[an error occurred while processing this directive]

This article by Brian Kahin appears in the April 1990 issue of Technology Review (Building W59, MIT, Cambridge MA 02139, (617)253-8250). It may be copied for noncommercial purposes provided that it is copied, along with this statement and the bio at the end of the article, without any modification whatsoever. (Copyright (C) 1990 by Brian Kahin)

The Software Patent Crisis

An explosion of patents on software processes may radically change the programming industry—and our concept of human expression in the computer age.

Last August, Refac International, Ltd., sued six major spreadsheet publishers, including Lotus, Microsoft, and Ashton-Tate, claiming they had infringed on U.S. Patent No. 4,398,249. The patent deals with a technique called "natural order recalc," a common feature of spreadsheet programs that allows a change in one calculation to reverberate throughout a document. Refac itself does not have a spreadsheet program and is not even in the software industry. Its business is acquiring, licensing, and litigating patents.

Within the last few years, software developers have been surprised to learn that hundreds, even thousands, of patents have been awarded for programming processes ranging from sequences of machine instructions to features of the user interface. Many of the patents cover processes that seem conventional or obvious, and developers now fear that any of the thousands of individual processes in their programs may be subject to patent-infringement claims.

The Refac suit demonstrates the vulnerability of the industry to such claims. Patent no. 4,398,249 was applied for in 1970, granted in 1983, and only recently acquired by Refac. In the meantime, software developers have been busily creating spreadsheets and other new products unmindful of patents. The industry accepted copyright and trade secret as adequate protection for its products, and most programmers assumed that patents were not generally available for software.

Never before has an industry in which copyright was widely established suddenly been subjected to patenting.

Never before has an industry in which copyright was widely established suddenly been subjected to patenting. As it is, only a few companies that create microcomputer software have the resources to try to defend against patent infringement claims. Most small firms will be forced to pay license fees rather than contest the claims, even though many software patents may not stand up in court.

In the long run, the costs of doing business in a patent environment will radically restructure the industry. Many small companies will fold under the costs of licensing, avoiding patent infringement, and pursuing patents defensively. The individual software entrepreneur and inventor may all but disappear. There will be fewer publishers and fewer products, and the price of software will rise to reflect the costs.

Especially disturbing is that the broad claims of many recent software patents appear to establish monopolies on the automation of such common functions as generating footnotes and comparing documents. Some claims even cover processes for presenting and communicating information, raising troubling questions about the effect of patents on the future of computer-mediated expression.

Patent vs. Copyright

Software patents, like all patents, give an inventor the right to exclude all others from making, selling, or using an invention for 17 years. In return, the patentee discloses his or her "best method" of implementing the invention, thereby relinquishing trade secrets that might otherwise be enforced forever (like the formula for Coca-Cola).

To obtain a patent, an applicant must convince Patent Office examiners that the invention would not be obvious to a "person of ordinary skill in the art" who is familiar with all the "prior art," which includes previous patents and publications. In contrast, copyright inheres in books, poems, music, and other works of authorship, including computer programs, from the moment they are created. Registering one's work with the Copyright Office is a simple, inexpensive procedure that has important benefits (it is a precondition for filing suit, for example), but the copyright itself is automatic when the work is fixed on paper or on disk.

Copyright and patent protect different things. Copyright protects expression but not underlying ideas. Patents protect useful processes, machines, and compositions of matter. Traditionally "processes" have included methods of physically transforming materials but not business methods or mental steps. Thus, computer programs fall somewhere between the traditional territories of copyright and patent.

From the 1960s to the early 1980s, the Patent Office and the courts grappled with the question of whether algorithms—the elemental processes on which computer programs are built—are patentable as either processes or machines. Early on, the Patent Office granted some patents for processes built into computer hardware that today would be contained in software, but it was reluctant to grant patents for programs per se. As the 1966 Report of the President's Commission on the Patent System pointed out, the Patent Office had no system for classifying programs. The report also noted that even if this were remedied, the volume of programs being created was so enormous that reliable searches of "prior art" would not be feasible or economical.

