Stockfish: Our lawsuit against ChessBase

[syzygy]

2. If there were no impediment to making exact copies of new machine parts, then I would have expected this to have happened many times before

There are so many examples of this. Screws, nails, springs, batteries, ping pong balls, tea bags, light bulbs, pencils. Someone made them first and as long they are not or no longer protected by an intellectual property right, they can be copied.

2. Per my point, in general, companies don't make exact copies of parts of another company's machinery (obviously there have been exceptions in other jurisdictions as there have been for all IP protection rules).

So you are saying you need IP protection rules after all?
So your point is that the USA must have some SECRET, UNKNOWABLE intellectual property right that MAGICALLY keeps companies from making exact copies of products they did not develop themselves? An intellectual property that is hidden from Wikipedia and law books, and which so far has never been invoked in any court case? Right.

Give an example of a court case that was won against a manufacturer of an exact copy of a machine part that was not/no longer protected by any intellectual property right and that fell outside passing off as defined on its Wikipedia page.

That's my point: I cannot do that because I'm simply not aware of it having happened (apart from in "far away" jurisdictions where IP is not respected),

So this SECRET, UNKNOWABLE intellectual property right MAGICALLY keeps companies from making "exact" copies, while at the same time there are patent infringement suits for copying patented products all the time?

Do you consider this realistic? Or maybe there simply is no such secret intellectual property right?

and my question is: why aren't manufacturers doing it more often if there's no legal impediment to doing it?

Really, how can you still be asking this question...

As now has been stated many times: because companies make sure to protect EVERYTHING they can with utility and design patents and any other applicable IP right. When those run out, the product has typically already been replaced. Unless it becomes a commodity like nails, springs, light bulbs, pencils.

There also many business and technical reasons why companies make their own product. Producing an exact copy (whatever you mean by that) will often be harder and more expensive since a company won't have the same production line and expertise. Also, the copier may want to differentiate its product from the original in price or quality. The IBM-PC compatible clones cost only a fraction of the price of an original IBM PC, guess why. And in case of consumer products there is the tort of "passing off": a company is not allowed to misrepresent its copied product as the original. This will be about the outside of the product, not the internals.

[syzygy]

2. If there were no impediment to making exact copies of new machine parts, then I would have expected this to have happened many times before

There are so many examples of this. Screws, nails, springs, batteries, ping pong balls, tea bags, light bulbs, pencils. Someone made them first and as long they are not or no longer protected by an intellectual property right, they can be copied.

Another example:
https://en.wikipedia.org/wiki/Post-it_Note

Although 3M's patent expired in 1997, "Post-it" and the original notes' distinctive yellow color[2] remain registered company trademarks, with terms such as "repositionable notes" used for similar offerings manufactured by competitors.

The patent ran out, so other companies now produce them too. But they have to work around the trademarks. If these copies are not "exact copies" by your undisclosed definition, then it is because aspects (name, color) of the original are stil protected by non-imaginary rights. Rights that you can find mentioned on Wikipedia, in law books, and in court cases.

Neural networks cannot be protected by a trademark (but their name can).

[towforce]

2. If there were no impediment to making exact copies of new machine parts, then I would have expected this to have happened many times before

There are so many examples of this. Screws, nails, springs, batteries, ping pong balls, tea bags, light bulbs, pencils. Someone made them first and as long they are not or no longer protected by an intellectual property right, they can be copied.

OK - let's take nails as an example. They have existed from a time before human history started to be recorded - so there's no way to patent them.

Suppose company A comes up with a nail that's superior in some way. Company B has a choice:

a) research for themselves how to make a nail with this new specification

b) tool up to make exact copies of company A's nails

Option (b) is going to cost less, and it's going to have a faster time to market - so obviously the better choice, right? So that's what all manufacturers are going to do, right?

Anyway, this discussion is going on ad-nauseum, and is reaching the limit of my willingness to make the same point over and over - but as I'm writing now, I'll state it again (and IMO this simple point hasn't been answered):

An NN is part of a machine, and if it was OK to make exact copies of other companies' machine parts, then we'd expect there to many examples of this having happened.

[syzygy]

Suppose company A comes up with a nail that's superior in some way. Company B has a choice:

Company A will patent their nails. Do you still not get it?
Company B will take a patent license or wait until the patent runs out.

If company A forgets to file for a patent, company B will copy the invention. This is why company A should and will file for a patent.

This is truly elementary, Stop running in circles.

[towforce]

Suppose company A comes up with a nail that's superior in some way. Company B has a choice:

Company A will patent their nails. Do you still not get it?
Company B will take a patent license or wait until the patent runs out.

If company A forgets to file for a patent, company B will copy the invention. This is why company A should and will file for a patent.

This is truly elementary, Stop running in circles.

