Open Source AI: Propagation of Open Source Licenses in the Age of AI

Open source AI models have the potential to foster innovation and technological progress. Nevertheless, the definition of “open source” in the AI space is hotly discussed. A related – yet no less central – issue concerns the propagating effect of copyleft (or ShareAlike) clauses embedded in training data or upstream code. Do these clauses require downstream AI models, systems and their output to be released under the same open source terms, e.g. as copyright derivatives? This has major repercussions as such propagation would render entire AI projects fully open. The presentation will examine this question and find that this is presently unlikely to be the case, save for training data. Further to this finding, and in order to protect the effectiveness of ShareAlike clause, it will thus advocate for a new definition of copyright derivatives specific to the AI-context.