29

I'm new to the domain of understanding the law behind patenting, and I want to start a company that uses machine learning algorithms. Unfortunately, almost every concept is patented by the top companies.

Aren't computer algorithms mathematical equations? Any algorithm can be represented mathematically. People may argue that even physical objects like engines can be expressed mathematically, but physical objects are obviously different from mathematical expressions. But how are algorithms different from mathematical equations? Many machine learning algorithms exactly satisfy the definition of mathematical equations, as they were directly derived from statistical methods used in the past (which I believe were not patentable before the software era). Of course, an animation created by an artist, cannot be considered as a mathematical work, even if the software uses multiple mathematical methods to make it function.

I have been reading the article in this link: https://www.gnu.org/philosophy/software-patents.en.html and here the author states that the public key encryption method was patented. Don't encryption algorithms precisely come under the definition of mathematical equations?

Sreram
  • 405
  • 4
  • 8
  • 5
    Programmes aren't patentable, they're copyrighted. If you mean algorithms, say algorithms. And no, programmes clearly aren't equations any more than humans are, regardless of how they can be represented. –  May 04 '18 at 20:00
  • "Any program can be represented mathematically" - not really. They are lists of instructions. – mark b May 04 '18 at 21:52
  • 5
    @markb: Turing-Church thesis, yes they can. There are multiple way to represent programs, but there no specific way that is most powerful. There is a very broad class ("Turing complete") of programming languages that are equally powerful. – MSalters May 05 '18 at 01:37
  • 1
    We have an sx for patents and it has questions like this: What constitutes an original patentable idea in software?. – Martin Schröder May 05 '18 at 12:12
  • 4
    “Aren't computer algorithms mathematical equations?” No. – Andrea Lazzarotto May 05 '18 at 19:33
  • Please tag your question with the appropriate country as there are countries where this is different and where software patents do not exist – PlasmaHH May 05 '18 at 19:52
  • 1
    Computer programs are just mathematical equations in the same way paintings are just colors and books are just letters. You can't patent letters so why does plagiarism even exist!? – Clay07g May 06 '18 at 03:45
  • Computer programs are lists of instructions to a CPU. Nothing more. As with any other arbitrary list of items, it's ability to be condensed into an equation is not always true. The instructions could be chaotic or have no pattern. – mark b May 07 '18 at 14:48
  • @Nji I certainly agree that humans are equations but engines aren't. You can patent a novel engine design, but it doesn't mean you are prevented from having an electronic copy of the patent or the detailed method to construct the engine. As for algorithms, they are just a human-friendly representation of the program. A computer program is just an alternate way of writing an algorithm. Isn't an electronic record of a computer program indistinguishable from the algorithm? Even the assembly instructions are just another language, giving a different way to express the same thing. – Sreram Dec 11 '18 at 00:29
  • Though in the practical word this may not be a problem, logically, if possessing the electronic record of the algorithm described in the patent is considered to be indistinguishable from the same algorithm described in a computer language called Python, and if the algorithm written in Python violates the patent's terms (remember, it is not a manufactured object. Just another electronic record of the algorithm, in a different form) shouldn't possessing an electronic copy of the patent itself be a violation? – Sreram Dec 11 '18 at 00:41

3 Answers3

31

Aren't computer programs mathematical equations? Any program can be represented mathematically. . . . But how are programs different from mathematical equations? . . . Don't encryption algorithms precisely come under the definition of mathematical equations?

In the context of patent law, the "mathematical equations" that aren't patentable are those that are discovered and are mathematical laws of Nature.

For example, you cannot patent a mathematical equation which is the equivalent of the Fundamental Theorem of Algebra (which tells you how many solutions there are to all possible polynomial equations), because it is a statement about what is fundamentally true about something, not about how to do something that works better than other possible ways of doing something.

An algorithm, in contrast, in patent law, is a method of doing something that is not inherently the only way of doing something as a consequence of the laws of mathematics or other laws of Nature (although it could very well be the "best" way of doing something).

