"In the second case, you're not patenting the algorithm, because people remain free to use the algorithm to do things other than sorting lists of restaurant recommendations."

That doesn't make the situation any better, in my mind. In fact, in many ways that makes it worse.

Step 1) File patent for using algorithm (everyone knows) to do (some obvious application of that algorithm).

Step 2) Wait around for some other poor sucker to do the hard work of actually implementing "your" idea.

Step 3) Profit!

1) In theory you wouldn't be able to file patents without a working prototype--we've unfortunately gotten away from this requirement;

2) That assumes "implementing the algorithm" is the hard part.

IMO/IME, implementing is always the hard part.

I agree with your #1 though in that I think you should absolutely be required to have a working prototype to file a successful patent. A lot (though not all) of my objections to software patents would go away if that were enforced.

I disagree. If you look at something like the software in your cell phone that handles the power control loop or hand-offs, etc, I guarantee you it's not very many lines of code. But the underlying algorithms (the specification), is the result of months if not years of experimentation and empirical validation.

That's the sweat of the brow argument: If somebody worked really hard then they should get a cookie. But that doesn't work. If Bob is a savant and he designs a brilliant hand-off algorithm in five minutes, should that not be patentable because he didn't have to work very hard, but an algorithm that isn't even as good should be because a much worse engineer spent a lot longer to design it? The result would be to reward mediocrity and waste.

Which is why we look at the invention instead of the effort. So even though I'm sure Einstein worked really hard on E=mc^2, laws of nature are still unpatentable, because it's a description rather than an invention.

The problem is that no matter how hard it is to discover it, we can't give you E=mc^2, because it only describes things. You can actually have implementations that predate the discovery, because it's possible to build something without being able to describe how it works. We had X-ray machines before we had E=mc^2 even though it describes their operation. Allowing a patent on anything following the description would allow too wide a scope. You would be claiming too many things you didn't invent.

But algorithms are the same way. They're not inventions, they're abstractions. There is a really cool ad hoc routing algorithm where you send packets along random routes but tag the routes, and then the more/faster packets get to a destination using a particular route, the more you prefer that route for future packets. Sounds like something that should be patentable if algorithms are patentable, right?

But it's called the ant algorithm. Guess why.

Algorithms are pure abstractions. Those little buggers crawling around leaving scent tracks everywhere have certainly never heard of your algorithm and have no capacity to produce any sort of a binary, but there they are infringing your would-be algorithm patent.

So you say, a-ha! We'll just make it an invention by applying a context. Then the wildlife will be able to escape liability (and hopefully stop serving as pesky prior art).

The issue is that the context is totally facile. The interesting thing is the algorithm, not the context. It's just taking the unpatentable thing and adding "on a computer" and claiming it as an invention. Sometimes they try to be subtle and instead of saying "computer" they say "ad hoc wireless network" or "anonymizing P2P network" or some other thing the meaning of which is always "computer that could benefit from using this unpatentable algorithm" as if to pretend that the limitation is meant as an actual restriction rather than merely a vague description of the circumstances under which the algorithm is expected to be beneficial.

> IMO/IME, implementing is always the hard part.

I'm actually surprised anybody would say that. To use an example, you're saying it's more difficult to implement quicksort than it was to come up with it in the first place?

IME, compared to thinking up an algorithm, implementing them is very easy.

>In the second case, you're not patenting the algorithm, because people remain free to use the algorithm to do things other than sorting lists of restaurant recommendations.

I know this wasn't your example originally, but it does exemplify a common problem: There is nothing special about sorting restaurant recommendations vs. sorting some other such thing. If you have an unpatentable algorithm which is good at sorting lists of anything it shouldn't become patentable just because you claim it as "sorting restaurant recommendations." Mixing an unpatentable algorithm with a totally obvious application of it shouldn't get you a patent.

>Ultimately you have to tie it back to the purpose of the property right. Why does someone have a legal right to a particular glob of atoms, or a particular string of bits, or a particular series of steps of some algorithm in some specific context?

So why should someone have a legal right to a particular series of steps of some algorithm in a specific context? Why should adding an arbitrary context make it patentable if it wouldn't have been otherwise?

> I know this wasn't your example originally, but it does exemplify a common problem: There is nothing special about sorting restaurant recommendations vs. sorting some other such thing. If you have an unpatentable algorithm which is good at sorting lists of anything it shouldn't become patentable just because you claim it as "sorting restaurant recommendations." Mixing an unpatentable algorithm with a totally obvious application of it shouldn't get you a patent.

Totally correct. The first requirement to patent something it that it should not be a trivial invention (something that is so easily derived from something else). But most patents are, however, trivial in the software world. That is a big issue.