Under the #AIAct, there is now a legal obligation (Art. 53 (1) d) to disclose information about content used to train GPAI models according to a template prepared by the AI Office. Over the next 12 months, we're going to see some heated debates about what this template should look like. Civil society and researchers should use this opportunity to push for more transparency.
As a contribution to these discussions, together with Maximilian Gahntz and Paul Keller, and with input from a group of top experts, I wrote a paper describing the categories of rights and "legitimate interests" that GPAI training data transparency would serve (tl/dr - it's about more than copyright). The paper is accompanied by a blueprint for the upcoming template.
Special thanks to Yacine Jernite, Shayne Longpre, Abeba Birhane, Stefan Baack, Kris Shrishak, Aviya Skowron, Alex Hanna, Ph.D. Mark Dingemanse, Frederike Kaltheuner, Claire Pershan, Michał “rysiek” Woźniak, Natali Helberger, João Pedro Quintais, Toni Lorente
🔎 Transparency of the data used to train AI models is a prerequisite for improving accountability in AI development and can strengthen people’s ability to exercise their fundamental rights. Under the AI Act, there is now a legal obligation to disclose information about content used to train GPAI models according to a template prepared by the AI Office.
As the Office takes its final shape, we share a policy brief arguing for a robust transparency of content used to train GPAI models.
This paper and the accompanying “Blueprint of the transparency template that the AI office is tasked to develop” was created in collaboration with Mozilla Foundation, drawing on input from experts.
You can access the brief using the link in the comment 👇
#AI #GPAI #AIAct #DigitalRights
Link to the brief 👉 https://openfuture.eu/publication/towards-robust-training-data-transparency/