Are neural nets (the weights file) copyrightable?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

syzygy
Posts: 5563
Joined: Tue Feb 28, 2012 11:56 pm

Re: Are neural nets (the weights file) copyrightable?

Post by syzygy »

towforce wrote: Wed Mar 03, 2021 12:39 am
syzygy wrote: Wed Mar 03, 2021 12:09 am
towforce wrote: Wed Mar 03, 2021 12:02 am
syzygy wrote: Tue Mar 02, 2021 11:57 pmIf I take covered work = SF...
I don't think you can do that. I think that if you didn't want your NN included in the covered work, you'd achieve this by not using SF in your project.
Of course I can do that. People are distributing SF by itself all the time, for which they have permission from the GPLv3 with covered work = SF.

So SF by itself certainly is a covered work licensed under the GPLv3. (Covered work = the program or a work based on the program. There you have it.)
This license gives me permission to distribute SF under certain conditions.
Which condition does the GPLv3 text specify that is not fulfilled when I bundle SF with an NN?
Please spell out for me the relevant portion of the GPLv3. I don't want the usual "you can't do that!", I want to see the text from the GPLv3 that tells me I can't do that.
As I was told earlier, "...not by their nature extensions of the covered work, and which are not combined with it such as to form a larger program...".
Read the whole sentence. It is only a definition of the word "aggregate".
User avatar
MikeB
Posts: 4889
Joined: Thu Mar 09, 2006 6:34 am
Location: Pen Argyl, Pennsylvania

Re: Are neural nets (the weights file) copyrightable?

Post by MikeB »

A little bit dated, but still an interesting read ..portions of the article are applicable to this discussion...

https://www.wipo.int/wipo_magazine/en/2 ... index.html
Image
gonzochess75
Posts: 208
Joined: Mon Dec 10, 2018 3:29 pm
Full name: Adam Treat

Re: Are neural nets (the weights file) copyrightable?

Post by gonzochess75 »

syzygy wrote: Wed Mar 03, 2021 12:47 am
towforce wrote: Wed Mar 03, 2021 12:39 am
syzygy wrote: Wed Mar 03, 2021 12:09 am
towforce wrote: Wed Mar 03, 2021 12:02 am
syzygy wrote: Tue Mar 02, 2021 11:57 pmIf I take covered work = SF...
I don't think you can do that. I think that if you didn't want your NN included in the covered work, you'd achieve this by not using SF in your project.
Of course I can do that. People are distributing SF by itself all the time, for which they have permission from the GPLv3 with covered work = SF.

So SF by itself certainly is a covered work licensed under the GPLv3. (Covered work = the program or a work based on the program. There you have it.)
This license gives me permission to distribute SF under certain conditions.
Which condition does the GPLv3 text specify that is not fulfilled when I bundle SF with an NN?
Please spell out for me the relevant portion of the GPLv3. I don't want the usual "you can't do that!", I want to see the text from the GPLv3 that tells me I can't do that.
As I was told earlier, "...not by their nature extensions of the covered work, and which are not combined with it such as to form a larger program...".
Read the whole sentence. It is only a definition of the word "aggregate".
Indeed it is. And the license goes on to spell out exceptions for aggregate to the previous requirements. If you don't meet the definition of aggregate, then the full requirements are yours to meet or you can't distribute the covered work. This is not that hard.
Michel
Posts: 2272
Joined: Mon Sep 29, 2008 1:50 am

Re: Are neural nets (the weights file) copyrightable?

Post by Michel »

gonzochess75 wrote: Wed Mar 03, 2021 5:52 am
syzygy wrote: Wed Mar 03, 2021 12:47 am
towforce wrote: Wed Mar 03, 2021 12:39 am
syzygy wrote: Wed Mar 03, 2021 12:09 am
towforce wrote: Wed Mar 03, 2021 12:02 am
syzygy wrote: Tue Mar 02, 2021 11:57 pmIf I take covered work = SF...
I don't think you can do that. I think that if you didn't want your NN included in the covered work, you'd achieve this by not using SF in your project.
Of course I can do that. People are distributing SF by itself all the time, for which they have permission from the GPLv3 with covered work = SF.

So SF by itself certainly is a covered work licensed under the GPLv3. (Covered work = the program or a work based on the program. There you have it.)
This license gives me permission to distribute SF under certain conditions.
Which condition does the GPLv3 text specify that is not fulfilled when I bundle SF with an NN?
Please spell out for me the relevant portion of the GPLv3. I don't want the usual "you can't do that!", I want to see the text from the GPLv3 that tells me I can't do that.
As I was told earlier, "...not by their nature extensions of the covered work, and which are not combined with it such as to form a larger program...".
Read the whole sentence. It is only a definition of the word "aggregate".
Indeed it is. And the license goes on to spell out exceptions for aggregate to the previous requirements. If you don't meet the definition of aggregate, then the full requirements are yours to meet or you can't distribute the covered work. This is not that hard.
The relevant sentence is this
gplv3 wrote:Inclusion of a covered work in an aggregate does not cause this License to apply to the other parts of the aggregate.
If we consider the covered work to be SF then FF2 is either an aggregate or it is not. If it is an aggregate then it is ok to distribute it in whatever way you want. If it is not an aggregate then this sentence has nothing to say about it.

