts by policymakers to retain but reform software patents.

16

patentability that, in their view, best promote the progress of Scienceand useful Arts throughout the economy.

Conversely, the CAFC is the archetype of the specialist court, andthe ruling reads as such. Only a few judges on the CAFC bench hearpatent cases,41 and as is natural, most of those are former prominentpatent attorneys.42

The fact that the CAFC’s patent judges are primarily formerpatent attorneys does not necessarily imply a pro-patent or anti-patent bias in specific cases, since every attorney is sometimes onthe prosecution side of a patent case and sometimes on the defense.But the CAFC rulings do demonstrate a patent-oriented worldview.44

They read like a patent attorney conversing with a client, askingwhether the item in question shows creativity and effort that competi-tors could free-ride upon. It is notable that none of the key CAFC rul-ings listed above (from Arrhythmia to AT&T) use the words progressor innovation.

Since every patent attorney has seen patents where a creative indi-vidual combined two common and obvious steps to form an innovativeprocess, it is logical that the CAFC’s members are reluctant to viewPatent N as two separate and unpatentable steps.

With regards to the above unresolvable question of natural philo-sophy—are mathematical results a discovery about nature or an in-vention of man?—one who chose a profession regarding the study ofinventions is likely to lean toward seeing mathematical results as man-made; the Alappat ruling clearly reflects this.45

The second type of insignificant postsolution activity is prefac-ing a pure mathematical expression with a real-world subject matter.Above, Flook took pains to indicate that this is not sufficient to makean algorithm patentable. But in State Street, the court ruled that analgorithm to calculate a share price, a numeric algorithm that takes

41 John R Allison and Mark A. Lemley, 2000. How Federal Circuit JudgesVote in Patent Validity Cases, Florida State University Law Review 27 (Spring):745–66, footnote 30.

42 The CAFC bench boasts a former member of the Board of directors of theAmerican Patent Law Association, a former Chairman of the Patent Committeeof the Law Section of the Pharmaceutical Manufacturers Association, a multitudeof other such patent association titles, and over a hundred years of collectiveexperience in private patent practice.43

43http://www.fedcir.gov/judgbios.html44Jaffe, Adam B., and Josh Lerner, Innovation and Its Discontents: How Our

Broken Patent System Is Endangering Innovation and Progress, and What to Doabout It, Princeton University Press (2004).

45This ruling also follows the lead of Chakrabarty by interpreting the aboveCongressional statement regarding how “[a] person may have ‘invented’ a machineor a manufacture, which may include anything under the sun that is made by man,but it is not necessarily patentable under section 101. . . ” to indicate that Congressintended that there be no limits on patentable subject matter.

17

in one set of numbers and outputs another via simple matrix algebra,is patentable subject matter. The ruling stated:

. . . the transformation of data, representing discrete dollaramounts, by a machine through a series of mathematicalcalculations into a final share price, constitutes a practi-cal application of a mathematical algorithm, formula, orcalculation, because it produces “a useful, concrete andtangible result”–a final share price momentarily fixed forrecording and reporting purposes and even accepted andrelied upon by regulatory authorities and in subsequenttrades.

It is very difficult to reconcile this ruling with the passage fromFlook above that stated “the Pythagorean Theorem would not havebeen patentable, or partially patentable, because a patent applicationcontained a final step indicating that the formula, when solved, couldbe usefully applied to existing surveying techniques.” It seems this isexactly the sort of thing the patent in State Street is doing, indicatingthat a series of matrix multiplications, when solved, could be usefullyapplied to pricing a set of mutual funds. Unfortunately, the StateStreet ruling only mentions Flook once in passing, so we do not knowhow the author of the ruling (Judge Giles Rich) would reconcile thetwo cases.

Again, from the narrow perspective that a business method re-quires creativity and effort to design, and others could free-ride onthat effort, the ruling makes sense. A client walking in to an at-torney’s office with the State Street algorithm in hand would haveblueprints and a story much like those of an electrical engineer client.A patent attorney hearing both stories would disdain rules dictatingthat one may receive a patent and the other may not. That is thespecialist view; but as discussed below, a business method patent anda new circuit design are likely to have very different effects in thegeneral economy.

The rulings make no effort to address the problem of distinguishinginsignificant postsolution activity. State Street simply rejected theFreeman-Walter-Abele test as obsolete: “[a]fter Diehr and Chakrabar-ty, the Freeman-Walter-Abele test has little, if any, applicability todetermining the presence of statutory subject matter.”46 It is difficultto discern whether the CAFC felt that the Freeman-Walter-Abele testwas based on erroneous law or was simply out of fashion.

Because the Supreme Court’s caveats regarding insignificant post-solution activity have been basically ignored, patents on pure math-

46The word after is not to be taken as chronological, since the ordering of therulings is Freeman, Walter, Chakrabarty, Diehr, and finally Abele. Thus, theFreeman-Walter-Abele test both precedes and follows Diehr and Chakrabarty.

18

ematical algorithms, prefaced by mention of application and loadedonto a stock computer, are routinely patented. For example:

• Patent 5,886,908: Method of efficient gradient computation

• Patent 6,434,582: Cosine algorithm for relatively small angles

• Patent 6,664,975: Cheap well-behaved affine transformation ofbounding spheres

• Patent 6,721,771: Method for efficient modular polynomial di-vision in finite fields f(2m)

• Patent 6,904,421: Methods for solving the traveling salesmanproblem

Inspection of all of these patents show that they are of the form ofPatent N : a mathematical algorithm loaded onto a stock computer.Beyond the luck of timing, it is difficult to understand why thesepatents are valid while the Benson algorithm was ruled as an invalidattempt to patent a mathematical algorithm.47

It is clear, from these patents and a multitude of others, thatthere is no mathematical algorithm exception left. If an algorithmis too computationally intensive to be easily done by hand, or if thepatent draftsman competently uses technical language, the applicantcan evade the formerly recognized limitations on the patenting ofmathematical algorithms.

7 The logical conclusion

Now that the “insignificant postsolution activity” rule has been thrownin the dustbin, what is the new line between patentable and un-patentable subject matter?

Unfortunately, there is none to be found: when all that is requiredfor a process to be physical is something as ethereal as a cash paymentor a recorded share price, there is no type of human endeavor that isnot patentable. All of the traditional examples of what is not patent-able, such as a story written to paper, an inventive method of playingan F chord on the guitar, or a method of drawing a pictorial likenessof an object, all involve both an inventive step and a step which is as

47One could perhaps argue that the algorithms can still be executed using paperand pencil rather than a computer, and therefore they are not truly pre-empted.However, the same could readily be said of the Benson algorithm—the rulingeven instructs the reader on the steps. We should recognize that a law statingthat mathematicians may use a computationally-intensive algorithm provided theydo all calculations by hand is (in the words of Benson) to wholly pre-empt thealgorithm “in practical effect.”

