The confusion surrounding patent eligibility is not surprising when one takes a moment to consider some of the individuals and communities contributing to the debate—and how their frameworks for analysis are simply inadequate to address the real world issues inherent in this topic. Some economists, like our friend above, are often awfully good at tallying up things, but sometimes not so good at recognizing what they are talking about—like knowing the difference between what’s patentable (sheep) and what’s not (dogs). Gary Becker’s recent Becker-Posner Blog post of suggestions for reforming the U.S. patent system is another example of misguided, reductionist logic common to econometricians. In reducing the complex issues at hand to a simple cost-benefit analysis, he displays a woefully misinformed understanding of the actual nature of patents and the mechanics of innovation.
Arguing that the current system is “too broad, too loose, and too expensive,” Becker grounds his criticism in the view that patent litigation is “expensive and time-consuming” and that “many individuals and small companies do not have enough resources … to litigate against bigger companies that infringe their patents.” As a result, such companies sell their patents to “patent assertion entities” leading to “excessive litigation.” To stem this tide, Becker suggests “narrowing the types of innovations that can be patented.” “The first step,” Becker says,
is to recognize that many innovations presently cannot be patented. A major example is the inability to patent scientific theories and concepts, such as Einstein’s theory of relativity, Darwin’s theory of evolution, or Keynes’ model of the macro economy. The presumption in excluding basic scientific knowledge from the patent system is that the cost of restricting open access to such discoveries far exceeds any gains in encouraging the development of scientific concepts through granting temporary monopolies to the creators. To offset the effect of this exclusion from patent protection on the incentive to discover, individuals and governments have created prizes and awards, like the Fields Medal and Nobel Prizes, for mathematical and scientific discoveries.
Becker correctly points out that scientific theories etc. are not eligible for patents, but immediately frames this in the familiar cost-benefit analysis: We exclude scientific theories, etc. from patenting because "the cost of restricting open access to such discoveries far exceeds any gains." Becker then uses this cost-benefit analysis to draw a bead on his real target: software patents. Becker cleverly extends his “exclusion” argument by next suggesting that “one can start by eliminating the patenting of software” because “disputes over software patents are among the most common, expensive, and counterproductive.” Becker believes that while eliminating patent protection for software “would discourage some software innovations … the saving from litigation costs over disputed patent rights would more than compensate the economy for that cost.”
What is wrong with this? Let me count the ways:
First, all patent litigation, not just software patent litigation, is expensive. The problem of patent litigation costs arises from the intersection of technology with hyper-complex rules of claim construction, infringement and validity. “This type of litigation is typically very expensive for several reasons: (1) heavy reliance upon expert witness testimony during trial; (2) extensive discovery of documents; (3) a significant number of pretrial depositions; and (4) the inevitable disputes regarding the discovery process that must be resolved by the court.”
Second, Becker asserts that the cost savings from reduced patent litigation will “more than compensate” for the cost of reduced innovation. In this regard, Becker is likely relying on the research of James Bessen and Micheal Meurer, who have argued that the NPE litigation imposed $29B in direct costs on U.S. companies in 2011. Bessen and Meurer place the blame most heavily on software patents, arguing that 62% of the patents involved in NPE litigation in 2011 were software patents.
The problem here is that Bessen’s analysis has been shown to be faulty for a number of reasons, including that it is based on a biased sample of patent litigation cases. Jay Kesan and David Schwartz point out that Bessen and Meurer relied on just 82 responses to a survey of RPX Corporation of “about 250 companies,” including “RPX clients and nonclient companies with whom RPX has relationships.” This is not a properly designed random survey and is suspect precisely because RPX is in the business of soliciting enterprises that are frequent targets of patent troll litigation to purchase a subscription to RPX’s defensive portfolio. As Kesan and Schwartz note “It seems extremely likely that RPX’s clients have experienced high litigation costs, perhaps much higher than the average company. As such, the survey has a strong selection bias here in favor of companies that are repeat defendants in NPE litigation and thus need the services of RPX to reduce future patent liabilities.” Further, given that the data was based on just the 82 companies that responded to the survey, there is an inherent selection bias. Kesan and Schwartz argue:
Without more information, a reasonable assumption is that the responding companies likely had easier access to the information (i.e., better electronic record keeping), which likely means larger companies, and/or were more motivated to respond (i.e., they have higher exposure and costs). Thus, it is very likely that there were selection effects on multiple levels: the solicited companies had higher costs and expenses than the average company, and the responding companies had higher costs and transfers than the universe of companies solicited. Our view is supported by Bessen & Meurer’s disclosure that 72% of the 82 respondents are publicly-traded companies, while only 14% of all NPE defendants are.
Kesan and Schwartz conclude that the $29B figure is at best “the highest possible limit” and that “the true number is very likely to be substantially lower.”
The problems in Bessen and Meurer’s research run deeper than simply bad methodology. They reflect either a complete naiveté or an irrational belief in rationality. To estimate the costs to public companies from NPE litigation, Bessen and Meurer look to the drop in the stock prices of about 4,000 companies within a 25-day window following the filing of a patent lawsuit. They assert that such NPE patent litigation resulted in a “mean wealth lost per lawsuit” of $122M. The grand assumption here is the rational market hypothesis, that investors can accurately assess both the outcome of the patent litigation and its impact on the firm’s profitability—all within a 25-day window after the litigation is filed. A lovely and standard economist assumption, but not very realistic. Bessen and Meurer offer this unabashed and utterly navel-gazing justification: “While we accept the idea that investors do not always act rationally, we have found no explanation consistent with the evidence for why investors should persistently overreact to lawsuit filings.” They can’t think of an explanation of why investors would dump the stocks of companies that have just been sued? Apparently Bessen and Meurer don’t know the second rule of the market: Buy the rumor, sell on the news. Sophisticated patent attorneys can’t predict the outcome of a patent litigation, years into the future, within 25 days of its filing—especially not with a 50% reversal rate by the Federal Circuit. Apparently the mythical, perfect, omniscient investor can. Right.