However, the Court of Customs and Patent Appeals (CCPA) maintained that computer programs were patentable and overturned numerous Patent Office decisions denying patentability. The Supreme Court vindicated the Patent Office in two decisions, Gottschalk v. Benson (1972) and Parker v. Flook (1978), holding that mathematical algorithms were not patentable subject matter. Still, the CCPA continued to uphold patentability in other cases. Finally, in Diamond v. Diehr (1981), a sharply divided Supreme Court upheld the patentability of a process for curing rubber that included a computer program. The majority concluded that programs that did not preempt all uses of a computer algorithm could be patented—at least when used in a traditional process for physically transforming materials.

That case has been the Supreme Court's last word on the subject. But despite the narrowness of the ruling, the Patent Office underwent a radical change of heart. Until very recently, there were no reported appeals of adverse Patent Office decisions, leading observers to conclude that the office was eventually granting almost all applications for software patents. Although articles began appearing in legal periodicals a few years ago noting that patents were being routinely granted for many software processes, not until 1988 did the industry realize that the rules were changing, or had already changed, in the middle of the game. By the spring of 1989, the patents that entered the pipeline after Diamond v. Diehr were starting to flow out in significant numbers—by one count, nearly 200 in the first four months of that year.

Processing Problems

Unfortunately, the Patent Office classification system remains unchanged, and the volume of software being created has grown exponentially. This makes searching for prior art—processes already in public use—time-consuming and expensive.

The search is extraordinarily difficult because the field's printed literature is thin and unorganized. Software documents its own design, in contrast to physical processes, which require written documentation. Also, software is usually distributed without source code under licenses that forbid reverse engineering. This may amount to suppressing or concealing the invention and therefore prevent the program from qualifying as prior art. The search for prior art may require securing oral testimony from people who developed software at universities many years ago, an expensive proposition.

Many programmers suspect that patent examiners lack knowledge of the field, especially since the Patent Office does not accept computer science as a qualifying degree for patent practice (it accepts degrees in electrical engineering). Moreover, attracting and holding individuals with expertise in a field like software, where industry demand is high, is not easy for a government agency. Less qualified examiners create problems because they naturally have a lower standard in determining the hypothetical "person having ordinary skill in the art," and are thus more apt to grant patents for obvious processes. Since the examination process is conducted ex parte (as a private matter between the Patent Office and the applicant), less qualified personnel are also more likely to be influenced by sophisticated patent attorneys and the apparent expertise of the applicant.

The quality of software patents being awarded has aroused concern even among patent lawyers and other advocates of the new regime. But it will be left to firms being sued for infringement to prove that a process should not have been patented because it was obvious in view of the prior art. Meanwhile, software patents stand as intimidating weapons for those who hold them.

Restructuring the Industry

Perhaps because of some of these problems, applications for software patents take an average of 32 months to be approved and published. That's significantly longer than the overall average of 20 months, and a very long time given the short product cycles of the software business.

Unlike copyright, independent creation is irrelevant to patent infringement. Every developer is charged with knowledge of all patents. Even if someone is not aware of a patent, he or she can still infringe against it. Furthermore, patent applications and the examination process are confidential, so there are ordinarily several years of patents in the pipeline that no search will reveal. Although no infringement occurs until the patent issues, an inventor may find that a newly awarded patent covers a feature he or she has already incorporated and marketed in a finished product. While this is a problem for the patent system as a whole, it is intolerable for software developers because of the industry's rapid pace of innovation and long patent-processing period.

The problem is compounded by the fact that a modern software package may contain thousands of separately patentable processes, each of which adds to the risk of infringing patents that are already in the pipeline. Since software functions are interdependent and must be carefully integrated, developers can find it difficult to excise a process built into the original program.

The patent system exacts a high penalty in an industry as decentralized as software.

