Software patents, part II.

In my last post, I promised a post about what I think needs changing in the patent system. This necessarily gets considerably more detailed, so it's likely to be harder to apply it to different patent systems. Nonetheless, when people talk about problems with software patents, the US seems to be (by far) the most often cited culprit, so I doubt that being specific to it is really a major problem. Anyway, on with the show.

The first problem I see is that the US patent office is basically run as a business -- it's profitable to the government, and most of the profit comes from issued patents. Worse, what's really intended to be the primary job of the patent office (searching through prior art to figure out whether something is really original) is apparently run at a loss. The profit comes from "maintenance fees" on patents after they're issued.

The patent office has suggested a way to try to streamline the patent examination process that would involve the inventor either citing only a few sources of related art, or (if they know of too many) specifically pointing to those they think are most relevant, including specific citations of the parts they consider relevant.

At least to me, this seems utterly insane. First of all, it's simply attempting to shift the burden of the patent office's job from the patent office to the inventor. Second, it's asking the inventor to predict what the patent examiner will find the most useful. If, for example, the patent examiner looks at some other reference and finds it more relevant than the ones the applicant pointed to as most relevant, would the patent become invalid? If so, the patent system basically becomes a lottery. But, if there's no penalty, what's to prevent an applicant from citing something utterly irrelevant as the most relevant art related to his patent?

The simple fact is fairly simple: as it stands right now, the patent office simply isn't complying with the requirements of patent law. The law says the fees they charge should cover the cost of examining the patent. If their fees aren't high enough to do that, then the fees need to be raised. Another point that's routinely cited is that an examiner has only about 20 hours to examine a given application (there's some disagreement about the exact number, but at least it's at what I'd consider the noise level -- there's little question that it's between  15 and 25 in any case). Another common complaint is that patent examiners simply aren't competent -- that they often just don't know enough about the technology to sort out (for one example) truly new techniques from simply new and different terminology for existing techniques.

These have a simple solution though: simply increase the fees for a patent application to the point that they can/will cover the costs of hiring competent examiners, and giving them enough time to do a good job. Along with funding the process properly, higher fees, all by themselves, would probably do quite a bit to encourage companies to look more carefully at their applications before filing, to ensure that what they're filing is truly novel and useful, before wasting others' time finding prior art to show that it's invalid.

The second major problem I see is that it's much easier to get a patent granted than to get that same patent invalidated after it's been granted. The patent office has been taking some steps that attempt to balance this a bit better, but at least personally, I don't think they've done quite enough yet.

Classically, the situation was that the inventor filed his application, and the patent examiner (usually only one, though sometimes two) tried to find prior art. Under most circumstances, almost nobody else participated in the process at all. The patent office does now publish (most) patent applications, so normal people can review them, but it's still basically provided as a one-way communication -- if you look at an application on the patent office's web site, there's not (for example) a button on the page to let you send an email to that examiner citing what you think is prior art on that patent application. Right now, however, instead of treating prior art submissions as what they really are (normal people doing them a favor to make their job easier) they seem to treat it as if they're doing me a big favor by allowing me to submit anything at all, and it's perfectly reasonable to make me jump through flaming hoops before I can be granted such a privilege.

Even so, the patent office follows a "preponderance of the evidence" rule in deciding whether to issue the patent. That basically means only a simple majority of the evidence needs to point toward its being novel for the application to be accepted, and the patent to issue. If 51% of the evidence says it's original, and 49% says it's not, the patent can be issued. Keep in mind that in most cases, it's up to one person to find any and all evidence of invalidity (in roughly 20 hours of work or less), but a large company may easily have a dozen people working hundreds of hours (if necessary) to get the patent issued, and it's quickly apparent that the system is heavily loaded in favor of issuing the patent unless it's quite obviously bad (and sometimes even if it is quite obviously bad).

Once the patent issues, however, the courts are required to give the patent a presumption of validity. To invalidate the patent, you don't simply need to provide a preponderance of evidence in the other direction. It's not enough at that point to show that 51% of the evidence points toward invalidity, and only 49% toward validity -- instead, the courts require "clear and convincing" evidence that the patent is invalid before they will declare it invalid. I've never seen a court attempt to specify a percentage of the evidence necessary to meet the clear and convincing hurdle, but it's clearly a lot higher than a mere preponderance of the evidence. If I had to characterize it, I'd probably put the number at around 90% -- still somewhat short of the "beyond a reasonable doubt" (that's supposed to be) used in criminal courts, but much higher than than 51% used as the basis for issuing the patent in the first place.

We end up with a system that deprives most people of much ability to provide evidence that a patent application isn't valid, issuing the patent even when/if there's a substantial chance that it's not valid, and maintaining it as valid even when most of the evidence indicates that it's really not.

Though I don't think those are (probably) the only two problems with the patent system as it stands today, I think addressing those two points would go a long ways toward redressing much of the current imbalance in the patent system. Interestingly, neither of these requires any change to the actual patent laws at all, only changes in procedures about how the existing law is enforced. Over the past several years I've seen many proposals that included what initially seem like much larger changes (nearly all including substantial changes to the patent law itself). I believe most of these would do little to cure existing problems, and quite a few would add many more problems of their own as well.

From a programmer's point of view, most of the proposals strike me as doing nothing to try to find or fix actual bugs, and instead advocate changing indentation and naming conventions throughout the code. They'll change how the code looks, and might (conceivably) make it more readable, to at least some degree, but they're almost certain to introduce at least a few new bugs in the editing, but fail (except, perhaps incidentally) to fix existing bugs. Just as with code, to do real good, you need to identify the real problems, prioritize them, and (usually) attempt to make the smallest changes that fix those problems.

