Liesenfeld A, Dingemanse M. Rethinking open source generative AI: open-washing and the EU AI Act. FAccT '24: Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. 2024; Pages 1774-1787. https://doi.org/10.1145/3630106.3659005
Summary: This paper examines the rise of generative AI systems labeled as "open source" and critically assesses the authenticity of such claims. In light of the EU AI Act, which treats open-source AI differently, the authors propose a 14-dimension framework for evaluating openness, covering aspects such as dataset transparency, licensing, and technical documentation. After analyzing over 45 generative AI systems, they found that many models are only superficially open, often withholding key information like training data. The study argues that openness in generative AI is composite and gradient, requiring nuanced assessment to avoid "open-washing" and ensure regulatory compliance and accountability.
Comments