Re: Google opening the door to a discussion about AI opt-out

Hi, 

The current situation is that content owners either control their websites (this is the case for newspapers) or have close ties with the organizations that host their content (this is the case for publishers that distribute books on bookseller's websites, also the case for authors of scientific articles who delegate the distribution of their articles to companies like Cairn.info <http://cairn.info/>). 

So, in the majority of existing professional cases, applying the opt-out at the level of the website is a very good solution, and there is no problem linking the opt-out action to the organization that has the right to rule an opt-out.

In France, organizations representing newspapers, music and cinema platforms, scientific publications and books (*) have decided to use TDMRep for signaling their opt-out choice to TDM and AI actors, and proposing licensing solutions via our ODRL structure. They are now looking to attract the attention of these actors, and the discussion now starting with Google will make this company aware of its duty vs European companies using this legally backed opt-out solution. 

(*) the organizations are GESTE, CNC, SNE, CNL. 

I hope that other European countries will soon follow this path. 

Newspapers, publishers, and booksellers are not technical experts. They can easily raise a flag in an HTTP or HTML header, but most won't take the time to setup digital signatures and verifiable credentials. For those who can, or can spend the money to get a contractor doing the job, that's fine. But it must not be required to set up a complex system in order to signal that you don't want your content to be used for TDM/AI. Simple is beautiful. 

We'll see what happens to the discussion with Google, and I hope that other big players (OpenAI first) will also open a discussion. If the adoption of a profoundly modified robots.txt is the price for getting them on board, my position is that it is acceptable. 

Meanwhile, I'll soon propose an integration of the TDMRep flag and policy url in EPUB files, as this is something requested by members of this group.

Re. ISCC codes, we'll see what happens with the EU AI Act and the requirement to publish the list of every resource used in a training set. If this happens, I firmly believe that ISCC codes are the way AI actors should identify these resources at the time of publication. 

Best regards
Laurent


> Le 31 oct. 2023 à 20:19, Sebastian Posth <posth@iscc.foundation> a écrit :
> 
> Expanding on Leonard's point, not only is it important to recognize that the content owner might not always be the same entity that owns, manages, or controls the website where the content is shared, especially in cases like content distribution on social media. It is a necessity that the declaration is bound to the content and not the location where the content is published. 
> 
> Furthermore, a critical consideration is whether the entity making a claim, such as an opt-out, is both identifiable and possesses the legal authority to enact that claim. 
> 
> The use of digital signatures and/or verifiable credentials ensure that declarations are verifiable and facilitate certified and accurate attribution (in a scalable and machine-readable way) – which currently is supported by the C2PA system and ISCC declarations, externally inseparably binding the opt-out to the content-derive Identifier. 
> 
> With best regards, 
> Sebastian
> 
> 
> On Mon, 30 Oct 2023 at 03:40, Leonard Rosenthol <lrosenth@adobe.com <mailto:lrosenth@adobe.com>> wrote:
>> I also attending the webinar and am glad to see Google getting into the act and doing so by taking input and wanting to work through a standards process…All good!
>> 
>> The current problem with the robots.txt direction is the same as the current TDM specification – which is that it assumes that the owner of the content also owns/manages/controls the web site on which it is hosted.  That may help the professional publisher who maintains their sites, but it’s not helpful for the average user putting their content up on social media, stock image services, etc.
>> 
>>  
>> 
>> Leonard
>> 

Received on Thursday, 9 November 2023 10:03:13 UTC