The Open Source Initiative (OSI) has launched the first official Open Source AI Definition (OSAID), setting out a clear framework to determine if AI systems can be considered open-source. Announced at the “All Things Open” conference, this definition results from years of research, numerous international workshops, and a year-long community design process involving experts worldwide.
As the authority on open-source definitions, OSI collaborated with various industry stakeholders to create a standard that applies equally to complete AI systems, models, weights, parameters, or other essential components. The OSAID ensures open-source AI projects meet four essential freedoms:
- Freedom to Use: Users can operate the AI system for any purpose without needing permission.
- Freedom to Study: The AI’s components and workings can be examined and understood.
- Freedom to Modify: The system can be altered for various purposes, including changing its output.
- Freedom to Share: Users can distribute the system, with or without modifications, for any purpose.
For these freedoms to be meaningful, OSAID specifies that users must have access to detailed data, complete source code, and model parameters, ensuring transparency and modifiability. The definition establishes guidelines for a fully open-source AI, from functional systems down to individual elements.
Framework Development and Industry Endorsement
“The collaborative process behind OSAID v1.0 was thorough and balanced,” OSI board chair Carlo Piana noted, affirming the definition aligns with the open-source standards defined by the OSI. A notable element of the framework is that open-source models must provide comprehensive training data, enabling skilled users to recreate comparable systems. This was described by Ayah Bdeir, Mozilla’s AI strategy lead, as a practical compromise that balances idealistic and realistic goals.
The framework also garnered support from organizations like the Digital Public Goods Alliance (DPGA), which plans to adopt OSI’s standards into their Digital Public Goods Standard for AI. The EleutherAI Institute, dedicated to open-source AI research, also endorsed OSAID, with Executive Director Stella Biderman noting it meets the needs of independent AI researchers and promotes transparency among top AI developers.
Transparency and Continued Development
OSI Executive Director Stefano Maffulli acknowledged the challenges faced in developing OSAID, noting that heated debates often highlighted the diverse perspectives within the community. Despite this, Maffulli expressed optimism about the ongoing collaboration that will refine the framework as AI evolves.
The OSAID framework does not specify a mandatory legal approach for making model parameters freely accessible, but future developments in legal frameworks may provide clearer guidelines. For now, OSI anticipates further community engagement to fine-tune the definition and adapt it as AI technologies and practices continue to advance.
For More latest tech news, follow Gadgetsfocus on Facebook, and TikTok. Also, subscribe to our YouTube channel.