For submission to the Patent & Trademark Office

by Richard Stallman

Each year the government creates new bureaucratic programs. Each is created for a purpose. That doesn't mean it serves that purpose or that it is worth the cost.

It's hard to close down unnecessary bureaucracies, because people presume that they must do some good. It's easy to admit that a government program has drawbacks, but many won't seriously consider whether it does more harm than good.

Thus we see, in the announcement of these hearings, the supposition that software patents are helpful. We are asked whether they "protect" software developers "enough". Patent lawyers chose the word "protection" to imply that patents are good.

Now commentators feel pressure to use the same prejudiced word, lest they appear unschooled. This does its part to impede consideration of the possibility that patents are actually "too much".

I'm not a lawyer. I am a programmer, and considered a good one. I'm here to explain why software patents impede software development and retard software progress.

Software is like other fields of engineering in many ways. But there is a fundamental difference: computer programs are built out of ideal mathematical objects. A program always does exactly what it says. You can build a castle in the air supported by a line of zero thickness, and it will stay up.

Physical machinery isn't so predictable, because physical objects are quirky. If a program says to count the numbers from one to a thousand, it will do exactly that. If you build the counter out of machinery, a belt might slip and count the number 58 twice, or a truck might go by outside and you'll skip 572. These problems make designing reliable physical machinery very hard.

When we programmers put a while statement inside an if statement, we don't have to worry about whether the while statement will run such a long time that it will burn out the if statement, or that it will rub against the if statement and wear it out. We don't have to worry that it will vibrate at the wrong speed and the if statement will resonate and crack. We don't have to worry about physical replacement of the broken if statement. We don't have to worry about whether the if statement can deliver enough current to the while statement without a voltage drop. There are many problems of hardware design that we don't have to worry about.

The result is that software is far easier to design, per component, than hardware. This is why designers today use software rather than hardware wherever they can. This is also why teams of a few people often develop computer programs of tremendous complexity.

People naively say to me, "If your program is innovative, then won't you get the patent?" This question assumes that one product goes with one patent.

In some fields, such as pharmaceuticals, patents often work that way. Software is at the opposite extreme: a typical patent covers many dissimilar programs and even an innovative program is likely to infringe many patents.

That's because a substantial program must combine a large number of different techniques, and implement many features. Even if a few are new inventions, that still leaves plenty that are not. Each technique or feature less than two decades old is likely to be patented already by someone else. Whether it is actually patented is a matter of luck.

The only way a programmer can avoid this mess is by sticking to things that are obsolete. You may not recall the state of the computer field seventeen years ago, since most people didn't pay attention back then. There were no personal computers. If you were a hobbyist, you might get a computer with a few thousand bytes of memory. With luck, it might run Basic.

This shows another way software is different: it progresses very quickly. A program three years old is becoming obsolete, and one six years old looks stone-age. A twenty-year monopoly for anything in computers is absurd.

When I spoke to the American Bar Association Science and Technology section, the moderator asked the audience whether they saw a need for any software monopoly lasting more than ten years. Most said no. Some of them said, "No, aside from copyright." Some said, "No, not even copyright."

In other fields, a new technique may require development--building one device after another until you understand how to make the technique work. If a steel part works badly, and you think copper might be better, you can't type "replace steel with copper" and try it ten minutes later. The need to recoup this cost of this development is part of the usual argument for patents.

In software, an individual technique usually doesn't need much development. What we develop are products that combine a new technique with dozens of other techniques.

When a second programmer decides to use the same technique, he will have to do just as much development work as the first programmer--firstly, because he's probably using a different combination of techniques, and secondly, because the first programmer probably kept the results of development a trade secret.

The patent system is supposed to help by discouraging trade secrecy. In software, patents don't do this. Today, just as in 1980, most developers publish general ideas and keep the source code secret.

The Kodak Photo CD shows how this can work. The general ideas used in the Photo CD are patented; the details of the algorithm and data format are secret. Even twenty years from now, we will not know how to read a Photo CD unless someone does reverse engineering--and he will probably be sued for copyright infringement.

To manufacture a physical device, you need to set up a factory--a big investment. The need for this investment is one of the arguments for patents in other fields. But mastering a CD-ROM costs just $1000.