19

physical (or more so) than many business method patents, meaningthat any of them could be written in a form like Patent N .

Knight48 argues exactly this point. He explains that the CAFC’srulings (especially In re Lowry, regarding a data structure written toa hard disk) have effectively overturned the “printed matter doctrine”that dictates that writings are protected by copyright and not patent.Not one to merely publish in journals, his firm has submitted applica-tions for patents on storylines and even new words;49 as of this writingthey are still pending.

In Ex parte Lundgren,50 a ruling regarding a method of payingmanagers, the USPTO’s Board of Patent Appeals and Interferences(BPAI) failed to reject such applications:

Our determination is that there is currently no judiciallyrecognized separate ‘technological arts’ test to determinepatent eligible subject matter under §101. We decline tocreate one.

While it implies that the BPAI sees no reason to exclude patents onstorylines and neologisms, this determination may also offer a way out:could one write a technological arts test to exclude storylines and legalmachinations but still allow share prices and virtual machines? Thiswould be a novel rule: as the BPAI notes, there is no precedent for sucha test in U.S. patent law. Further, one would be hard-pressed to writea section of the Manual of Patent Examination Procedure that woulddraw a clear distinction. Revisit the statement by Knuth above: theonly difference between the logical processing of a system of rules (taxlaw, social norms) and the processing of numeric data (share prices) isin the symbols used. Like many Economics papers, patent draftsmencould easily rewrite their claims to include a wealth of Greek symbols,thus making a simple chain of logic look like a treatise on advancedmathematics. To give a concrete example, the databases used bymillions of organizations to keep track of customers, finances, andso forth, are based on a formal algebra, and so one could describedatabase processes via either the relatively quotidian language of listsof names, dates, and sums, or via the language of abstract algebra.Since mathematics is invariant to changes of notation, patent lawregarding mathematics should be as well.

To give another contentious example, consider patents on legalmethods: business methods aimed at creative uses of laws such as

48op cit.49 Application 2005/0244804, “Process of relaying a story having a unique

plot” (about a character who suffers a peculiar case of deja vu); application2004/0249626, “Method for modifying English language compositions to removeand replace objectionable sexist word forms” (in which HIM + HER becomesHIR, for example)

50Appeal No. 2003-2088 (BPAI 2005).

20

tax regulations. To date, the USPTO has granted over a dozen suchpatents, but legal scholars and practitioners feel that such patents arebad policy.51

Schwartz52 makes a number of policy arguments as to why legalmethod patents are bad patents, all of which apply directly to soft-ware and business methods. He points out that “no attorney wantsto pause before advising a client in order to run a patent search tomake sure that no one ‘owns’ the advice she is about to give;” thesame certainly holds for a programmer sitting down to write a pageof code, or a business manager considering how to arrange the fur-niture in her shop. Schwartz is pointing out that (non-patent) lawis a decentralized industry that is not focused on patented methods,and therefore imposing patents will create immense transaction costs.This point is explored fully in the next section.

He explains that “the experience of the past two centuries hasconfirmed the theory that economic incentives are not needed to in-centivize attorneys to innovate,” and it takes absolutely no leap oflogic to apply the same reasoning to both business methods since thedawn of civilization and software since the dawn of transistor arrays.Business methods—methods of saving costs and generating profits—especially do not need the implicit subsidy of a government-grantedlimited monopoly for development.

But if our intuition and policy analysis indicate that storylinesand legal methods should not be patented, and their lack of nontriv-ial physical manifestation does not preclude patentability, how is thelaw to exclude them? Schwartz proposes distinguishing “laws of man”from “laws of nature.” Here, a legal method and a storyline would bemanipulations of laws of man, while software and business methodsare manipulations of laws of nature. Unfortunately, such a distinctionis a rehashing of the debate regarding whether mathematics is a dis-covery from nature or an invention by man, and therefore the sameunresolved issues of categorization crop up again. The patent in StateStreet was over a method of calculating a price for an agglomerationof agglomerations of shares in legally-constructed corporations—whatcould be more man-made? Since the price is “accepted and reliedupon by regulatory authorities,” does that make it a method for com-pliance with law? Is the human nature revealed in a storyline a lawof man or a law of nature? If you prick the characters, do they notbleed?

By eliminating the line that patented devices must be bona fidephysical devices, it has become impossible to preclude legal meth-

51 Andrew A. Schwartz, “The Patent Office Meets The Poison Pill: Why LegalMethods Cannot Be Patented” Harvard Journal of Law and Technology, Vol. 20,No. 2, Spring 2007

52ibid.

21

ods, storylines, or other such structured writings from patentability.No new frontiers delimiting the scope of patentable subject matterhave come forth to replace that erased by the CAFC, nor does itseem that a clear and enforceable candidate is forthcoming. Knuth’srevenge rears its head again: information, being infinitely more mal-leable than physical materials and devices, can easily be recast to usewhatever symbols and jargon the law requires, regardless of the un-derlying meaning. Simply put, once some categories of informationare patentable, all categories are patentable.

The rulings that allowed software to be patentable inexorably ledto the patenting of legal methods and other such pure-informationdesigns. But it is difficult to determine whether storyline and legalmethod patents are just amusing novelties, a brewing storm, or thedawn of a new age in tax loophole innovation. Better would be tolook at the effects that patentability has had on software; since theAlappat ruling is over a decade old, there has been time for the effectsof patents to be felt on this multibillion dollar industry.

II The economic perspective

This section re-asks the same question as above: where should onedraw the line between what is patentable and what is not? Instead ofconsidering the statutory and judicial history, this section instead con-siders the structure of the software industry, and information-based‘industries’ in general. It will show why allowing software and busi-ness methods to be patentable creates transaction costs that easilydwarf the benefits that such patent protection may provide.

The key concept behind the discussion is that these pseudo-in-dustries are massively decentralized, and patents do not efficientlypromote progress in a decentralized industry. Patents are the onlytype of intellectual property protection where independent inventionis not a valid defense against claims of infringement. If there are mil-lions of potential independent inventors, then the waste and economicloss associated with restrictions on independent inventors becomesinevitable.

1 Empirical evidence

The National Research Council (one of the National Academies) ex-pressed concern that the expansion of patentable subject matter “hasoccurred without any oversight from the legislative branch,”53 andthat “the long-term effects of this trend [increasing software patentabil-

53 The Digital Dilemma: Intellectual property in the information age, NationalResearch Council (2002).

22

ity] are as yet unclear, although the near-term consequences are wor-risome.” The study points out that no research had been done onwhether software patents will indeed foster innovation.54

So, now that software has been widely accepted as patentable forover a decade, have proxies for innovation such as the rate of softwareR&D investment shown an appreciable shift?

