It’s encouraging to see that the market is moving towards licensing deals for dataset creation, but the current model, where tech companies make one-time payments and never compensate creators again, is fail. Clearly unsustainable by not accounting for the ongoing value generated from the datasets. 

Compensation mechanisms cannot be put in place with the current modes because the data was initially trained on unlicensed IP. But it should be clear that no amount of new deals can rectify the fundamental compensation issue, or come close in depth and volume. 

Content creators should be actively involved in the dataset creation process, establishing transparent licensing agreements that include provisions for ongoing royalties, rather than one-time payments. By involving creators from the start, allowing them to review and approve the use of their work, and maintaining open communication builds trust through fairness and transparency. Music licensing is a good analog.