Here is a copy of the compiler that I wrote with a few friends. This program is mainly the work of four people. Another dozen helped substantially and others occasionally. The two principal developers were not working on this full-time.

This compiler and its output are probably being used on more than a million computers today. Major companies such as Intel and Motorola have adopted it and now add to it. The US Air Force is funding extensions to it. Many widely used systems are compiled with it.

Just a few lines of code can be enough to infringe a patent. This compiler has ten thousand pages. How many patents does it infringe?

I don't know. Nobody does. Perhaps you can read the code and tell me.

I know of one patent it infringes. I found it--along with some near misses--in a list I saw by luck. I believe I have prior art for that patent, but I can't be sure what a court would say. I don't dare tell you the patent number, because if I were sued, I couldn't pay for the defense. I would lose by default.

An invalid patent is a dangerous weapon. Defending a patent suit typically costs a million dollars, and the outcome depends mostly on legal technicalities.

I've had at least one patentable idea in my career. I know this because someone else patented it years later. It is a feature for using abbreviations in a word processor.

A couple of years ago, the users of the word processor XyWrite received a downgrade in the mail. XyWrite had an abbreviation feature; the developer removed it when threatened by the patent holder.

They knew about my earlier published work; why didn't they fight the patent? Sometimes a business can't afford to have a lawsuit that will drag on for years. At those times, even if the patent is invalid, you lose.

These patents are invalid because of luck. It was pure luck that these ideas were published by one person before they were patented by another. And it was luck that the ones who published didn't patent instead.

This is an important point: what is patented, and what is not, is mainly a matter of luck. When you develop a large system, and you need to combine a large number of techniques and features, whether or not you can use each one is a matter of luck.

The carelessness of the Patent Office in dealing with software is well known. So some people assume that, if the PTO does a better job, everything will be OK. They say that we should wait while the invalid patents spew out, and eventually the PTO will "understand software" and do the job right.

There are two flaws in that suggestion: the PTO will not do a better job, and it would not solve the problem if they did.

Some years ago, a professor I know patented Kirchhoff's current law, which says that the electric currents flowing into a junction equal the currents flowing out. He did this to confirm, privately, his suspicion that the PTO could not handle the field of electronics. He never tried to enforce the patent, which has since expired. (I will disclose his name if you give assurances that he and his lawyer will not get in trouble.)

Kirchhoff's laws were formulated in 1845. If the PTO couldn't understand electricity after a century, how can we expect it to understand software in another decade or two?

Computer scientists look at many software patents and say, "This is absurdly obvious." Defenders of the patent system reject our opinion. "You're using hindsight," they say. "You're more skilled than the typical practitioner." What we consider obvious patents are not errors; they reflect a different definition of "obvious". It won't change.

The most outrageously absurd software patent of all covers using "exclusive or" (an operation of Boolean arithmetic) to modify the image on a bit-map display screen.

This technique is so basic that it was done with just one instruction on the PDP-10 computer I used in the 1970s. I'm told this patent was upheld twice in court.

When the PTO issues two patents for the same technique, as it did for the LZW data compression algorithm, that is surely a mistake. These mistakes can't be avoided because one technique can be described in different ways.

It can take careful thought to see that two verbal descriptions describe the same computation. Mathematicians Atiyah and Singer once noticed that the same computation appeared, with different descriptions, in two branches of mathematics. This led them to the Atiyah-Singer Index Theorem, for which one received a presidential medal and the other a knighthood. The PTO expects an examiner to compare an application with not just one other thing, but everything known to man, in just 17 hours (average). Even if examiners were geniuses like Atiyah and Singer, they couldn't do this.

What if the PTO stopped making mistakes and issued no more invalid patents? That would not solve the problem, because all new techniques and features--those not known today--would be patented just the same.

Suppose the PTO were perfect. Suppose that it is the year 2010 and you are a software developer. You want to write a program combining 200 patentable techniques. Suppose 15 of them are new--you might patent those. Suppose 120 of them were known before 1990--those would not be patented any longer. That leaves 65 techniques probably patented by others--more than enough to make the project infeasible. This is the grid lock we are heading for.