The answer is no.Finding a jump in software innovation post-1994 would be difficult,

since the computing world was so innovative without patents. The pe-riod when software was widely understood to not be patentable sawthe invention of the Internet at large, the World Wide Web, email, in-stant messaging, word processors, relational databases, spreadsheets,graphical user interfaces, audio compression/decompression software,image manipulation software, some nifty games, and countless otheroptions. One could even think of a few businesses that prospered be-fore business methods were recognized as patentable in 1998. It isthus little surprise that no analyst has yet found that software pat-ents are necessary (or even conclusively beneficial) for innovation insoftware.

Many analysts put a positive spin on the lack of findings. Lernerand Zhu optimistically point out that “. . . little evidence can be foundfor harmful effects. . . .”55

Ronald Mann took the qualitative approach and interviewed ven-ture capitalists and programmers, and found that none of them haltedwork because of software patents, but (as discussed further below) justignored them.56

Mann and Sager looked at software startups, using a database offirms involved in seeking venture capital.57 How did patents affectthe odds that a startup would receive financing? “...[W]e found nosignificance to the existence of pre-financing patents.”58 A suite oflinear regressions found that the presence of patents toward the endof the study period, in 2005, was closely related to total investment,longevity of firm, rounds of financing, late-stage financing, and the exit

54 ibid., p 228.55 Josh Lerner and Feng Zhu, What is the Impact of Software Patent Shifts?:

Evidence from Lotus v. Borland, NBER Working Papers #11168, March 2005.Their analysis does indicate that organizations that patent more show greaterprofits. However, this does not show a social welfare improvement, but merelythat those firms that are granted limited monopolies by the government makemore money than those that are not. The search for overall benefits found onlythe lack of harmful effects cited in the main text.

56 Ronald J. Mann, “Do Patents Facilitate Financing in the Software Industry?”Texas Law Review, Volume 83, Number 4, March 2005.

57Mann, Ronald J. and Sager, Thomas, “Patents, Venture Capital, and SoftwareStart-Ups” (September 2005). University of Texas Law, Law and Econ ResearchPaper No. 57. Available at SSRN: http://ssrn.com/abstract=802806.

58ibid., p 12.

23

status of the firm (closed down, went public, bought out, et cetera).For these regressions, p-values were 0.001 or smaller, indicating strongstatistical significance.

But only one of these indicators was closely related to pre-financingpatents: exit status (p = 0.0157). However, the fact that four out offive regressions found no significant relation indicates that to take thisone significant regression out of context would be to engage in a bit of“data snooping.” The Bonferroni correction, common in biometrics,suggests that if five regressions were run, the p-values should be ad-justed by a factor of five. The p-value should thus be interpreted notas 0.0157, but as 0.0785—not significant at customary levels. Takenas a whole, the results, that indicators of success are strongly cor-related to late-stage patent rates but are at best weakly related topre-financing patents, indicates that causality is not in the directionof patents fostering startup venture investment, but the other wayaround: successful firms eventually begin patenting, perhaps for de-fensive or other strategic purposes.

Robert Merges, another software patent advocate, makes this tepidendorsement: “I will venture a prediction: [p]atents will not cause anyreal and lasting problems. [p 12] ... Something good may come of it.[p 13]”59 Merges also produced an empirical paper60 along similarlines: it found that the software industry has not grown more con-centrated since patents were introduced, and that more experiencedplayers are more likely to patent. However, the empirics again offeredlittle evidence to back up any claims that patents foster innovation,and Merges was again left with a tepid endorsement: “Whatever theeffects of patents on the software industry, this paper concludes, theyhave not killed it.”

However, the purpose of patents is not to “fail to hinder theprogress of Science and useful Arts,” but to promote their progress.Lessig61 argues that the “promote the progress” clause in the Con-stitution is not mere decoration, but a check on how Congress maygrant potentially lucrative limited monopolies. A patent used to beany monopoly granted by the King to a favored subject, but thisclause of the Constitution disfavors such broad usage and allows pat-ents only when needed to promote progress. As the National ResearchCouncil noted above, patents were expanded to software and businessmethods without consideration of whether they promote the progress;the analysis below will indicate that there are reasons to believe that

59 The Uninvited Guest: Patents on Wall Street. Merges, Robert P. EconomicReview (Federal Reserve Bank of Atlanta), 2003 4th Quarter, 88:4, pp 1-13.

60 Merges, Robert P., “Patents, Entry and Growth in the Software Industry”(August 1, 2006). Available at SSRN: http://ssrn.com/abstract=926204.

61 Lawrence Lessig, Code and Other Laws of Cyberspace, 1999, Basic Books, p133.

24

patents in these fields will have more of a hindering effect than in otherfields; even the most pro-patent empirical investigators were unable tofind evidence of the promotion of progress; and software and businessboth flourished without patent protection. All of this indicates thatthe expansion of patentability to include these new subject mattersmay be unconstitutional.

2 Patent efficiency in massive industries

The lack of findings is rather easy to explain. Imagine a study toconsider the effect of business method patents on the business methodindustry—by which I mean all businesses. It would be absolutelyastonishing if the productivity of United States business could beproven to hinge on the few thousand business method patents grantedto date. The story with software is similar.

Define a potential innovator as someone who has sufficient humanand physical capital that they could conceivably independently inventthe innovation in question. For the typical pharmaceutical patent, theset of potential innovators includes only experienced chemists withaffiliation to a research lab.

On the other end of the spectrum, consider the famous Southwest-ern Bell Corp (SBC, now merged with AT&T) patents for a “Struc-tured Document Browser,”62 described in SBC’s own words below.Here, the set of potential innovators includes anyone with a home com-puter and a basic knowledge of a web-design tool such as a word pro-cessor. Netcraft, an Internet data and consulting company, counted106.9 million web sites in January of 2007.63 By the definition here,the administrator of any one of these sites is a potential innovator.64

SBC actually pursued its claim against a broad range of companieswith the misfortune of having a web site. For example, they wrote toone toy company:

We recently observed several useful navigation featureswithin the user interface or your site www.museumtour.com.For example your site includes several selectors or tabsthat correspond to specific locations within your site doc-uments. These selectors seem to reside in their own frame

62Patents 5,933,841 and 6,442,574.63 One web site equals one domain name. Netcraft January 2007 Web

Server Survey, http://news.netcraft.com/archives/2007/01/05/january2007 web server survey.html

64 RIM v. NTP65 found that if there is “control and beneficial use” of a patentinside the U.S.A., then the device is beholden to U.S. patent laws. Therefore, wemay include a web site hosted anywhere in the world in our tally of sites thatmust take U.S. patent law into account. This may be reversed by the currentlypending case of AT & T v. Microsoft.