The patent system exacts a high penalty in an industry as decentralized as software. Programming requires no special materials, facilities, or tools: to design software is to build it. Because barriers to entry are low, the industry attracts many small players, including hundreds of thousands of individuals who work as consultants or short-term employees. Rather than a handful of competitors working on the same problem, there are likely to be dozens, hundreds, even thousands. Since under the patent system one winner takes all, many others—including developers without lawyers—are deprived of the fruits of their independent labor and investments.

Patent proponents argue that this uninhibited duplication of effort wastes resources. But the "waste" could be cut only by reducing the number of players and slowing the pace of development to fit the cycles of the patent system. The result would be a handful of giants competing on a global scale, bidding for the ideas and loyalty of inventive individuals.

However, many programmers believe that there are diseconomies of scale in software development—that the best programs are authored rather than assembled. The success of Visicalc, Lotus 1-2-3, WordPerfect, and other classic programs testifies to the genius of individuals and small teams. Certainly there has been no evidence that they need more incentives. Quite the contrary, the freewheeling U.S. software industry has been a model of creative enterprise.

A Costly System

Even software developers and publishers who do not wish to patent their products must bear the costs of operating under a patent system. While these costs may initially come out of the software industry's operating margins, in the long run, they will be borne by users.

At the first level is the expense of analyzing prior art to avoid patent infringement. A precautionary search and report by outside patent counsel can run about $2,000—that's per process, not per program.

Next are the direct costs of the patent monopoly—the license fees that must be paid to patent holders. If the patent holder refuses to license at a reasonable fee, developers must design around the patent, if that is possible. Otherwise, they must reconceive or even abandon the product.

The third set of costs are those incurred in filing for patents. Searching for prior art, plus preparing, filing, negotiating, and maintaining a patent, can total $10,000 to $25,000, not including internal staff time. Seeking foreign patents can make the bill substantially higher.

The notoriously high costs of patent litigation must be borne by both sides. Just the discovery phase of a lawsuit is likely to cost each side a minimum of $150,000, and a full trial can cost each from $250,000 to millions. Again, these figures do not include internal staff time, which could easily double the real cost. While a small patent holder may be able to secure a law firm on a contingency basis or sell an interest in the patent to speculators, the defendant has no such options.

Litigation also involves the possibility and further expense of an appeal. All appealed patent cases now go directly to the Court of Appeals for the Federal Circuit (CAFC, successor to the CCPA), where panels in patent cases are usually led by patent lawyers turned judges. Whereas patents once fared poorly on appeal, the CAFC has found patents to be both valid and infringed in over 60 percent of the cases that have come before it. The CAFC has greatly strengthened the presumption of patent validity and upheld royalties ranging from 5 to 33 percent.

While a large software company may be able to absorb these costs, they will disproportionately burden smaller companies. The first to suffer will be independent developers who cannot afford to market their own products. These developers typically receive royalties of 10 to 15 percent from publishers who serve as their distributors. Such modest margins, out of which developers must recoup their own costs, would be wiped out by the need to pay royalties to a few patent holders.

The high costs of a patent environment give patentees considerable leverage over small firms who will, as a practical necessity, pay a license fee rather than contest a dubious claim. To establish credibility, the patentee will settle for small fees from the initial licensees. The patent holder can then move on to confront other small firms, pointing to such licensings as acknowledgments of the patent's validity and power. This tactic has a snowballing effect that can give the patent holder the momentum and resources to take on larger companies.

Cross-licensing—where firms secure patents to trade for the right to other patents—seems to work reasonably well in many industries and has been touted as the answer to these problems. However, cross-licensing is of little value to smaller companies, which have little to bring to the table. And cross-licensing may prove of limited value even to large companies, since it does not protect against companies like Refac that have no interest in producing software and therefore no need to cross-license.

Of course, the power that software patents afford may induce some venture capitalists to invest in them. But investing in software patents is one thing; investing in robust, complex products for a mass market is another.