For example, the simplex method of linear programming would probably be patentable if it hadn't been publicly disclosed prior to anyone obtaining a patent on this method.

Most computer programs are the digital embodiment of algorithms, rather than of equations that were discovered that represent fundamental truths that are always and uniquely true about numbers or reality.

Notably, recent governing U.S. Supreme Court case law has now made it categorically true that there mere fact of embodying an algorithm or other business method in digital form with a computer program, as a matter of law, does not make an idea that would not be patentable if not embodied in a computer program patentable. The underlying idea that is embodied digitally in a computer program must, itself, be patentable.

A method for doing something, even if it can be or is represented mathematically, is not inherently not subject to being patented. But, not every method of doing something is patentable either. The "business method" must still clear the hurdles of being original (you can't patent ideas that have already been disclosed publicly by someone else, or by yourself for that matter), of having "utility" rather than merely aesthetic value, and of not being obvious to a person of ordinary skill if the field in which the invention arises.

For example, anything that would follow as a lemma or corollary of an existing aspect of prior art would not be eligible for a patent either, even if it is novel and useful.

Now, there is some room for sophistry here.

For example, you could call a computer equation that calculated a Lorentz transform a method for calculating the relationship between two coordinate frames to which relativistic mechanics apply. But, because that relationship is a law of Nature which is always true, even if it isn't the only means to calculating that relationship, it would not be patentable even if it were not prior art.

I want to start a company that uses machine learning algorithms.

While it is a risky business model, it is worth considering that over the last 10-15 years the legal threshold necessary to patent a computer program has gotten much, much higher.

The Patent and Trademark Office used to routinely grant these applications, but due to a string of important statutory and case law changes, software patents are now denied at a very high rate, and existing software patents are routinely determined to be invalid in litigation.

If there is serious reason to doubt that an existing patent that would be important for your business is really patentable, you should consult a patent lawyer to consider the possibility of bringing an interpartes review action in the Patent and Trademark Office to have a previously issued patent revoked. This is an expensive and time consuming process, but it is much cheaper than losing a patent infringement lawsuit after having invested time and money into running a business using a patent you know to be infringing because you believe that it won't hold up in court.