65392 F.3d 1336, (Dec. 14, 2004).

25

or part of the user interface. And, as such, the selectorsare not lost when a different part of the document is dis-played to the user [. . . ]. By separating the selectors fromthe content, Museum Tours has truly simplified site navi-gation and improved the shopping experience for its users.As you review the Structured Document Patent you willnotice that the above-discussed features appear to infringeseveral issued claims in our patent.

The salient feature of this letter is not that it describes what seemsto be a rather obvious design—that issue will be discussed below.Rather, the key point is that SBC, a telephone company, was sendinga software patent infringement letter to a toy company. It is unlikelythat the owners of Museum Tour thought of themselves as a computersoftware firm, but by having a web site, they effectively were.

The latest accounting from the Bureau of Economic Analysis di-vides the software market into three parts: retail, consultants, andin-house, which are evenly split in the U.S. economy. Of the $232.5billion spent on software in 2002, 32.6 percent bought prepackagedprograms, 36.4 percent custom-built ones, and 31.0 percent paid forsoftware written in-house.66 A report commissioned by the EU used adifferent method and different definitions that indicate that softwaresales represents an even smaller portion of the software economy: bytheir estimates, 16% of 2002 U.S. software sales went to proprietarysoftware sales, 41% went to development and customization services,and 43% paid for internal development.67

These statistics indicate that it is wrong to think of software pat-ents as only affecting companies that vend software. The majorityof software designers and authors are at companies where software issimply necessary for daily operation. Whereas the set of people whohave the know-how, equipment, and inclination to work on a mechan-ical device may be limited to a few hundred organizations, the set ofpeople who work full-time on writing new and potentially innovativesoftware is distributed among millions of organizations.

And the set of businesses that strive to use innovative businessmethods is simply the set of all businesses.

The size of the software and business method pseudo-industriesderives directly from the elimination of the concept of insignificantpostsolution activity. Information and the tools to manage informa-

66 U.S. Department of Commerce, Bureau of Economic Analysis, “Recognitionof Business and Government Expenditure for Software as Investment: Methodol-ogy and Quantitative Impacts, 1959-98” (www.bea.gov/bea/papers/software.pdf). Updated with 2002 data at www.bea.gov/bea/papers/table11.xls.

67Study on the: Economic impact of open source software on innovation andthe competitiveness of the Information and Communication Technologies (ICT)sector in the EU, final report, 20 November 2006, p 124.

26

tion are truly ubiquitous, but the equipment needed to engage in sig-nificant postsolution activity, such as testing a chemical’s propertiesor verifying that a machine actually runs, are available to a more lim-ited range of people. Therefore, when inventions with no significantpostsolution activity are allowed, the number of potential innovatorsgrows exponentially.

This also provides justification for the mathematical algorithm ex-ception to patent law. The set of people who use mathematics is sim-ply the set of all humans, so all of the problems regarding applyingpatent law to pseudo-industries that cover the entire economy alsoapply to the set of all mathematics users. Thus, even if one believesthat there are no ethical objections to the patenting of mathemati-cal results, there are pragmatic reasons for treating it exceptionally.Information processing is simply different from the processing of tan-gible ingredients.

Scaling Let us say that a producer has a product that embodies apatent. On the one hand, additional research could provide benefit toa larger number of people—patents provide more benefit in a massivemarket. On the other hand, the fact that the potential market ismillions instead of thousands means that it is much easier to findcustomers and it is possible to profit on smaller per-unit margins. Sothere is less need for government intervention in the market—patentsprovide less benefit in a massive market. One could write a modeland numbers to tip the scale either way.

We can say more about how the number of potential innovators(rather than potential customers) will affect the value of a patent.

The likelihood that someone will stumble upon and reveal an in-novation is larger when there are more agents searching. If, in a no-patent world, there are already millions of people in a field searchingfor a new innovation, then patents are less likely to add value.

Some argue that “patent races” induce faster innovation. The ideais that if there are multiple parties that could all potentially achievean invention, but the first will be granted a monopoly on the result,then all parties will invest more in research—and do so more quickly.The drawback to the patent race, however, is that the investmentsmade by the losers in the race are basically unproductive, lost costs.

In the economic literature, different papers present different re-sults, depending on how the patent race is modeled. Dasgupta andStiglitz68 find that “. . . competition in R&D increases the level of in-novation, possibly beyond the socially optimal level.” Conversely, Sahand Stiglitz state that “. . . the number of firms has no effect on the

68 Partha Dasgupta and Joseph E Stiglitz, Uncertainty, Industrial Structure,and the Speed of R&D, Bell Journal of Economics, 11:1, pp 1–28.

27

pace of innovation in a market economy.”69 I take the economic liter-ature to give weak support to either claims that patent races raise ordecrease the benefit from patents.

Now consider the transaction costs of patents. Transaction costsinclude the cost of negotiating licenses with the monopolist, the costsof patent searches, and the cost of legal disputes.

For a clear patent, the cost of negotiating licensing could be small:the monopolist may develop a standardized take-it-or-leave-it pricelist, and those who know they will need a license will quickly getthrough the necessary paperwork. Of course, the paperwork riseswith the number of actors, but we can generally expect that the effectof a larger number of innovators is positive but small.

This is especially the case if there is a bottleneck where users canbe collectively licensed. For example, Apple paid Creative Technology$100 million for the right to implement its method of sorting songs onthe iPod,70 and so the iPod’s millions of users implicitly licensed thetechnology from Creative.

There is also the cost of determining that there is a patent andfinding the patent owner—patent search. To truly eliminate the riskof patent infringement, all industry members must do regular patentsearches. If the cost of a patent search were fixed, twice as many par-ticipants means twice as much spent on searching. This is especiallytrue because subtle differences in operations can make for entirelydifferent patent situations, so it is difficult for multiple parties topool patent searches. One would be hard-pressed to find an attorneywho would work for such rates, but say that a minimal patent searchcosts about $10,000. Then if every potential infringer of a softwarepatent—all 106 million web sites—took U.S. patent law seriously, thetransaction costs would total about $1 trillion. As it is, the world’sweb economy progresses by mostly ignoring U.S. patent law.

On top of this, add those businesses that engage in business meth-ods—that is, all businesses. When deciding how to set price lists, amanager is a potential innovator who must do a patent search andtake action on the results to ensure that the price list is not infringingany patents. Again, this would impose astronomical costs if takenseriously; the economy moves by simply ignoring the requirements ofpatent law.

69 Raaj Kumar Sah and Joseph E Stiglitz, The Invariance of Market Innovationto the Number of Firms, RAND Journal of Economics, 18:1, pp 98–108.

