We're a big step closer to defining open source AI - but not everyone is happy
The open-source community is actively establishing a formal definition for open-source AI to address the growing discrepancy between marketing claims and actual transparency. Major organizations like the Open Source Initiative and Mozilla argue that existing software licenses are ill-suited for AI, necessitating a new standard. This effort aims to counter "open washing," where companies label proprietary models as open, and to provide lawmakers with clear criteria to regulate AI risks and protect consumers. Central to this new framework is the Model Openness Framework, which categorizes AI systems into tiers of transparency. This approach acknowledges the practical complexities of AI development, such as legal and privacy barriers surrounding training data. By distinguishing between fully reproducible models and those with restricted datasets, the definition seeks to balance ideal open-science principles with the realistic constraints of data sharing, ensuring that claims of openness are measurable and consistent. This initiative is highly relevant to open data as it sets a precedent for how data provenance and accessibility define openness in complex technological ecosystems. By rigorously evaluating not just the model weights but also the training data and usage rights, the emerging standard emphasizes that true openness requires full transparency of the entire system. This collaborative, multi-stakeholder process ensures that the resulting definition promotes genuine innovation and accountability, shaping the future of ethical AI development.
Source: zdnet.comPublished on 2024-08-24