The LNM is presented as the first open-source generative AI model trained exclusively on data derived from the natural world and obtained through lawful and ethically responsible means. More specifically, the artist and his team secured the necessary permissions to use data originating from tropical rainforests, coral reefs, and other ecosystems. These data were either collected directly by the artist and his collaborators or obtained through partnerships with research and cultural institutions, including the Smithsonian Institution, the Cornell Lab of Ornithology, the Getty, iNaturalist, and the Natural History Museum in London. [1]
The approach underlying the LNM differs fundamentally from that adopted in the development of many widely used generative AI models, which have allegedly been trained on musical, literary, visual, and other copyright-protected works without the prior authorization of the relevant rightsholders. The LNM therefore breaks new ground by demonstrating that generative AI systems can be developed using lawfully sourced and properly licensed training material.
It also shows that the creation and use of large-scale generative models whose training complies with intellectual property law is not merely an abstract aspiration but a technically achievable objective. Consequently, the demand made by collective management organizations and other rightsholders that AI model providers obtain the necessary licences is not only reasonable and fair, but also practically feasible, provided that those AI model providers are willing to adapt their policies and business models accordingly.
Zoi Mavroskoti
Attorney-at-law, Dr. Juris
* The title of this article is a paraphrase of the title of John Berendt’s book and its film adaptation of the same name, directed by Clint Eastwood, Midnight in the Garden of Good and Evil. In the story, midnight is regarded, according to local beliefs, as the most ambiguous hour, when the balance between good and evil is precarious and fragile.
[1] https://dataland.art/about/large-nature-model , https://dataland.art/faq
[2] https://dataland.art/about