70 Apple Agrees to Pay $100 Million to Resolve IPod Patent Litigation Los An-geles Times, 24 Aug 2006. p. C.2. Apple pays $100 million U.S. to settle patentsuit Toronto Star, Aug 24, 2006, p. C3. Nick Wingfield, Apple, Creative Tech-nology Set Pact Ending iPod-Tied Dispute, Wall Street Journal, Eastern Edition.Aug 24, 2006. p. A.2.

28

3 Patent trolls

One final transaction cost is the possibility of litigation. The cost oflitigation, when it occurs, easily dwarfs the cost of a patent search. Inan ideal world litigation never happens, and parties are all sufficientlyaware of the law that they can all stay out of the courtroom.71

The worst case would be a situation where the members of onesubset of the industry hold patents, and the members of another sub-set are entirely ignorant of those patents. In such a case, the likelihoodof wasteful infringement suits rises significantly over a situation whereeveryone is equally informed of what is patented.

This is exactly the case in software and business methods. Becausethe set of potential innovators is so large, the likelihood of independentinvention is high and the likelihood that an independent inventor hasdone a patent search is low. SBC had little difficulty finding a numberof web sites like that of the toy company that infringed its patent.

It is an easy syllogism: there are 100 million potential independentinventors; independent invention is not a valid defense against claimsof patent infringement; therefore opportunistic infringement claimsare basically inevitable.

In this context, we see the rise of the patent troll. One preva-lent definition of the term72 is that a patent troll is a company thatholds a patent not to produce a product, but to extort royalties fromothers. However, non-producing patent holders are common. Untilan inventor finds an investor to provide capital, she remains a non-producing patent holder, and if she shows the blueprint to an investorand he uses it without permission, then she has every reason to sue.Although such suits are wasteful by themselves, one can argue thatthey are beneficial in the long term because they support the systemthat allows inventors and capital-holders to fruitfully interact.

But it is much more difficult to argue that a lawsuit against anindependent inventor in a different field entirely provides any socialbenefit. At best, the suit is simply court costs wasted on rent-seeking;at worst, such a suit could create the stifling impression that onemay not engage in any but the most obvious activity without firstconsulting a lawyer.

Thus, a more useful definition is that a patent troll is one whounfairly takes advantage of informational asymmetries by suing inde-pendent inventors who are ignorant of the field of patents in which thetroll works. Informational asymmetries often lead to outcomes whichallow the agent with more information to extract rents from agentswith less information, and which are socially suboptimal.73

71 See, e.g., Richard A Posner, Economic Analysis of Law, 3rd ed, Little, Brown& Co. (1986) §21.

72Attributed to former Intel Assistant General Counsel Peter Detkin.73See, e.g., Hal Varian, Microeconomic Analysis, 3rd ed, W. W. Norton & Co

29

In a symmetric situation, both the patent holder and potentiallicensees are aware of the patent, and therefore the potential licenseescan plan accordingly, by either negotiating a patent license beforeembarking upon any major investments, or by passing on to otherprospects.

In an asymmetric situation, the potential licensee invests its assetsinto its business without knowing about the patent. This produces anasymmetric bargaining position, where the patent holder may sue forinfringement damages up to the full value of the company. To give oneexample, Kodak demanded a billion dollars from Sun Microsystemsdue to an patent infringement regarding its Java compiler and virtualmachine—which Sun gives away for free.74

Looking at the list of companies that the media have dubbedpatent trolls, including Acacia, BTG, DataTreasury, Eolas, Forgent,and NTP, all of them work in software and business methods, and thisis no coincidence. There are no asymmetries of information in phar-maceuticals or mechanical engineering, because everyone is aware ofthe patent process, and for any given field there are a limited num-ber of players (tens or hundreds, not tens of millions) so the costs ofpatent search are limited. Because the number of parties are small,the likelihood that another party independently invented a patenteditem is relatively small. Conversely, for software or business methods,the question is not whether someone ignorant of the patent has alsoimplemented it, but how to find that party and what to bill them.

The problem with patent trolls has clearly taken hold in the me-dia75 and even in Congress,76 but is it all merely anecdote?

Even if it is, the anecdotes by themselves are impressive. In 2006alone, we saw payments of $612.5 million to NTP from RIM (the

(1992) Chapter 25: Information.74 John Oates, Kodak Wins Sun Java Patents Case, Wants $1bn, The Regis-

ter, October 4, 2004 (www.theregister.co.uk/2004/10/04/kodak wins java/).Ashlee Vance, Sun Settles Java Spat with Kodak for $92 Million, The Register,October 7, 2004 (www.theregister.co.uk/2004/10/07/kodak sun settle).

75 The Washington Times, ’Patent trolls’ and market dominance, WashingtonTimes, October 4 2006. Alan Cane, Trolls control the rickety-rackety bridge ofintellectual property, Financial Times, September 20 2006. Julie Creswell, A WallStreet Rush to Patent Profit-Making Methods, New York Times, August 11, EastCoast Late Edition. William M. Bulkeley, Freedom Wireless To Get $55 Million InPatent Accord, Wall Street Journal, July 22 2006, Eastern Edition. Patent trollinga thorn in the side of hi-tech giants ’Spectators’ build up an IP portfolio and thenwait for a big firm to infringe upon it, South China Morning Post, May 23 2006.Richard Waters, Invention shop or troll factory?, Financial Times, April 26 2006Bruce Sewell, Troll Call, Wall Street Journal, March 6, Eastern Edition. TimHarper, America’s patent trolls: Are they out of control?, Toronto Star, February3 2006.

76Based on interviews with members of the Senate Subcommittee on IntellectualProperty.

30

makers of the Blackberry),77 $92.3 million to Acacia from variouscompanies throughout all industries (in first 3Qs of 2006, plus $27.6min 2005),78 and $8 million to Fogent from a coalition of licensees.79

So in 2006 to date, these few opportunistic lawsuits took in $712.8million. This includes only opportunistic threats against companiesthat independently invented the software in question—and it shouldbe no surprise at this point that these cases are entirely about soft-ware. To give a sense of scale, Google made $733 million in net rev-enues in the third quarter of 2006.80

Since Microsoft is perceived as the deepest pockets in the tradi-tional software industry, it is frequently the target for software in-fringement suits. Settlements and rulings against the giant include$521 million to Eolas (not yet settled),81 $440 million to Intertrust,82

$60 million to SPX,83 and $60 million to Burst.com.84 We may wantto include in this the $900 million Microsoft paid to Sun for patentlicensing, but it is unclear whether the claims were over independentinvention or Microsoft actually imitating a competitor’s products.85

Adding up the settlements, we find a variety of companies, some intraditional software and many elsewhere, paying billions of dollars forthe right to use software that they conceived and wrote without out-side assistance—and that’s just the headlines. I exclude settlementslike DataTreasury’s 2006 settlements with BankOne and JP Morgan,where the exact amount paid to license independently written soft-ware is surely in the hundreds of millions, but the exact amount isnot disclosed to the public.