In fact, software publishers hold very few patents. The vast majority are held by large hardware companies, computer manufacturers that have in-house patent counsel and considerable experience in patenting and cross-licensing. Nearly 40 percent of the software patents that the U.S. Patent and Trademark Office now issues go to Japanese hardware companies. It is quite possible that the separate software publishing industry may cease to exist as companies find that they need the patent portfolios and legal resources that the hardware giants can provide. The result will be a loss of diversity in software products, reduced competition, and, many believe, a less productive software industry.

Protecting Ideas and Information

A deeper, more disturbing problem in patenting programs was barely evident before computers became ubiquitous personal tools and software became infinitely versatile. More than a "universal machine," the computer has developed into a medium for human expression and a mediator of human experience. Software is designed to satisfy specific needs for shaping and delivering information. Thus, what is increasingly at stake in software patents is the generation and flow of information. This becomes more threatening when the claims in a patent extend far beyond the disclosed means of implementation to cover general ideas.

Broad patent claims covering abstract processes are not limited to software, or even to computer hardware. Consider patent no. 4,170,832, granted in 1979 for an "interactive teaching machine." The patent discloses a clumsy-looking combined videotape deck and television with a set of push buttons.

The patent includes a process claim for a procedure commonly used in interactive video: showing an introductory video segment, presenting the viewer with a limited number of choices, registering the viewer's decision, and then revealing the likely outcome of that decision. The disclosed machine, which was never marketed, contributes nothing to the public domain: it simply reveals one person's way of implementing a basic instructional technique.

In a notorious 1983 case, a federal district court upheld the patentability of Merrill Lynch's Cash Management Account system, a procedure for moving investment funds among different types of accounts. Acknowledging that the system—essentially a method of doing business—would not be patentable if executed with pencil and paper, the court nevertheless upheld the patent because it made use of a computer.

The Patent Office has taken this principle one step further. Besides granting monopolies on new procedures such as the Cash Management Account system, the office is also awarding patents merely for automating familiar processes such as generating footnotes (patent no. 4,648,067) and comparing documents (patent no. 4,807,182). But software developers have been routinely automating such common office functions, bookkeeping procedures, learning strategies, and modes of human interaction for years. The principle that patents are granted to induce inventors to disclose trade secrets has no relevance here. These processes are part of everyday life, and can and should be computerized in a number of ways.

What's more, information per se is traditionally the substance and territory of copyright. The intelligent ordering of information is the very heart of grammar, rhetoric, and graphic design.

Why should information be subject to the pervasive restraints of patent simply because it is interactive rather than linear? Should human expression that is assembled, communicated, or assimilated with the aid of a computer be restrained by patents? If the computer is seen as an extension of the human mind rather than vice versa, the answer is no.

Changing Patent Policy

Software developers who understand the impact of patents are demoralized. Lawyers assure them that patents are here to stay, and that programmers must seek new patents to protect against other patents. These lawyers point to the growing torrent of software patents, the presumption of patent validity, and the fervidly pro-patent record of the Court of Appeals for the Federal Circuit. Smaller companies that cannot afford this advice can only hope that companies with deeper pockets will afford more visible and attractive targets for patent holders bringing suit.

But the narrowness of the Supreme Court decision in Diamond v. Diehr remains. The Court never explicitly rejected the traditional doctrines against the patentability of mental steps and business methods, doctrines that may yet defeat many of the patents that have issued. If the hue and cry grows, Congress could amend the Patent Act to make it clear that the scope of patenting is still limited to physical processes.

The software industry was not broke, but it is in the process of being "fixed." The question is whether the fixing will be done by the gush of awards from private proceedings in the Patent Office—or by a public decision about whether software patents serve "to promote the Progress of Science and useful Arts," as the Constitution requires.

Brian Kahin is an attorney specializing in information technology and policy. An adjunct research fellow in the Science, Technology and Public Policy Program at Harvard University's Kennedy School of Government, he was formerly affiliated with the MIT Research Program on Communications Policy and the MIT Communications Forum. He is a graduate of Harvard College and Harvard Law School