Despite high demand for transparency from rights holders, an analysis by the Rights Alliance shows that AI providers are becoming less transparent.
Transparency about what content has been used to train artificial intelligence is crucial for rights holders to enforce their rights.
However, none of the most prominent AI providers (so-called State of the Art models) seem to be sufficiently transparent about what their AI models are trained on. This is the conclusion from the Rights Alliance in a new report based on a systematic review of the largest AI models in text, image, video and music generation.
Transparency is decreasing
Despite an increasing focus on the need for public transparency, the analysis shows that the transparency of AI providers is decreasing over time. In addition, terms such as ‘open source’ that give the impression of openness and transparency are being used by modelling providers, even though there is no transparency in the training data.
Read the report on AI modelling providers’ training data transparency and copyright enforcement here
The analysis is based on a systematic review of the major AI models, mapping as much as possible what the models are trained on and where the content originates from, as well as whether AI developers’ transparency gives rights holders sufficient opportunity to identify whether their work has been used and thus to enforce their rights.
Read also: The Books3 case highlights the need for transparency in the training of artificial intelligence