ohwilleke
  • 211,353
  • 14
  • 403
  • 716
  • 1
    I agree with your remarks about equations; however, in fairness (if not in law), sometimes systems of equations seem like they ought to be patentable. Consider, e.g., Maxwell's equations, which immediately led to Heinrich Hertz's early radio experiments, followed by Marconi's radio. Lots and lots of radio-related patents. But seems like Maxwell deserves some slice of that pie -- the engineers ultimately baked the pie, so to speak, but it was Maxwell who came up with its necessary ingredients. The engineers get all the patents, but they totally rely on foundations developed by physics/math. – John Forkosh May 04 '18 at 23:21
  • 4
    @JohnForkosh That is a feature, not a bug. If Maxwell had patented Maxwell's equations, nobody else could create any electromagnetic system without paying him a royalty until the patent expired, bringing every field of applied science involving electromagnetic forces to a screeching halt for a couple of decades. And, honestly, most top scientists live pretty decent lives with a very different tenured professor business model, and lots of top scientists also invent things that they patent and do earn royalties upon. – ohwilleke May 05 '18 at 02:58
  • 1
    This is a good answer, but I think that in some programming languages, the distinction between algorithm and equation is more blurred. For instance, the length of a list could be expressed by the equations length([]) = 0; length([head|tail]) = 1 + length(tail). It could also be expressed by the equations length(list) = len(list,0); len([],x) = x; len([head|tail], y) = len(tail, y+1). In a language like Prolog, while these are logically equivalent, the algorithm is distinct (one is a simple recursive definition requiring linear stack space, and the other is tail recursive and essentially – Joshua Taylor May 05 '18 at 04:00
  • 2
    a loop). If your programming language is essentially a system for evaluating equations, then different ways of writing the equations really are the different algorithms. – Joshua Taylor May 05 '18 at 04:01
  • So theoretically, Newton's laws of motion could have been patented in the right circumstances, because they are not strictly correct laws of nature, but they are good ways to approximate physical phenomenon. Yes? – 16807 May 05 '18 at 16:38
  • @16807 No. Newton's laws of motion are still laws of nature, not ways to doing something. The Patent office is not charged with evaluating the absolute accuracy of proposed patents, just their character, originality, obviousness, and utility. – ohwilleke May 05 '18 at 17:24
  • @JoshuaTaylor I'm not convinced that the example you give would be patentable, although there are multiple reasons one could offer for why that is the case (the argument that it is obvious might be stronger than the argument that it is a law of Nature, although the ideas overlap). – ohwilleke May 05 '18 at 17:27
  • @ohwilleke absolutely, I don't think that it necessarily makes it patentable. But I think that the fact that the programming language, as the background, can make so much implicit, can mean that the way that a is phrased, really can be a significant difference. I'm not going to make any Claim about whether, or how, patent law should address that. I just wanted to point it out. – Joshua Taylor May 05 '18 at 17:34
  • 2
    This may well be the legal perspective, but I'm not convinced that it's a proper grasp of the nature of math, science, and their relationship. Consider this excerpt of a Feynman lecture: https://youtu.be/YaUlqXRPMmY It also doesn't seem to consider that all mathematical theories must necessarily start from a set of unproven axioms, which means none of them are really "fundamental properties of nature." – jpmc26 May 06 '18 at 01:11
  • 1
    @ohwilleke Is there anything that characterizes it as a law of nature besides scientific consensus? Were it originally cast as a heuristic then I fail to see how it it differs from a patentable algorithm. I can expect this happens all the time in physics, where an ansatz is established as an approximation first and then later found to be an actual reflection of reality. – 16807 May 06 '18 at 01:32
  • "Notably, recent governing U.S. Supreme Court case law has now made it categorically true ... does not make an idea that would not be patentable if not embodied in a computer program patentable." This sentence is too long and due to the triple negation, quite hard to understand, especially for non-native English speakers. I eventually got my head around it, but would you mind breaking it down into simpler sentences? The sentence following that helps to an extent, but a simpler alternative would be useful as well. –  May 06 '18 at 13:06
  • There's also the issue of "obviousness", which in the past the patent office did a poor job. One example I've seen cited as a potential standard is if 7 or more out of 10 engineers were assigned the same problem that the algorithm solves and they come up with essentially the same algorithm, then it's "obvious" and should not be patented. – rcgldr May 06 '18 at 16:07
  • The worst case that I'm aware of as an example of a patent that should not have been granted was when a prior patent stated than when content addressable memories become cheap enough, then they should be used instead of hashing to implement a compression algorithm, yet the patent office granted a patent for doing just that, using a content addressable memory instead of hashing for that algorithm. The patent office now does a much better job of preventing patents like this from being granted. – rcgldr May 06 '18 at 16:09
  • This post makes me very, very sad. While I acknowledge your answer, the very idea of "patenting" an algorithm, which is just part of a constructive existence proof, is maddening. But then I remember that in the U.S. you can even patent round edges, so that madness on their part makes me feel better again. – SK19 May 06 '18 at 18:10
  • @16807 Scientific consensus have nothing to do with it whatsoever. A claimed law of nature is not patentable whether or not it is true. – ohwilleke May 07 '18 at 01:06
  • @SK19 "an algorithm, which is just part of a constructive existence proof," Not all algorithms, even if implemented, are necessarily patentable. To be patentable an algorithm must be capable of being implemented in a manner that has utility for the owner. Not every algorithm is part of a constructive existence proof. If patentable it is merely a step by step process for doing something in a way that is useful that no one had come up with before. – ohwilleke May 07 '18 at 01:12
  • @ohwilleke My point is that a law of nature could instead be billed as a heuristic for approximating solutions to certain real world problems, and in that case, would such a thing be patentable? – 16807 May 07 '18 at 15:36
  • @16807 No. Even as a heuristic for approximately solutions to real world problems statements about how reality is, even approximately, are never patentable. Also, the distinction is usually easier in practice than in theory. This is rarely a tough rule to apply compared to other patentability rules. – ohwilleke May 07 '18 at 16:10
8