Further, there is no reliable way to document the constant streamof letters and settlements that fly under the radar, though there area few rules of thumb presented by experts that may help us get a

77 Settlement reached in BlackBerry patent case: Research in Motion pays NTP$612.5 million; devices to stay on Associated Press, via MSN, http://www.msnbc.msn.com/id/6448213/did/11659304?GT1=7935

78Acacia fact sheet, accessed 28 December 2006. http://www.acaciaresearch.com/pr/AcaciaFactSheet.pdf

79Michael Kanellos, Forgent settles JPEG patent cases, CNET News.com,November 1, 2006, http://news.com.com/2100-1014 3-6131574.html

80Last available quarter. Google Inc. Condensed Consolidated Statements ofIncome, available from the SEC’s EDGAR.

81Michelle Kessler, The 521-million-dollar man, USA Today, 30 September 200382Intertrust press release, available at http://www.intertrust.com/main/ip/

settlement.html.83Keith Regan, Microsoft Settles Patent Lawsuit for $60M, 29 December 2003

http://www.technewsworld.com/story/32479.html84Underdog Or Patent Troll? How Burst.com went from making software to su-

ing tech giants BusinessWeek Online, 24 April 2006, http://www.businessweek.com/magazine/content/06 17/b3981070.htm

85Stephen Shankland Sun settles with Microsoft, announces layoffs CnetNews, 2 April 2004, http://news.com.com/Sun+settles+with+Microsoft,+announces+layoffs/2100-1014 3-5183848.html.

31

ballpark figure. Chip Lutton of Apple Computer86 offers a “rule of25,” that for every case litigated, there are 25 that are filed, and forevery 25 that are filed, there are 25 notice letters claiming a patentinfringement sent out. Simply receiving a notice letter imposes com-pliance requirements that will cost between $30,000 and $50,000;87

multiplying together these two rules of thumb, we get around $18.7 to$31.2 million in legal fees spent on handling notice letters and otherlow-level compliance matters for every patent lawsuit filed. With 600software patent suits filed per week88, these costs quickly grow torather large amounts. Although it is near impossible to get a solidestimate of software patent compliance costs, the above rough guidesindicate that an estimate in the hundreds of millions would not beunreasonable.

It may be that opportunistic lawsuits against independent inven-tors are mere anecdotes in the sense that they are not a sufficient dataset for rigorous time series analysis, but when the anecdotes total bil-lions of dollars, they are worth paying attention to nonetheless.

A decade from now, can we expect that there will be more or feweropportunistic infringement suits? There are two competing forces thatwill determine whether we will be seeing more or fewer opportunisticinfringement claims in the future.

One the one hand is patent reform. This is slow, and may notwork, because it is simply impossible to halt patent trolls—by eitherdefinition of the term—while allowing the rest of the patent world tocontinue unabated. As discussed below, any type of patent reform inCongress has been stalled since software patents became an importantissue.

On the other hand, the key bottleneck to the opportunistic law-suit is capital, and we can expect that that will be easier to obtain.A fully-prosecuted infringement case can cost millions of dollars, andruns some risk of losing it all. Each of the half-billion dollar settle-ments above bolsters the business case for those who want to enterthe opportunistic infringement lawsuit industry. That is, successfulpatent trolls breed other patent trolls, so we can expect that the prob-lem will only get worse in the near future.

Patents have been wed to massively decentralized industries foronly a decade, and a fully prosecuted patent suit could take severalyears. As such, it is impressive and noteworthy that we have alreadyseen the opportunistic infringement suits that we have. But we can

86Testimony before the House Subcommittee on Courts, the Internet, and In-tellectual Property, April 20, 2005. Thanks to Brian Kahin for pointing out thistestimony.

87Daniel Ravicher Software and Law: Is Regulation Fostering or Inhibiting In-novation?, presentation at the Brookings Institution, 7 December 2005.

88ibid.

32

expect that in the future, more people are going to comprehend thesyllogism that when there are a hundred million potential independentinventors, and independent invention is not a defense, then profit iseasy to obtain.

4 Contributory infringement

If a car manufacturer is infringing on a patent on, say, its windshieldwipers, then every owner of the car is a contributory infringer, andis theoretically equally liable under the law.89 Thus, one could ar-gue that the car industry creates patent liability that is as massivelydecentralized as the software industry.

There are three key differences between such an argument and thesituation with software, that further elucidate the many ways infringe-ment can occur in a massively decentralized industry. First, unlike thetheoretical contributory infringement suit above, software patent in-fringement lawsuits have actually been taken out against a very broadrange of companies who do not claim to be software companies. Togive another example, Acacia Media Technologies has brought suitagainst Johns Hopkins University, the University of Virginia, the Uni-versity of Wyoming, and a number of other schools claiming that someonline components of their classes violate Acacia’s patents.90

Novell, Inc. is one of many companies that package the Linuxoperating system with a number of other components and providesupport to corporate clients. Microsoft has arrived at an agreementwith Novell regarding patents it claims that are infringed by Linux.The implication is very clear: companies that use a non-Novell versionof Linux in their data centers could face a patent infringement suit.91

Even if no lawsuits are ever filed, such threats are clearly intended toinfluence how companies in all industries select the software they willuse for their accounting database and email server.

Second, under the logic of In re Alappat, the act of loading soft-ware onto a general purpose computer creates a “new machine,” soloading software onto a computer is direct infringement, unlike thecontributory infringement of driving a car with unlicensed windshieldwipers. Acacia is not suing universities for contributory infringement,but for direct infringement, and it is likely that if Microsoft were tosue a Linux-using company, they would do the same.

Third, and most importantly, there is very little unmodified soft-ware running in the world’s server rooms. Without fail, one needs

8935 U.S.C. §271(c) (2006).90Corey Murray, “Schools targeted in streaming video patent claim,” eSchool

News, http://www.eschoolnews.com/news/showStory.cfm?ArticleID=4937.91Robert A. Guth and Don Clark, “Microsoft, Novell Reach Accord on Linux”,

Wall Street Journal, 3 November 2006, p A3.

33

to reconfigure off-the-shelf software to work with the local situation,perhaps even rewriting the program’s internal algorithms, which iswhy a third of software spending is in-house. Such behavior createsanother opportunity for direct infringement. The toy company’s website may have been written using an off-the-shelf web design program,but it was a company employee who designed the page in a mannerthat infringed the SBC patent. Thus, it is wrong to describe the worldas consisting of a few centralized software authors and a wide arrayof downstream consumers: the downstream users are actively writ-ing new code and modifying existing code, blurring any distinctionbetween the author and user.

