On Opt-Out Language for AI and Text Mining
Open Future Policy Brief
A newly-released policy brief from this European-based group serves to "facilitate the development of robust, interoperable standards for machine-readable rights reservations on copyrighted works used for AI training." The group seeks to develop a common vocabulary for machine-readable opt-outs:
...it seems desirable that any compliance policies should be based on a vocabulary that distinguishes between a full TDM opt-out (no-tdm) and an opt-out from generative AI training that applies to the use of works for no-generative-ai is a more specific version of the no-tdm opt-out, for the purpose of training generative AI models, either of these two would signal an opt-out to the model training the subset of AI models described in recital 105 of the AI act (no-generative-ai).
The authors view this as a fundamental and necessary first step in developing acceptable guard-rails.
On its most basic level, a vocabulary would provide a standardized set of terms that describe types of use of protected works and other subject matter in the context of text and data mining and the training of AI models.
The proposal is not created in the expectation that the vocabulary would be used primarily by either creators or rightsholders; rather the expectation is that providers of opt-out systems and solutions would implement the vocabulary.
More Details Here
📢 A proposal for a vocabulary for machine-readable opt-outs from AI Training and Text and Data Mining. @paulk.bsky.social proposes a vocabulary aimed at making opt outs from AI training more interoperable: openfuture.eu/publication/...
— Open Future (@openfuture.bsky.social) March 10, 2025 at 8:13 AM
[image or embed]