*To*: galactus@stack.nl*Cc*: debian-legal@lists.debian.org*Subject*: Re: libdts patent issue?*From*: Nathanael Nerode <neroden@twcny.rr.com>*Date*: Tue, 19 Jul 2005 08:05:46 -0400*Message-id*: <200507190805.46713.neroden@twcny.rr.com>

Arnoud Engelfriet wrote: >I agree with you that the distinction may seem artificial. But it >does seem logical to me to say "you can't patent A XOR B but you can >patent a computer program that does that." If you can patent the class of computer programs which do A XOR B, you have patented the abstract operator which does that. >Then the formula remains >public domain; you just can't make, use or sell a program that >implements the formula. Were the formula patented, then you couldn't >even publish a textbook. Unfortunately, that's a distinction without a difference. If you're prohibited from making a computer program implementing the algorithm, you're prohibited from writing a formal description of the algorithm, which is a standard textbook technique. (A computer program *is* a formal description of an algorithm.) If you're prohibited from selling such a program, you're prohibited from selling such a textbook. The use prohibition is at least different: if only the use prohibition were present, you could indeed publish a textbook, but nobody would be allowed to use its techniques without a license. According to this "distinction", we could distribute Debian as a "computing textbook" rather than as a "system", and we would then be exempt from these patent considerations. (The current US rule is that that every such patent is for a "program plus a generic computer", so this should actually work. Only the silly people who use Debian on their computers who are violating the patent, although Debian might be in trouble for encouraging them. Actually, the users might be fine too, because they're not using the process on an industrial scale. Hey -- there's a solid line of argument: the program itself doesn't violate anything, only the program plus a computer, and it's only combined with the computer at the end-user's house. This is silliness, of course, but that's what you get for making a distinction without a difference.) I am sure ill-informed judges are making this "distinction", but it's something like saying "The ideas are free; it's only thinking them which is illegal". It's complete and utter bullshit. Most likely judges are going by some sort of incompetent gut instinct, and allowing "dead-tree" publication of algorithms while prohibiting online publication. A deterministic algorithm which takes one bit sequence as input and produces another as output is a piece of pure mathematics. And that is *exactly* the sort of thing which is being patented under the name of "software patents". The IEEE magazine (Spectrum) had an article about this recently, which made much the same point: you cannot make a valid distinction of this sort. It went into some detail on why various "distinctions" used by the US courts are hopeless and illusory. The current US excuse is that any mathematical algorithm "plus a generic computer" is patentable. Is that what's being used in Europe too? There's a reason the FFII preferred standard is that the inventive part of a patent must be on some method of manipulating the physical world. That's the last place you can make a distinction where there's a difference. Once you allow patents on abstract algorithms (even if you require the pro forma utterance "plus a generic computer"), you allow patents on large swathes of mathematics, whether you admit it or not. (Not *all* mathematics is algorithms, of course. If European law said that "mathematical methods, except algorithms" were unpatentable, that could be different. But algorithms are quite definitely mathematics. However, there are lossless transformations to convert much of mathematics into algorithms and back.) >It's the same in my eyes as saying "you can't patent a discovery >but you can patent a machine that applies this discovery in >practice." It's not, though. Look again. Now, all algorithms are indeed processes, and most processes can be patented. However, they're purely *mathematical* processes. They must be excluded from patentability if mathematical methods are excluded from patentability. "Mathematica" allows mathematical theorems to be expressed formally (as programs). So according to the theory used currently in the US -- and perhaps in Europe as well -- I can patent a mathematical theorem (plus a generic computer) right now. As it is, all pure mathematics is patentable by law in the US (via the "just add generic computer" method), thanks to the incompetent rulings of the Federal Circuit and others. The basic problem is this: a computer is a machine for doing mathematics. A computer program is a formal description of a piece of mathematics. It is a "method" for computing something, composed of mathematical elements (all the basic bit operations are purely mathematical) -- which means that it is a piece of mathematics. Judges, and apparently you :-), have been fooled by the language of "a method for doing X", which looks like the language for ordinary patents, and haven't noticed that "doing X" is in fact a type of "doing mathematics". I guess "the mathematics is public domain, but any use of it is patented"? >Probably. Still, the EPC mentions "computer programs as such" >and "mathematical methods as such" as two separate categories. They are, in fact, different in a specific technical way. A computer program is a specific formal description of a mathematical algorithm. The patents routinely cover the underlying algorithm, not the formal description of it. In other words, they are patents on mathematical methods as such. The formal description can't be patented due to the computer program rule. The underlying algorithm shouldn't be patentable due to the mathematical methods rule. > but I would not >advise my client to oppose a European patent on the ground that >it's a computer program as such. Again, the grounds should be that it's a *mathematical method*. The "method" or "process" in such a patent is in fact a *mathematical* method. You get mathematicians specializing in fields like theory of computation to testify to that. The inventive step is solely in the new mathematical method (not in the generic computer). Of course, if Europe uses the current totally-broken US standard, under which the "inventive step" doesn't have to be in a patentable area (!), anything goes -- according to that standard, I should be able to patent artwork (combined with a generic computer). I think this standard is obviously wrong, and I think even a non-techie judge would figure that out in Europe, as it renders EPC article 51 section (2) a dead letter (just combine anything on the list with a generic computer). >Not even if I had Knuth as >expert witness. Knuth is the wrong authority. You want the inventors of lambda calculus, combinatory logic, information theory, etc. Well, they're mostly dead, but mathematicians in those fields who can quote the masters, anyway.

**Follow-Ups**:**Re: libdts patent issue?***From:*Arnoud Engelfriet <galactus@stack.nl>

**Re: libdts patent issue?***From:*"Michael K. Edwards" <m.k.edwards@gmail.com>

- Prev by Date:
**Re: libdts patent issue?** - Next by Date:
**Re: libdts patent issue?** - Previous by thread:
**Re: libdts patent issue?** - Next by thread:
**Re: libdts patent issue?** - Index(es):