So the only way to argue that FF2 is a GPL-3 violation is to argue that FF2 is a covered work (and not just SF). That would be my position since it is analogous to dynamic linking which is what the FSF considers to be a covered work. So one has first to accept the FSF's position (which has never been tested in court, one way or the other), and then one has to reason by analogy (which I believe is sometimes admissible in court).
Ideas=science. Simplification=engineering.
Without ideas there is nothing to simplify.
syzygy
Posts: 5563
Joined: Tue Feb 28, 2012 11:56 pm

Re: Are neural nets (the weights file) copyrightable?

Post by syzygy »

gonzochess75 wrote: Wed Mar 03, 2021 5:52 am Indeed it is. And the license goes on to spell out exceptions for aggregate to the previous requirements.
What do you mean by that? Still as part of the definition of "aggregate"? Which text are you referring to?
If you don't meet the definition of aggregate, then the full requirements are yours to meet or you can't distribute the covered work.
If I have already satisfied conditions a) to d), then what does the definition of "aggregate" have to do with that? Please use the text of the GPLv3 in your arguments instead of 'you can't do that!".
GPLv3 wrote:You may convey a work based on the Program, or the modifications to produce it from the Program, in the form of source code under the terms of section 4, provided that you also meet all of these conditions:

a) The work must carry prominent notices stating that you modified it, and giving a relevant date.
b) The work must carry prominent notices stating that it is released under this License and any conditions added under section 7. This requirement modifies the requirement in section 4 to “keep intact all notices”.
c) You must license the entire work, as a whole, under this License to anyone who comes into possession of a copy. This License will therefore apply, along with any applicable section 7 additional terms, to the whole of the work, and all its parts, regardless of how they are packaged. This License gives no permission to license the work in any other way, but it does not invalidate such permission if you have separately received it.
d) If the work has interactive user interfaces, each must display Appropriate Legal Notices; however, if the Program has interactive interfaces that do not display Appropriate Legal Notices, your work need not make them do so.
SF is the Program, FF2 SF is a work based on the Program. So I may distribute both provided I meet a), b), c) and d).
Since i may certainly distribute SF by itself, SF is an "entire work" within the meaning of c).
So there is no problem complying with a), b), c) and d).
The permission to distribute SF includes the permission to distribute SF in return for money, or as part of a zip file, or under any other circumstance, unless the permission was conditional on such a circumstance being absent. None of a), b), c) or d) make the permission conditional on SF not being part of a zip file.

Now the next sentence:
A compilation of a covered work with other separate and independent works, which are not by their nature extensions of the covered work, and which are not combined with it such as to form a larger program, in or on a volume of a storage or distribution medium, is called an “aggregate” if the compilation and its resulting copyright are not used to limit the access or legal rights of the compilation's users beyond what the individual works permit.
This is just the difficult-to-parse definition of "aggregate". For the sake of argument, let us assume that a zip file with SF and an NNUE net is not an "aggregate".
Inclusion of a covered work in an aggregate does not cause this License to apply to the other parts of the aggregate.
OK. But clearly the inclusion of SF in a "non-aggregate" zip file with an NNUE net cannot cause the GPLv3 apply to the NNUE net. The author of the NNUE net gets to decide what license applies to it. So there is no possibility of arguing a contrario here.

Where do I go wrong? Please use the text of the GPLv3, not "you can't do that!".
syzygy
Posts: 5563
Joined: Tue Feb 28, 2012 11:56 pm

Re: Are neural nets (the weights file) copyrightable?

Post by syzygy »

Michel wrote: Wed Mar 03, 2021 8:32 amThe relevant sentence is this
gplv3 wrote:Inclusion of a covered work in an aggregate does not cause this License to apply to the other parts of the aggregate.
If we consider the covered work to be SF then FF2 is either an aggregate or it is not. If it is an aggregate then it is ok to distribute it in whatever way you want. If it is not an aggregate then this sentence has nothing to say about it.
Indeed we are reading this in the same way (as already became clear in the beginning of this thread).
So the only way to argue that FF2 is a GPL-3 violation is to argue that FF2 is a covered work (and not just SF).
But why would "just SF" not be a covered work that can be distributed with permission? It seems universally accepted that one can distribute the SF executable by itself (together with the promise to provide source code on request etc.).

