[License-review] Notice requirement for model output: OSD-compliant or not? (ModelGo)

[Pamela Chestek]

On 3/2/2025 7:26 PM, Moming Duan wrote:
> Hi Pamela,
>
>
>> 2. I believe the nature of open source licenses is that they impose 
>> requirements on the licensee only to the extent they assure that the 
>> licensor receives the benefit that is their motive for putting 
>> software under an open source license and to protect the licensor 
>> from harm when they have generously made a gift. However, this term 
>> is for the benefit of the public and I don't see what value the 
>> licensor gets from it. If that's true, I think it is just another 
>> barrier that a licensee has to take into account, thus introducing 
>> friction when we are trying to make licensing as frictionless as 
>> possible.
>
> I acknowledge that requiring AI-generated content to be declared may 
> introduce friction, but the absence of such a provision can also harm 
> open source. Nowadays, it is common to transfer knowledge by improving 
> a model based on content generated by other models. If we do not 
> address this behavior, the sustainability of open source cannot be 
> guaranteed. Even though the ModelGo License considers knowledge 
> transfer as creating a derivative work, people can still circumvent 
> this by publishing generated content as a dataset without disclosing 
> which model was used (e.g., a model with an open source requirement).
>
> I am also considering whether we could narrow this clause so that it 
> only applies when distributing a collection of outputs, requiring 
> disclosure of the info of original model and its license. Any 
> suggestions on this? Thanks.

Yes, I think you are at least being overinclusive, because you are 
requiring for everyone who creates outputs, even text, audio or visual 
works. I'm not sure how you would even determine compliance. If I have 
some AI images and some Getty Images as illustrations for my blog posts, 
what if I put in my Terms of Use "some images created using AI." Am I in 
breach or not?

As to using a model for training, I'll leave it to others who are 
better-versed in AI than I am. But just being a good practice doesn't 
mean that it must or should be required by the license. If a model 
doesn't give me adequate information about its provenance, then I expect 
good practice says be very wary of it. Having a license that requires 
that you tell me if AI was used doesn't mean someone actually will tell 
me; licenses aren't self-enforcing. The licensee could have simply 
elected to be in breach of the license and no one, including the 
licensor for purposes of enforcement, will be any the wiser.  So, in my 
opinion, it's greater friction for no benefit.

Pam


Pamela S. Chestek (in my personal capacity)
Chestek Legal
PLEASE NOTE OUR NEW MAILING ADDRESS
4641 Post St.
Unit 4316
El Dorado Hills, CA 95762
+1 919-800-8033
pamela at chesteklegal
www.chesteklegal.com

>
>
> Best,
> Moming
>
> _______________________________________________
> The opinions expressed in this email are those of the sender and not necessarily those of the Open Source Initiative. Communication from the Open Source Initiative will be sent from an opensource.org email address.
>
> License-review mailing list
> License-review at lists.opensource.org
> http://lists.opensource.org/mailman/listinfo/license-review_lists.opensource.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20250302/c326d0c9/attachment.htm>