Today's PTO mistakes are bringing us to grid lock sooner, but the ultimate result is grid lock even with a perfect PTO.

I've explained how patents impede progress. Do they also encourage it?

Patents may encourage a few people to look for new ideas to patent. This isn't a big help because we had plenty of innovation without patents. (Look at the journals and advertisements of 1980 and you'll see.) New ideas are not the limiting factor for progress in our field. The hard job in software is developing large systems.

People developing systems have new ideas from time to time. Naturally they use these ideas. Before patents, they published the ideas too, for kudos. As long as we have a lot of software development, we will have a steady flow of new published ideas.

The patent system impedes development. It makes us ask, for each design decision, "Will we get sued?" And the answer is a matter of luck. This leads to more expensive development and less of it.

With less development, programmers will have fewer ideas along the way. Patents can actually reduce the number of patentable ideas that are published.

You will probably hear someone today raise the issue of whether software "inventors" are "entitled" to patents just like other inventors. The answer is simple: like other inventors, they are not entitled to patents at all.

Our Constitution rejects the idea of patents as an entitlement. Patents and copyrights are justified as a restriction on the public only for the sake of the public--only for promoting progress.

Every two years, we elect our Congress. We don't ask who deserves to win, because it's impossible to deserve this, and that's the wrong question anyway. Instead, we ask which Congress the nation deserves.

The same applies to patent policy. We shouldn't ask who deserves a patent, because it's impossible to deserve this, and that's the wrong question anyway. The right question is how the policy affects the public. Do software patents "promote the progress of science and useful arts" as the Constitution requires? Do they promote progress enough to justify their cost--the imposition on the freedom of individuals to use what they know?

A decade ago, the field of software functioned without patents. Without patents, it produced innovations such as windows, virtual reality, spreadsheets, and networks. And because of the absence of patents, programmers could develop software using these innovations.

We did not ask for the change that was imposed on us. There is no doubt that software patents tie us in knots. If there's no clear and vital public need to tie us up in bureaucracy, untie us and let us get back to work!

Responses to the specific questions posed by the PTO:

1. The best definition for a "software patent" is, any patent that restricts the development of new software. Regardless of the form in which this restriction is expressed, the practical effect of such a patent is to create an obstacle to software development.

For the most part, the harmfulness of a patent depends on the substance, not the form, of the patent. More precisely, the extent to which a particular software patent impedes software development depends on the class of programs it prohibits. A patent that covers a broad class of programs, or that covers a vital function of programs, is very harmful. A patent that covers fewer programs, or only obscure programs, is less harmful.

A patent can be written in various ways while still prohibiting the same class of programs. The most extreme example is that a patent written to describe hardware can apply to software implementations, under the doctrine of equivalents.

As a result, new criteria for patentability, based on the sort of distinctions that question 1 asks about, would have little practical effect. Forbidding patents of one form would encourage applicants to use another form, with no change in substance.

Solving the problem of software patents probably requires changing the scope of patents rather than the criteria for patentability. More precisely, the best solution is to exempt software implementations running on general purpose computers from patent law--so that a developer of software for general purpose computers would ipso facto never need to worry about patents.

Nonetheless, here are responses to question 1 as it was asked.

1-A. A patent covering the implementation of a particular algorithm on a general purpose computer obstructs software development and therefore is harmful.

In computer science there is no meaningful distinction between "mathematical" and "non-mathematical" algorithms. Algorithms in general are a branch of mathematics.

1-B. A patent that covers only special purpose hardware implementations is not really a software patent, in that it cannot cause trouble for software development.

Reportedly patents written in this form actually do cover pure software implementations, under the doctrine of equivalents. If this is true, a patent of form B is equivalent in its effect to a patent of form A, and is harmful to software development.

1-C-1. If the program covered by the patent is substantial, the patent would be narrower in breadth than the copyright on the program it covers. This means the patent would be harmless to people writing new software. Hardly anyone would apply for such a patent, since the copyright lasts longer and exists automatically.

If the program patented is very short--perhaps just a few lines--then a patent of this sort might cause trouble for software development. The practical effects of such a patent would depend crucially on the precise interpretation of the patent's scope.

1-C-2. The typical software patent of today has this form. These patents interfere with software development.

