3

I'm doing some experimentation with machine learning, for purely educational/scientific purposes, and one of the concerns that came up, is:

If I use training materials (images and other data) designated for non-commercial use, and then release the trained model, do I also need to give the model a non-commercial license, or can I mark it as completely free to use for any purpose?

Would I be responsible if someone uses this model for commercial use?

4
  • It sounds as if you are already violating the license agreement by, as a COMPANY, you are using the materials for commercial use. Perhaps I'm misunderstanding your question, so in that case, please edit it to make it more clear.
    – jwh20
    Oct 20, 2022 at 14:51
  • 1
    @jwh20 I'm an individual, I shouldn't have used the word "work", as I'm doing it for educational purposes only. My question is about the product of my experiment.
    – Kevin
    Oct 20, 2022 at 15:19
  • You should reword your question to make that clear. You said "in a company".
    – jwh20
    Oct 20, 2022 at 15:42
  • I didn't think I did, I wrote the question while someone else was talking to me, so I might have. Either way, it's gone now.
    – Kevin
    Oct 20, 2022 at 16:21

1 Answer 1

4

There is no fixed legal meaning to "non-commercial", instead, every license must define what is meant by "non-commercial", if it uses that term. CC BY-NC 4.0 says that

NonCommercial means not primarily intended for or directed towards commercial advantage or monetary compensation. For purposes of this Public License, the exchange of the Licensed Material for other material subject to Copyright and Similar Rights by digital file-sharing or similar means is NonCommercial provided there is no payment of monetary compensation in connection with the exchange.

That would mean that you cannot sell software of a dataset so licensed, but you could give it away.

Copyright does not prohibit using a work as raw data for some kind of analysis, such as counting occurrences of words, so it is legal to publish a table of word-counts from the Harry Potter novels, even though you can't publish the novels without permission. The word counts are facts, not "expression", and copyright doesn't protect facts.

It is possible that your initial access to the expressive work was contingent on not doing something with that work, for example a license to copy a book in the first place might be contingent on you not using the data for purposes of statistical analysis. For Harry Potter books, there's really no way to show that Smith could only have gained access to the work by agreeing to a certain license and showing that Smith did agree (Potter books are everywhere). It is possible that some work is so well controlled that you did indeed agree to "not subject the work to statistical analysis". It is highly unlikely that there is any such condition is attached to a work that has is "non-commercially" licensed. You can read the license and see what it actually says. Basically, an abstract model of X is not X itself, it is a fact about X, and facts are not protected.

1
  • 1
    The "license" that some of the data has is akin to a line of text in the description saying "You may use this for anything that is not commercial". It's definitely not nearly as well controlled as you stated.
    – Kevin
    Oct 20, 2022 at 16:39

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .