Update LICENSE to specify attribution ML training requirements

[itay-grudev]

Clarified conditions for using software as sample data.

Closes #210
Related to: #171

[itay-grudev]

@q234rty @hosiet Is this sufficient? It's no longer in any way forbidden, but attribution is enforced?

[hosiet]

@itay-grudev Can I suggest tightening your words?

 Permission is not granted to use this software or any of the associated files
-as sample data for the purposes of building machine learning models without
+as sample data for any purpose (including building machine learning models) without
 attributing the source material correctly as required by this license.

By tightening the license, we can avoid violating OSI OSD#6.

[itay-grudev]

@hosiet Done!

[hosiet]

Update: I had a second thought. Consider the following scenario:

• I myself use this project's source code to train a local machine learning model for personal use.
• I guarantee I do not redistribute my trained machine learning model in any way.
• I do not want to attribute source materials that I used during my training.

^ This usage shall be allowed.

In other words, attribution shall only be required along with redistribution. In other words, the usage/execution of a software project shall not be limited by whether or not properly attributed.

So the better version should be:

 Permission is not granted to distribute or redistribute this software, the derivative works
 of this software, or any of its associated files that was generated in any approach
 (including building machine learning models) for any purpose without
 attributing the source material by including its license.

Note that I am not a lawyer, and my proposed version may have flaws.

[itay-grudev]

@hosiet That makes sense. Done.

[hosiet]

Thanks. The current version looks good from my side. I am not sure about @q234rty 's thought at the moment.

[q234rty]

The current version looks OK-ish to me but I'm not a lawyer :)

[vejeta]

Hello everybody.

@itay-grudev : I understand that the main concern is about attribution when feeding SingleApplication code
in ML training. The current wording creates ambiguity about ALL distribution, not just ML-related.

Would you consider this alternative that addresses attribution while remaining
DFSG-compliant:

When distributing any work that incorporates or is derived from this Software
(including but not limited to machine learning models trained on this Software),
the above copyright notice and this permission notice shall be preserved and
included with such work in a manner consistent with how copyright notices are
customarily preserved for works of that type.

This ensures attribution for ALL derivative works (including ML models) without
discriminating against specific fields. It's legally clearer and achieves your
goal of requiring attribution when ML models incorporate your code.

The key difference: instead of conditionally denying permission ("permission is
not granted... without"), this unconditionally grants permission while requiring
attribution ("shall be preserved"). This removes ambiguity about what is and
isn't allowed.

Would this address your concerns while keeping the library usable in FOSS
distributions?

I think that with the above paragraph we are ensuring FOSS compatibility
and enforcing attribution, we could even ADD a stronger sentence,
but I believe it is not needed:

For machine learning models specifically, attribution should include
identification of this Software in training data documentation

although I believe the first one is pretty explicit on its own.

[itay-grudev]

@vejeta I think the current versionis simple enough - you include the license and cite the source - you're good.

[itay-grudev]

@hosiet Feel free to package v3.5.4 that has the updated license.