5 Is it all just obviousness?

Some contend that the entire problem with software and businessmethod patents—the reason that only these industries and none otherhas a serious problem with opportunistic lawsuits—is that patent ex-aminers for these industries allow obvious patents that would not beallowed in traditional industries where the patent record and PTOexperience is much more extensive. Of course, “patent quality” andthe line between obviousness and non-obviousness will always be aproblem, and I will not dispute that many seemingly obvious patentshave been granted. However, the obviousness problem does nothingto alleviate the problems with applying patents to a massive industry.

First, let us say that there were only 200 patents in the softwareindustry instead of 200,000. Every web site would still need to verifythat it was in compliance with those patents. One can think of thecost of patent clearance as an affine linear function: there is a fixedcost (say, $10,000) for retaining a lawyer to search one patent, plusan additional cost for searching each additional patent (say, a nickel).In such a case, it is the fact that the industry consists of about ahundred million actors, not the number of patents, that makes searchcosts potentially astronomical.

The granting of a patent can take several years, meaning that theabove discussion about independent invention is invalidated only if apatent is nonobvious both at the date of application and several yearslater at the date of granting. SBC’s above-mentioned patent is oftenmentioned as the perfect example of an obvious patent, but on thedate of application (May, 1996), graphical web browsers were still newand in limited use. But as millions of people began using the medium,what had been non-obvious and obscure quickly became obvious tomillions.

Some would contend that subjects like the aforementioned “Methodfor efficient modular polynomial division in finite fields f(2m)” is un-likely to have a massively decentralized audience like the “Structured

34

Document Browser” does. That is, the problem of massively decen-tralized software production could be alleviated via a technical artsrequirement. As discussed above, it would be basically impossible towrite such a test. The Structured Document Browser patents are al-ready couched in the technical language of GUI rendering of SGMLDTDs, and if the draftsman had to, he could likely express it all interms of the manipulation of byte arrays in certain hexadecimal reg-isters.

It is natural to presume that as one raises the standards for whatis obvious, the likelihood of independent invention falls. In an idealworld, we would define nonobvious to mean that the likelihood ofindependent invention is low, and assign patent examiners to evaluatethis likelihood.

In such a world (which is very different from the one in which welive), examiners would assign different standards to different fields.Eliminating the risk of independent invention with a few hundredintegrated circuit (IC) manufacturers involves a much lower standardthan what would be required to eliminate the risk of independentinvention among 100 million web sites. So the first step in eliminatingobvious patents in software would be to establish differential standardsfor patentability in different fields.

But recall the fiction that loading software on a computer creates a“new machine;” to maintain this fiction, the USPTO has no separatesoftware category. So the next step in implementing higher standardsfor software patents would be for the USPTO to categorize softwareseparately from ICs—but the courts base the validity of software pat-ents on the claim that such a distinction is impossible to make. Thiswould not be the first self-contradictory law in U.S. history, but suchthings are not to be encouraged.

Our options are thus to set an incredibly high standard for in-tegrated circuits for the sake of better software patents, continue totolerate lawsuits against independent inventors in software for thesake of allowing IC manufacturers to continue as they do now, or toacknowledge that software and ICs are distinct—which would be tooverturn the rulings that Patent N is a bona fide physical device likeany IC.

6 Disclosure

Campbell-Kelly92 points out that patents also serve the purpose ofdisclosing innovative ideas to the public that would otherwise havebeen kept a secret. With an astronomical number of potential inno-vators at a comparable level to the patent-holder, this means that the

92 Martin Campbell-Kelly, “Not All Bad: An Historical Perspective on SoftwarePatents,” 11 Mich. Telecomm. Tech. L. Rev. 191 (2005).

35

benefits to disclosure may also be great.But the disclosure story, while very sensible in theory, has a num-

ber of important problems in the real world. The first is that firmsin all industries tend not to search patent databases for technologicalinstruction. Arora et al.93 state that “patent disclosures appearedto have no measurable impact on information flows from other firms,and therefore no measurable effect on R&D productivity.” Arundel94

finds that “a consistent result in survey research on the use of patentdatabases is that they are among the least important external infor-mation sources available to firms.”

On top of this base, there are further problems with disclosureamong software and business methods. In Northern Telecom v. Data-point,95 the CAFC ruled that writing source code—in fact, authoringanything more detailed than the broad flowchart describing the overalllogic of the design, is a “mere clerical function.” Therefore, the dis-closure requirements of a software patent are minimal. So if a patentdoes not need to reveal any but a basic flowchart, what does it reveal?

On one end of the software spectrum are patents for user interfacedesigns such as the SBC patent, whose workings are evident frominspection, so the patent provides no disclosure. Any patent on anon-software design, such as a business method or storyline, wouldalso fall into this class, and would therefore provide no informationalvalue beyond the product itself.

On the other end of the spectrum, one finds the algorithm forLZW (Lempel-Ziv-Welch) encoding, which Campbell-Kelly cites as asuccess story for patents. However, as Campbell-Kelly notes, TerryWelch had published the algorithm in a peer-reviewed journal shortlyafter applying for the patent, but years before receiving it.96 If MrWelch had truly wanted to keep the LZW algorithm a trade secretshould patent protection be unavailable, then he would have waitedfor the ink on his patent application to dry before publishing thealgorithm.

To generalize from the anecdote of Mr Welch, computer scienceat the level of advanced algorithms for data processing is still moreabout journal publications and academic recognition than about pat-ents, and one is hard-pressed to find a programmer who searches thepatent literature for disclosure of ideas. Mann97 interviewed softwaretechnologists and found that the apathy toward learning from pat-

93Arora, A. et al., 2003. “R&D and the patent premium”, NBER WorkingPaper No. 9431, p 17.

94Anthony Arundel, Patents in the knowledge-based economy, BeleidstudiesTechnology Economie, 37: 67-88, 2001.

95908 F.2d 931, 940–41 (1990).96 Terry A. Welch, A Technique for High Performance Data Compression, Com-

puter, June 1984, at 8.97Mann, op cit.

36

ents revealed by the general surveys is evident in software as well:“. . .none of the startup firms to which I spoke suggested a practice ofdoing prior art searches before beginning development of their prod-ucts.”98

Thus, although disclosure is a positive benefit from patents in somefields, an overall perception among technologists that patent searchesare not an effective manner of learning new techniques, the weak in-terpretation of the enablement requirement, the self-evident nature ofmany software and business method designs, the difficulty of search-ing patents relative to the ease of searching journals and online codebases, the academic nature of the more arcane aspects of computerscience, and the speed of the industry all collude to make patentson software and other information-processing technologies virtuallyuseless for disclosure purposes.99

III Conclusion

To summarize the example of the software industry, the millions ofplayers basically ignore patents, and the empirical analyses show justthis. But a few lucky losers are sued for the software that they inde-pendently wrote, and must spend up to millions of dollars defendingagainst claims of infringement. Such lawsuits are not an artifact ofobviousness or details of patent procedure: if a patent holder hasa preeminently valid patent, and there are 100 million others work-ing in the field, the patent holder need only search to find someonewho is unknowingly infringing. As the current crop of litigators showhow easy it is to find unknowing infringers, one can expect that thenext generation of litigants will have an easier time funding their ownsearches for infringers.

