Recent research has exposed a concerning trend among major tech companies, including Meta and Google, who tout their generative AI systems as "open" while evading substantive scrutiny. This practice, termed "open-washing," misleads the public and avoids genuine transparency in the development and deployment of AI models.
Andreas Liesenfeld and Mark Dingemanse from Radboud University's Center for Language Studies conducted a comprehensive survey of 45 text and text-to-image models claiming openness. Their findings, published at the ACM Conference on Fairness, Accountability and Transparency (ACM FAccT 2024) and highlighted in Nature, reveal a stark reality: many corporations misuse terms like "open" and "open source" for marketing without providing adequate access to critical aspects such as source code, training data, or system architecture.
The study underscores disparities in true openness, contrasting major tech giants' ambiguous claims with smaller entities like AllenAI and BigScience Workshop + HuggingFace, which actively document and open their AI systems to scrutiny.
The introduction of the EU AI Act further complicates matters by offering exemptions for "open source" models without a clear definition, incentivizing open-washing to avoid stringent regulations and public oversight. According to Liesenfeld, clarifying what constitutes genuine openness in generative AI is crucial, advocating for a nuanced understanding that acknowledges openness as a multifaceted and gradational concept.
Dingemanse emphasizes the importance of transparency in AI to foster trust and comprehension, particularly regarding the capabilities and limitations of AI systems like ChatGPT, which have faced scrutiny for their training data and performance claims.
This research builds a compelling case for redefining openness in AI, advocating for rigorous standards that promote innovation, uphold scientific integrity, and enhance societal trust in AI technologies. Radboud University's Faculty of Arts has similarly called for enhanced AI literacy among researchers, reflecting a broader effort to navigate the complexities of generative AI responsibly and ethically.