A patent of this form covers a class of algorithms--the class of algorithms that have in common the specific steps or elements listed in the patent. It is ironic that patenting many algorithms at once in this fashion is regarded by many as less offensive than patenting a single algorithm.

1-D. I do not see a difference in practical effect between a patent of form D-1 and a patent of form C-1; likewise for D-2 and C-2. While a program can be thought of as an abstract mathematical entity, in practice it cannot be distributed or used unless stored on a physical medium.

1-E. A patent on a data structure is harmful to software development. It also directly harms the users of software; they would have data bases that they cannot read without infringing the patent.

In practice, a patent of form A or C or D can have the same effect as a patent of form E. For example, the patents covering LZW compression cover a computational process (that is to say, a class of similar algorithms); however, no one has found a way to produce the output of LZW except by using the LZW algorithm. The data format that LZW produces is thus, in effect, patented.

1-F. This sort of patent interferes with software development.

1-G. These patents interfere with software development.

2. One of the programs distributed by my project is an interpreter for the programming language PostScript. The revised "level 2" version of PostScript specifies the use of a particular data compression algorithm known as LZW. This algorithm is patented, as a result of which we cannot implement the full PostScript language.

It should be noted that the LZW algorithm has been patented twice. It was developed independently by two groups of people, and there is good reason to believe both groups would have published in academic journals in the absence of software patents.

We know of a superior compression algorithm that we believe is unpatented, and for some purposes we use the superior algorithm. However, no other algorithm can replace LZW for PostScript purposes, because other algorithms produce different data formats and thus do not interoperate with other implementations of PostScript.

The code to implement LZW is approximately 0.2 percent of the PostScript interpreter. The rest of the interpreter is full of other algorithms, data manipulations, and features, any one of which might be patented (and perhaps some of them are).

It was difficult to find any data compression algorithm of reasonable quality that was not covered by a patent. There are dozens of patents in data compression alone. Our first planned replacement for LZW was killed off by a newly issued patent, a week before the program's planned release. After another year and a half, we finally found an algorithm that seems not to be covered by the patents that we know of.

3. The current system is leading to an explosion of patents. Professor Merges of Boston University reports that, based on the amount of software business patent law firms seem to be doing and their interest in hiring lawyers with knowledge of software, the rate of filing of software patents must be accelerating rapidly. Each patent is a land mine for software developers.

Eliminating patents from the field of software would make it easy to develop software again.

4. The present patent system obstructs innovation in the software field. It appears that the appropriate level of monopoly for software-related inventions is none whatever; that level of monopoly produced, through the early 80s, a high rate of that sort of innovation without hampering the development of software.

I have never been involved in a patent lawsuit. I have had to withdraw programs from use because I knew of applicable patents for which I had no prior art--this has happened twice in data compression, as described above.

But main effect that I see is that there are steadily fewer areas of software where I will be able to develop in the future, due to the spread of patents. I forecast this effect when I hear about newly issued patents; I don't need to wait till I myself am threatened with them.

5. No additional form of intellectual property for software is needed. The forms in use in 1980--trade secret and narrow copyright--provided plenty of incentive for commercial software development; there is no public interest in having more.

However, a new form of intellectual property for software could be an improvement if it replaces all the forms that currently exist. Here is a proposal for such a system:

  • When a program is released to the public, the developer is required to place the complete source code in escrow at the Library of Congress, where it is kept temporarily secret.
  • The software is then trade secret and copyrighted for four years. The power of this copyright could be as strict as anyone wishes.
  • Four years later, the program enters the public domain, and the escrowed source code is released by the Library of Congress, also in the public domain.
  • No other form of intellectual property monopoly covers software.

    This form of intellectual property is sufficient to enable most software companies to make a profit, because typically a four-year-old version of a program is obsolete. When version 1 of a program enters the public domain, its developer will be selling version 3 or version 4. Version 1 as public domain software will not be much competition for them.

    A software company that has no new version in four years is doing a bad job. With this proposal, that company might fail; but it is in danger of failing under the present system also. Besides, a badly run business ought to fail; that is a necessary part of free enterprise.

    By contrast, successful software companies come out with substantial improvements every year, and they would be profitable under this proposal.

  • Copyright 1994