The problems with patents in a massively decentralized industrystem from the elimination of the “insignificant postsolution activity”stipulation on patentable subject matter. Virtually every companyin the country has a computer on hand, and those of a certain scalehave full-time employees actively writing code. Yet only a handfulof companies have employees actively synthesizing new drugs. Thisis closely related to the fact that information processing is a general

98op cit. p 1004, italics in original.99Returning to the scaling question, the fact that there is no evidence of com-

puter scientists or business method users searching patents for research purposesmeans that the benefits from a patent do not spread to other members of theindustry. So if the social benefit to a software patent’s disclosure in a world witha thousand potential innovators is $100,000, then the benefit in an industry ofa hundred million potential innovators is still $100,000. But as above, the riskof opportunistic and socially destructive litigation increases significantly as thenumber of potential innovators expands.

37

requirement for anyone with information, and that information pro-cessing tools are considered to be stock equipment.

Thus, the recommendation of this paper is not an innovation, buta regression. A great many of the problems with patents that fillthe newspapers and vex businessmen can be solved by reinstating thedistinction from Diehr and its predecessors that indicate a device ispatentable only if it is based on steps that are simultaneously noveland non-trivially physical.

By respecting the caveats about postsolution activity in the Free-man-Walter-Abele test, the printed matter doctrine would be re-estab-lished, because the process of printing to paper, a ribbon, or a harddrive is not physically innovative, regardless of how inventive the databeing written may be.

The rule that a patent be physically innovative in no way hampersthe traditional process patents that manipulate materials into newmaterials. A process such as a novel means of folding a sheet orprocessing flour can not be decomposed into a design step that makesno mention of the materials and an insignificant postsolution activitystep that can be executed with little regard to the algorithm.

Some will argue that processing a data structure and writing itto a hard drive are intrinsically linked—that this line is too difficultto distinguish in practice. Yet the USPTO somehow managed todistinguish this type of physical process from more traditional physicalprocesses for decades. The distinction certainly created more disputesand borderline cases than the CAFC’s position, but that is becausethe CAFC ended any debate by simply eliminating subject matterrestrictions entirely.

Patent applicants work hard on blurring all distinctions: withenough flowery jargon, stock computers and even the process of puttingpen to paper can look like innovations. But the patent examiner hasthe ability to limit the scope of any claims. For example, the examinermay add an annotation that what is claimed is the given inventionwhen executed using physical steps that are beyond those executed bystock equipment. This would be more than sufficient to distinguish,say, a novel computer chip fabrication from the same logic executedin software, or a new portable email device from a new means of usingstock networks to transmit emails.

However, the concerns about whether patent examiners will befooled are overblown by a simple observation bias problem. Lawreview articles and key court cases are always about the edges ofpatentability, but the great majority of software patent applicationsare clearly of the form of Patent N : an algorithm loaded onto what isobviously a stock computing device. Respecting the declaration that“insignificant postsolution postsolution activity will not transform anunpatentable principle into a patentable process” would easily elim-

38

inate the great majority of such patents without any need for finejudgment calls.

1 Politics

How can limits on patentable subject matter be restored? As of thiswriting, there are a number of venues in government for debate overthe course of patent law. In the Executive Branch, the USPTO hasintroduced a draft strategic plan that will introduce reforms such asa multi-tiered system that one patent attorney described as “a decon-struction of our unitary patent system.”100

In Congress, both the House and Senate have introduced bills, butgiven the current condition of patents, it is unlikely that any beneficialreform will come from them, for two reasons.101

First, the CAFC rulings that software companies and pharmaceu-ticals are beholden to the same patent law means that building con-sensus on any given patent reform is suddenly twice as difficult. Asa very broad rule, pharma companies generally seek stronger patentprotection, while software companies tend to seek more restrictions onpatentability; building consensus from two groups pushing in oppositedirections is especially difficult.

The second problem is the collective action problem. As above,there are about 100 million potential targets for a patent infringementsuit regarding a web page, and as many potential targets regardinga patent on a business method as there are businesses. However, arelatively minuscule number of individuals and organizations actuallyhold patents. The few who stand to profit significantly from theirpatent business are much more likely to lobby Congress than the dif-fuse masses.102

The other option for restoring limits to patentable subject matteris the courts—and since the CAFC has shown a specialist’s enthusi-asm for allowing patents on all subject matter, it is up to the general-ist Supreme Court to consider what patent limitations best promoteprogress. As discussed above, members of the Supreme Court haveshown interest in the question of defining patentable subject matter,via the dissent in the rejection of certiorari in LabCorp v. Metabolite.Given that technicalities precluded reaching a verdict in that case,

100 J Matthew Buchanan, “Friday food for thought: The chang-ing scope of ’patent reform’,” Promote the Progress, 22 September2006, http://promotetheprogress.com/archives/2006/09/friday food forthought the changing scope of patent reform.html.

101This section is primarily based on personal interviews with various Congres-sional staffers working in some capacity on the Senate Subcommittee on Intellec-tual Property.

102 Mancur Olson Jr., The Logic of Collective Action: Public Goods and theTheory of Groups Harvard University Press (1965).

39

one would expect that the Court is currently seeking a new test casein which to discuss the question of patentable subject matter. Espe-cially if the question is the patentability of designs such as a stockcomputer on which is loaded software, potential test cases abound.

All the pieces fit together neatly: patent law, the only IP law whereindependent invention is not a valid defense,103 is a terrible match tomassively decentralized pseudo-industries like software authorship orbusiness methods, where there are so many players that independentinvention is basically inevitable. But the reason that these pseudo-industries are so decentralized is that they handle pure information,with a trivial after-step to apply the information to human affairs.There is a history of court rulings that made it clear that pure infor-mation processing is not patentable, even when a patent draftsmanadds “insignificant postsolution activity” to apply the information toreal-world affairs. Thus, this judicial line distinguishing the patent-able from the unpatentable exactly matches the ideal economic linethat divides traditional industries that prospered with patents fromthe massively decentralized information-based industries that haveprospered without patents.

103Fortunately for all involved, there is a type of intellectual property protectionthat covers information and yet does allow independent invention as a valid de-fense: copyright. As I discuss elsewhere,104 an appropriately interpreted copyrightlaw provides exactly the level of protection needed by software algorithms.

104Ben Klemens, Math You Can’t Use: Patents, Copyright, and Software, Brook-ings Institution Press, 2005.

40