The Open Source Initiative (OSI) released Version 1.0 of its Open Source AI Definition (OSAID) after two years of development with contributions from global experts. The OSAID sets criteria defining open-source AI, aiming to bring clarity to the concept and establish benchmarks for transparency and accessibility in AI. This framework applies traditional open-source principles to AI, defining it as a system that permits users to freely use, study, modify, and share the technology.
For an AI system to be recognized as open-source under the OSAID, it must provide users with unrestricted access to several core components, each aimed at ensuring transparency and modifiability across the AI lifecycle. These requirements include:
- Data Information: Sufficient detail about the data used to train the AI model, including its source, selection, labeling, and processing methodologies.
- Code: Complete source code that outlines the data processing and training, under OSI-approved licenses.
- Parameters: Model parameters and intermediate training states, available under OSI-approved terms, allowing modification and transparent adjustments.
The framework notably does not mandate a specific licensing mechanism, especially for data and parameters where legal questions regarding copyright protections continue to evolve. Yann Dietrich wrote:
And now, we do have an approved OSI definition of Open Source AI with several important points: Data (sufficiently detailed information about the data used to train the system so that a skilled person can build a substantially equivalent system and not necessarily all such data to be made available) – Code (traditional) – Parameters or weights. Please note that data and parameters can be made available through OSI-approved terms (and not license) as the lack of copyrightability/copyright protection may limit the use of the license mechanism.
The release has received support from organizations, which view it as a significant step toward establishing standardized practices for open-source AI. However, industry voices have raised questions on the scope and focus of the OSAID, particularly around safety and ethical considerations. For example, Seth Khon commented:
I appreciate OSI’s effort to promote transparency and openness with the Open Source AI framework. However, I am concerned that it lacks explicit considerations for safety. While open access and modifiability are important, AI poses unique risks that transparency alone does not mitigate. I believe the framework could benefit from integrating safety standards or guidelines to address responsible and ethical use, especially in high-stakes applications.
With OSAID now available, OSI aims to provide a foundation supporting both innovation and accountability, equipping developers with standards that can foster both ethical AI development and effective oversight.