You shouldn't be able to get a patent for a harder/sharper/cheaper nail - but leaving that aside, you seem to be advocating that each time a company makes an NN that other companies will be able to see (one that gets copied to the user's device rather than running in the cloud) they need to apply for a patent for it.

The patent office is going to be awfully busy: it's bad enough them having to deal with all the slight changes to a standard nail!

[syzygy]

Suppose company A comes up with a nail that's superior in some way. Company B has a choice:

Company A will patent their nails. Do you still not get it?
Company B will take a patent license or wait until the patent runs out.

If company A forgets to file for a patent, company B will copy the invention. This is why company A should and will file for a patent.

This is truly elementary, Stop running in circles.

You shouldn't be able to get a patent for a harder/sharper/cheaper nail

So you are advocating that company B should be free to copy company A's invention of a harder/sharper/cheaper nail. Understood.

- but leaving that aside, you seem to be advocating that each time a company makes an NN that other companies will be able to see (one that gets copied to the user's device rather than running in the cloud) they need to apply for a patent for it.

If I am correct that NNs are not protected by copyright (and this is not just my opinion but the prevailing opinion even if - afaik - not yet tested in court), then indeed only a patent could protect an NN. Alternatively, the legislator could step in and create a sui generis right for NNs (as was done for chip masks).

However, NNs usually come with regular computer code, which is protected by copyright. E.g. an NNUE file for Stockfish does not do anything without Stockfish. If company A sells software that includes an NN, then company B is allowed to copy the NN (weights) but not the software. Company B could attempt to create an original functional copy of the software (as was done with the PC BIOS) and sell that in combination with the NN. If this starts to be a real problem, then perhaps the legislator will step in.

For now, I only see a problem if one considers it desirable to have a market for Stockfish NNUE files. Since SF will sometimes change the format of its NNUE, such a market seems anyway doomed.

Btw, it is not immediately clear to me what effective protection fo NN weights would look like. Protecting the exact same set of weights only helps if the NN has to run with the exact same (copyrighted) software. Once company B anyway has to write that software itself, it can easily slightly modify it and make corresponding modifications to the NN weights. What kind of modifications of company A's NN would still be protected by the patent/copyright/sui generis right on the unmodified NN? (This again shows that copyright is not suitable. Copyright would protect the arrangement of numbers, but it would be trivial to make a small change and have completely different number which share no "expression" with the original arrangement.)

The patent office is going to be awfully busy: it's bad enough them having to deal with all the slight changes to a standard nail!

The US patent office processes about 650,000 patent applications per year. They can deal with a few thousand more of them.

[hgm]

I understood that the U.S. Patent Office doesn't do much more than just file the patents (after checking that it is not for a perpetual-motion device). Whether they might be invalid because of prior art, is up to the courts to decide, when someone decides to challenge the patent.

[dkappe]

I understood that the U.S. Patent Office doesn't do much more than just file the patents (after checking that it is not for a perpetual-motion device). Whether they might be invalid because of prior art, is up to the courts to decide, when someone decides to challenge the patent.

The various claims in a patent filing are often rejected as unpatentable. That’s because patent applications are usually larded up with lots of claims to make the patent as broad as possible. About 89% of patent applications are rejected by the USPTO.

[syzygy]

I understood that the U.S. Patent Office doesn't do much more than just file the patents (after checking that it is not for a perpetual-motion device). Whether they might be invalid because of prior art, is up to the courts to decide, when someone decides to challenge the patent.

The USPTO does examine novelty and non-obviousness. In recent times they have become stricter and the quality of their searches has improved too. But it is true that the grant of a patent does not mean it will survive in court. I believe the stricter line started with the KSR v. Teleflex SCOTUS opinion (which rejected the mechanical teaching-suggestion-motivation test for obviousness as applied by the USPTO and the lower courts, which basically ignored the role of common sense).

Actually, since 2012 the most promising way to attack a granted US patent is by requesting inter partes review at the USPTO (which are dealt with by the Patent Trial and Appeal Board of the USPTO). The PTAB uses the "broadest reasonable interpretation" of a claim when examining for novelty and non-obviousness, whereas the federal courts tend to use a more narrow interpretation (which renders the claim more likely to be upheld but less likely to be infringed). A former (rather patent maximalist) chief judge of the CAFC has referred to the PTABs as death squads.

[syzygy]

I understood that the U.S. Patent Office doesn't do much more than just file the patents (after checking that it is not for a perpetual-motion device). Whether they might be invalid because of prior art, is up to the courts to decide, when someone decides to challenge the patent.

The various claims in a patent filing are often rejected as unpatentable. That’s because patent applications are usually larded up with lots of claims to make the patent as broad as possible. About 89% of patent applications are rejected by the USPTO.

89% sounds high to me. Maybe it is 89% of the originally filed claims?
Indeed it is very common for claims to be dropped or further restricted during prosecution, which for a third party can be just as good as the application being refused.