I'd also note that almost none of this is specific to software patents at all. Most of the problems I see apply at least equally to areas outside of software. As noted in my previous post on the subject, I think most of the arguments attempting to separate software patents from other patents lack merit.

Why most programmers are wrong about software patents.

Many (probably most) computer programmers favor elimination of patents that apply to software. I believe, as is often the case, that they're well on their way to repeating a dark history they've blithely ignored.

Patents were invented as a service to society -- the basic intent was (and remains) fairly simple: give an inventor exclusive rights to his/her invention for some limited period of time in exchange for agreeing that as soon as that limited period of time expires, the invention goes into the public domain -- anybody can freely use it. By itself, that wouldn't mean much: after all, if the patent didn't exist to start with, anybody would be able to make free use of the invention immediately, instead of waiting for the patent to expire.

The benefit to society comes from another requirement for a patent: that the patent describe the invention so others can use it. The exact phrasing varies by country, but in the US the requirement is that the patent contain a description of the invention sufficient for a person of ordinary skill in the art to implement the patented invention.

Going a step beyond that, US patent law also includes a "best mode" clause that says the inventor must reveal what he believes to be the best method of implementing/deploying/using the invention. For example, if he's thought up a new type of oil for lubricating automobile engines, he can't patent it, but in the patent claim that it's intended to be used as cooking oil (which could deprive society of the benefit, because they ignore it after finding that it makes food taste terrible). Obviously the example of a cooking vs. lubricating oil is a bit extreme, and probably easily seen through -- most people would probably think something was a bit odd about Exxon patenting what they claimed was a cooking oil, for example.

Anyway, let's take at least a quick glance at that dark history they're ignoring. Before patents were invented, most crafts were governed by guilds. To learn how to carry out a particular craft, you had to be admitted into the proper guild. The guild guarded the secrets of its craft quite carefully in most cases -- to the point of sometimes murdering people who were suspected of revealing guild secrets, and refusing admittance to others because they were seen as too dangerous. In other cases, there were literally wars between competing guilds. These not only get people killed, but in some cases when a guild was destroyed, its secrets died with it. Archaeologists have found artifacts showing techniques we don't quite know how to duplicate to this day.

As many problems as they had, guilds were a substantial improvement over the previous system: one of individual craftsmen keeping secrets entirely to themselves, to be passed onto their children, sometimes on their deathbed -- or taking their secrets to the grave, if their children weren't handy at the right time, or were estranged, etc.

Most arguments programmers level against this fall into three broad groups. The first is that programming techniques (algorithms, protocols, etc.) get published even when there are no patents. The second is rarely quite so clearly articulated, but basically comes down to the notion that if one person invented something, they can invent something at least equally good when and if they need to. The third is that software moves so quickly that by the time a patent has expired, it's no longer of any value.

The first is partly true. They can point, for example, to the fact that European companies continue to publish despite the European policy against software patents. This, however, ignores reality. What's happening right now is that Europe is basically "freeloading" off of the patent systems of other countries (US, Japan, etc.) If, for example, Seimens or Philips wants a patent in the US and/or Japan, they need to publish all the details of that invention in their US/Japanese patent application. That means the invention is no longer secret -- and given current communications, the minor detail that it's published in the US/Japan instead of somewhere in Europe adds (at most) a few milliseconds for a European citizen to download the patent application. As such, Europeans enjoy the benefit of inventions being published, without (themselves) giving the inventor anything in return.

The second strikes me as simple conceit. Yes, there are patents running the gamut from utter stupidity to simply being something nobody else thought was important or original enough to bother patenting. Being at all honest, those constitute only a fairly small percentage of patents overall though. Much of this stems from a problem I've seen many times: somebody will draw a badly mistaken conclusion about what the patent covers, often based on looking at nothing other than the patent's title. They see a title that says something like "A method for run-length compression", and they think something like "good lord what insanity -- everybody's known about RLE for ages -- obviously the patent office is a bunch of idiots!" Though it's barely possible they're (at least sort of) correct, and this really is a patent on something everybody's known about for decades, chances are it's something else entirely -- some (possibly minor) improvement that allows the job to be done a little faster, or using a little less memory, etc. In a typical case, it's not earth-shaking, but is original and at least mildly useful.

The third argument may sometimes have some degree of validity as well, but often doesn't. This argument, however, ignores that fact that if the invention becomes obsolete quickly, the cost of the patent (to the people) is also low -- if an invention is only meaningful for a year or two, the cost to society of the fact that an inventor (theoretically) gets exclusive use of it for years afterwards costs society essentially nothing. At the same time, people have invented and patented algorithms that retain value well after the patent expires. An obvious case in point would be RSA encryption. The US patent on RSA expired well over a decade ago, but it's still almost certainly the most widely used from of PK cryptography. RSA remains (roughly) as valuable today as ever, although there are alternatives that could be used if RSA were not available.

Now, don't get me wrong: I don't think the US patent system (or probably any other) is anywhere close to perfect. Some time soon, I'll write a post about what I see as problems in the system, and at least some ideas about how to fix those problems. To give fair warning: I believe many of the problems can be fixed with much less drastic changes than most people seem to think are necessary. In fact, some of the changes are (or at least initially seem) so minor that they're probably almost invisible except to attorneys who specialize in intellectual property law. Nonetheless, I think some seemingly trivial modifications can have fairly serious consequences.