To add to the existing answer, what a company will patent is what they think will cost too much for a competitor to try to challenge. It does not have to be original, novel, or even discovered by them. It just has to be enough of an obstacle for competitors to defend their space - or (more commonly) to negotiate a mutual non-aggression pact with a competitor who has a similar portfolio of equally iffy patents.

A case study here. Closed-loop control is a standard concept from before Victorian times - you decide where you want something to be, monitor the current state, and feed more or less juice to your output until it gets there. Suppose you wanted to run your car at a constant speed. It would be natural to think of closed-loop control. Measure speed, apply more or less torque depending on whether you're faster or slower than you want to be, and job done. It's a standard concept, after all.

Not so fast. GM own a patent for closed-loop control of vehicle speed by choosing a setpoint in position, velocity or acceleration and measuring that quantity such that you close the loop based on the error. Unless you've licensed GM's patent, they can sue you. So if you're in the business, you pony up and pay your vig like a good boy.

Or you can work out a way around the patent. Ford decided to avoid this. Based on legal advice, the controller is split into two parts. One block looks at measured speed versus desired speed, but instead of producing an error term directly, it goes into a map which produces a desired acceleration. The next block in the firmware takes that acceleration and uses it in the torque demand map, and then things go on as normal. The lawyers decided that this was not producing an error term in speed, therefore GM's patent doesn't apply. It's too much effort for GM to try to sue Ford, and the patent would probably be struck down anyway, so the result is a standoff.

I realise this sounds unbelievable, but I've actually worked in Ford on the exact code I'm talking about here. It's not just an urban legend, it's personal experience.

Graham
  • 2,711
  • 10
  • 14
  • 2
    An interesting story, but how does it relate to the OP's question? – Robert Harvey May 04 '18 at 22:28
  • 4
    He's asking about patenting a simple algorithm. I've given him an example of how the most basic control algorithm with centuries of prior art can still be patented, and why companies get away with it - and an example of how this can be countered by a similarly-motivated competitor. – Graham May 04 '18 at 23:59
  • 1
    I have a similar anecdote. A company had a patent on marking video snippets by using a start timestamp and end timestamp and then labeling it. We worked around the patent by flagging everything except the content to be marked, and by using a "not" label. It's stupid but the courts said it was different enough to get around the patent. – Thane Brimhall May 05 '18 at 01:06
4

Algorithms are not (in the USA) patentable.

You can build a machine that does something in a new way that hasn't been done before, and patent it. Part of building the machine may be an algorithm. The machine may actually use a novel algorithm and nothing else that is novel, and be patentable. That doesn't mean the algorithm itself is patentable, only in conjunction with the real invention.

The example of the Simplex algorithm was given. You might make an invention that solves a problem by finding an optimal solution to a system of linear equations and inequalities, using the Simplex algorithm, and patent the invention. I might then make an invention that solves a completely different problem using the Simplex algorithm, and patent it. It doesn't infringe on your patent. Algorithms are not patentable, only inventions which may use algorithms.

gnasher729
  • 34,028
  • 2
  • 46
  • 88
  • Sort of. This analysis of the U.S. Supreme Court in Alice v. CLS (2014) is a bit more precise: https://jenner.com/system/assets/publications/13136/original/U.S.Supreme_Court_Narrows_Patentability_of_Software_Algorithms-_ATTORNEY_ADVERTISING.pdf?1403297305 – ohwilleke May 05 '18 at 17:32