To this day I have yet to see to a useful software patent.
I worked in video coding back in the day, I can assure you that there are stuff in the HEVC codec that are pretty far from obvious.
AlphaGo is pretty high up there too if you ask me.
Whether patenting software is useful or not is a different debate...
Patents should be contrasted with trade secrets as that is their main "competitor". Is it useful to the general public to be shown the details of a new technology, but have it's use restricted for 20 years? Or is it more useful to prevent blocking of technology but these are going to be kept as trade secrets instead?
From my point of view patents are probably harmful here. Having people be inspired and come up with alternative (or rediscover the same) implementations is probably going to lead to much better for the public than actually learning exactly how the original company did it in exchange for 20y of exclusivity (and all of the pain that enforcing the exclusivity brings).
Remember that unlike copyright patents are not about encouraging the creation of work or innovation. They are a trade of information to the public for a period of exclusivity. For software that tradeoff seems far to far in the inventors favor.
What about AlphaGo? The Reinforcement Learning methods? The Monte Carlo tree search? The Deep Learning?
DeepMind didn't invent any of these things but applied them in a novel way. If everything was software patented, AlphaGo may not have existed.
Just answering the original question.
Did AlphaGo bring anything new to the table? Hell yeah.
It is incredible software and deserves the Turning Prize, no question. But what new patentable tech did they invent?
I asked twice what specific patentable tech they invented and you declined to answer.
It is a straw man to suggest I was arguing against about any specific thing being patentable.
That guy:
https://patents.google.com/patent/EP3459018A1/en?oq=EP345901...
First, most of them are math related, except you get the computer to do the math. You can't (or shouldn't) be able to own or outlaw math.
The last link is a Support Vector Machine as far as I can tell. Those were discovered in like 1964 according to Wikipedia. And then it was patented in 199x? That's absolutely wild
Nope. At least not for any reasonably sane definition that would not apply to all software.
> Those were discovered in like 1964 according to Wikipedia. And then it was patented in 199x?
Maybe before whining about how Vapnik got an unfair patent for something someone else did in 1964, try learning something about SVMs and their history. It's quite insulting.
I agree these are novel and not something immediately obvious. I think what's happening (what always happens) is that people overreact and have a hard time finding nuance when they are upset.
Amazon's one-click patent was ridiculous. Is Google's PageRank? Maybe. Is the RSA algorithm or how to train neural networks? I don't know. Encouraging the development of these by granting limited time monopolies on them seems like it would encourage people to develop them, but if there was a different way to reward the inventors I would be interested in the details (i.e. less blanket statements that it would be as good, but reasoned thought at a minimum and data if we are so lucky as to have some).
Sometimes the problem is not the idea but the implementation. Maybe that's the case with some types of patents.
> I think what's happening (what always happens) is that people overreact and have a hard time finding nuance when they are upset.
Yup. I agree that software patents, in the US at least, are a bit of a cluster-fuck, but it's evidently not true that nothing non-trivial was ever patented.
An application of Bayes' Theorem (1763) and heavily related to Markov chains (1906). Build a Markov chain of simulated users clicking around the internet and build its weights with Bayes' Theorem and you basically have PageRank. It's clever, but no more clever than what thousands of software engineers invent every day.
https://en.wikipedia.org/wiki/Bayes%27_theorem
https://en.wikipedia.org/wiki/Markov_chain
-- RSA
A pretty strong example, but largely based on Diffie-Hellman key exchange published a year before and known since 1969. At least this one was granted to its actual inventors.
https://en.wikipedia.org/wiki/Diffie–Hellman_key_exchange#Cr...
-- Support vector machine
"The original SVM algorithm was invented by Vladimir N. Vapnik and Alexey Ya. Chervonenkis in 1964. In 1992, Bernhard Boser, Isabelle Guyon and Vladimir Vapnik suggested a way to create nonlinear classifiers by applying the kernel trick to maximum-margin hyperplanes. The "soft margin" incarnation, as is commonly used in software packages, was proposed by Corinna Cortes and Vapnik in 1993 and published in 1995."
https://en.wikipedia.org/wiki/Support_vector_machine#History
Patent issued in 2004 to none of the above people. Fail. Patent has FORTY YEARS of prior art. Should never have been issued. I personally remember SVMs being considered "old school" AI in my college AI class in 2005.
> but largely based on Diffie-Hellman key exchange published
Novel applications are patentable.
If they're non-obvious. Remember, the challenge was: show me the strongest, most defensible examples of software patents you can. GP did a great job coming up with 3 very strong examples of patents (taking his word that the SVM patent he meant to show is much stronger). And yet even in those top examples, there are (IMO) reasonable arguments that they still do not deserve protection. I'm not convinced either way, but there are arguments.
My point is that "novel applications" are most of what software engineers do, every day. I contend that there are 10,000+ ideas equally as novel as PageRank invented every day. Senior software engineers probably come up with one every few weeks. This strongly challenges the idea that these are "non-obvious".
Whether or not software patents are a good thing is an entirely separate question (one on which I suspect we agree, in fact) but software patents can and do contain actual non-obvious inventions.
It simply seems like a multi-tier cache (which it is, after all, it's 4 LRU duct taped together in a useful way).
And, maybe there's nothing wrong with the 20 years, the problem is how hard much it costs to license it, how much the whole discovery process costs to the whole industry.
Small but significant correction: software patent that was not obvious to someone skilled in the field AS OF THE FILING DATE (or priority date).