The evidence that eliminating software patents would encourage innovation is at best mixed, and too broad to cover here in detail. Most of the research is directed to the patent system as a whole and the question of whether patents generally increase the rate of innovation. And studies on the impact of software patents on innovation support different interpretations. Thus, Cockburn and MacGarvie find that software patents both inhibit and promote market entry: “we find that a 10% increase in the number of patents relevant to market reduces the rate of entry by 3-8%” but “importantly, these negative effects on entry are mitigated when entrants come to market with their own patents: firms that have filed applications for patents relevant to a market are approximately twice as likely to enter as otherwise similar firms.”
Third, there is no principled distinction between software as an artifact of human creativity and any other form of innovation. Humans create artifacts, artificial products, to serve functional (and sometimes aesthetic) needs. Material artifacts are inherently physical, such as pencils and bridges, microprocessors and forks; immaterial artifacts are essentially symbolic, such as building plans and music, bank accounts and computer programs. Technology is the purposeful design and implementation of artifacts to address human needs. More deeply, technology is the result of the transmutation of subjective Ideas into objective Artifacts. Software is just one form of this transmutation. Like other artifacts, it begins with a functional need, a problem to be solved, leads to an idea of a solution—and sometimes an invention—and finally to an implementation in programming code. How we create computer programs is no different in kind from how we create any other artifact.
Further, there is no principled distinction between hardware and software. It is well known that anything that can be done in software can be performed by hardware circuit and vice versa; this is known as the Church-Turing thesis. Indeed, it is standard practice today to develop hardware by first defining the functional operations of the hardware using software languages, and then converting the software description to a hardware specification. This is known as electronic design automation. The vast majority of hardware, from smartphones to TVs to the computers that run cars, is designed using EDA. The boundaries between hardware and software implementations are highly permeable, and Becker would have to eliminate hardware patents as well.
But, above all else, Becker’s lack of understanding of software and technology in general is demonstrated by his assertion that "creators of new software may then try to protect their innovations through secrecy."
The vast majority of value in software comes not from some deeply embedded algorithm that can be protected by trade secret. Rather, it comes from the creation of new functionality that has immediate and apparent value to the end user, whether that's a consumer or an enterprise. For example, in consumer or professional software, it’s features that matter and features that sell: what the end user can do, how quickly and efficiently they can communicate, create, edit, search, identify, manage, etc. New and improved ways of providing these functions, as well as new functions, are almost by definition “visible” to the end user—and to the competitor. It's just as much an invention to enable users to use their smartphone in a new way or a new application as it is to invent a new algorithm realistically animating the way waves break at the beach (i.e., an algorithm the end user cannot see or determine from examination of the result). This is true of new machines as well: as soon as a product leaves the manufacturer it can easily be reverse engineered and copied. When an automaker sells a car with a new safety or performance feature, even if the consumer has no idea about its mechanical design, it’s no longer a trade secret, because a competitor can buy the car, tear it down and figure out how the new feature works. A modern technology company with a consumer-facing product or software similarly cannot rely on trade secret protection for some of the most valuable, market differentiating aspects of its technology: if they were smart enough to figure out the solution to a problem, their competition is likely to be just as smart.
Indeed, not only does being the first mover not necessarily provide a sustainable advantage, in the absence of intellectual property protection, it directly contributes to the “second mover advantage.” Indeed, as an economist, Becker should know that,
Secondary or late movers to an industry or market, have the ability to study the first movers and their techniques and strategies. “Late movers may be able to ‘free-ride’ on a pioneering firm’s investments in a number of areas including R&D, buyer education, and infrastructure development.” The basic principle of this effect is that the competition is allowed to benefit and not incur the costs which the first mover has to sustain. These “imitation costs” are much lower than the “innovation costs” the first mover had to spend, and also can cut into the profits which the pioneering firm would otherwise be enjoying.
Amazon.com is considered a classic case of a “second mover advantage,” coming second into the market for online book sales three years after the pioneering sites of BookStacks and books.com. Anyone remember them?
Finally, Becker, up in the Ivory Tower of Economica, apparently has no idea that trade secret misappropriation is against the law: “Even when their secrets might be learned and sold by employees, companies that innovate could gain because they then would pay less to employees in a position to profit from learning about these innovations by selling the information to competitors.”
Yes, that’s right: Becker suggests that companies like Microsoft, Oracle, Google, Adobe, etc. can “pay less to employees” who could supplement their income by selling the companies’ trade secrets to competitors. No comment.
Becker’s arguments reveal a deep lack of knowledge of the real world: of real patent law, of real software technology, and of real employment practices. Software patents “are a problem” which must be solved, and for an economist, the solution to any problem is to assume a can opener.
My original posting erroneously suggested that Becker "believes (or at least argues) that, under current patent law, 'scientific theories and concepts' can be patented and that society would be better off by their 'exclusion from patent protection.'" This error was graciously pointed out by my readers, and I have revised the text accordingly. The core thrust of my argument remains intact.