(In general, my position is that even a completely broken program can be distributed by itself. There is no requirement that GPL'd programs be functional. That would make "open" development of GPL'd programs pretty hard to do.)
Michel
Posts: 2272
Joined: Mon Sep 29, 2008 1:50 am

Re: Are neural nets (the weights file) copyrightable?

Post by Michel »

syzygy wrote: Wed Mar 03, 2021 10:34 am
Michel wrote: Wed Mar 03, 2021 8:32 amThe relevant sentence is this
gplv3 wrote:Inclusion of a covered work in an aggregate does not cause this License to apply to the other parts of the aggregate.
If we consider the covered work to be SF then FF2 is either an aggregate or it is not. If it is an aggregate then it is ok to distribute it in whatever way you want. If it is not an aggregate then this sentence has nothing to say about it.
Indeed we are reading this in the same way (as already became clear in the beginning of this thread).
So the only way to argue that FF2 is a GPL-3 violation is to argue that FF2 is a covered work (and not just SF).
But why would "just SF" not be a covered work that can be distributed with permission? It seems universally accepted that one can distribute the SF executable by itself (together with the promise to provide source code on request etc.).

(In general, my position is that even a completely broken program can be distributed by itself. There is no requirement that GPL'd programs be functional. That would make "open" development of GPL'd programs pretty hard to do.)
My idea is that "the right to distribute something" is not something that is acquired once and then kept forever. You can lose the right to distribute something.

If you accept that FF2 is a covered work then the distribution of it is a violation of the GPL-3 license attached to SF. Hence you can no longer legally accept this GPL-3 licence. So you cannot distribute SF and hence you can also not distribute FF2.
Ideas=science. Simplification=engineering.
Without ideas there is nothing to simplify.
Dann Corbit
Posts: 12540
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: Are neural nets (the weights file) copyrightable?

Post by Dann Corbit »

>>
If you accept that FF2 is a covered work then the distribution of it is a violation of the GPL-3 license attached to SF. Hence you can no longer legally accept this GPL-3 licence. So you cannot distribute SF and hence you can also not distribute FF2.
<<
what happens if I do not accept that?

I see no violations of GPL license. I guess that if you ask the GPL body themselves, they will say the same.
If they say something different, then I think that they are emotional and not logical.

Let me restate my position:
Albert has done absolutely nothing wrong (even his unbounded exuberance can be forgiven easily).
There are no violations and those who seek them are grasping at straws that do not exist (IMO-YMMV)
Albert is guilty of the following: Over-exuberance of his own offspring. Which of us is free from blame in like circumstance? Certainly not me.

Are any of us free from such over-exuberance? My sons have excelled at certain things and I took a swelling pride in it (though I myself had nothing to do with it).
We seem to target Albert who (as far as I can see) has done literally nothing illegal, and yawn at those who blatantly violate the law.

Sometimes I wonder, "Why do I spend time and energy on chess when the audience is full of hate and bile, for no apparent reason?"
Does anyone have a good answer for me?
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Michel
Posts: 2272
Joined: Mon Sep 29, 2008 1:50 am

Re: Are neural nets (the weights file) copyrightable?

Post by Michel »

Dann Corbit wrote: Wed Mar 03, 2021 11:45 am >>
If you accept that FF2 is a covered work then the distribution of it is a violation of the GPL-3 license attached to SF. Hence you can no longer legally accept this GPL-3 licence. So you cannot distribute SF and hence you can also not distribute FF2.
<<
what happens if I do not accept that?
When I was writing "if you accept that" then I meant "if you accept my reasoning, or any other, that FF2 is a covered work according to the GPL-3 license".

Whether FF2 is a covered work or not depends on copyright law so the courts would be the final arbiter.

Note that this is a purely theoretical discussion since the CB copyright over the FF2 net is probably invalid and moreover nobody is really interested in this net anyway since it is weaker than the one which comes with SF.
Last edited by Michel on Wed Mar 03, 2021 12:03 pm, edited 1 time in total.
Ideas=science. Simplification=engineering.
Without ideas there is nothing to simplify.
Dann Corbit
Posts: 12540
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: Are neural nets (the weights file) copyrightable?

Post by Dann Corbit »

Michel wrote: Wed Mar 03, 2021 11:58 am
Dann Corbit wrote: Wed Mar 03, 2021 11:45 am >>
If you accept that FF2 is a covered work then the distribution of it is a violation of the GPL-3 license attached to SF. Hence you can no longer legally accept this GPL-3 licence. So you cannot distribute SF and hence you can also not distribute FF2.
<<
what happens if I do not accept that?
When I was writing "if you accept that" then I meant "if you accept my reasoning, or any other, that FF2 is a covered work according to the GPL-3 license".

Whether FF2 is a covered work or not depends on copyright law so the courts would be the final arbiter.
This position is entirely reasonable and I do accept that.
This sort of thing is not where I have issues and problems.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.