The Open Source Initiative (OSI) launched the Open Source AI Definition Framework (OSAID), establishing 4 essential freedoms for open-source AI systems.
According to AI news, at the All Things Open conference, the OSI board, a global authority on open-source definitions, officially announced the first open-source AI definition framework after years of collaborative research efforts.
The OSAID definition framework was developed by OSI through close collaboration with industry stakeholders, affirming that open-source requirements apply to all components of AI systems, including fully functional systems, models, weights, and parameters.
According to OSAID, an open-source AI system must be provided under terms that ensure four essential freedoms:
Use the system for any purpose without needing permission.
Research how the system works and examine its components.
Modify the system for any purpose, including changing the system's output.
Share the system for others to use with or without modifications, for any purpose.
Stakeholders include OSI, the Digital Public Goods Alliance (DPGA), the EleutherAI Institute, and many other experts and organizations around the world. The announcement of OSAID is crucial for promoting transparent and collaborative development in the field of AI.
These freedoms apply to both the complete system and its individual components. A prerequisite for exercising these rights is access to the preferred format for making modifications, including detailed data information, complete source code, and model parameters.
Carlo Piana, Chair of the OSI Board, stated: 'The co-design process has led to a well-developed, comprehensive, inclusive, and fair version 1.0 of the open-source AI definition.' He also expressed confidence that this definition meets the standards of open source.
One of the most important requirements of the definition framework is that open-source models must provide complete information about training data, ensuring that 'a skilled person can reproduce a functionally equivalent system using similar data.'
Ayah Bdeir, head of AI strategy at Mozilla, acknowledged that while this approach may not be perfect, it represents a practical compromise between purity of idea and practical implementation.
The Digital Public Goods Alliance (DPGA) has expressed support for OSI's leadership in defining open-source AI. Liv Marte Nordhaug, the CEO of DPGA, confirmed that her organization will integrate this foundational work into updates of digital public goods standards for AI applications.
On the other hand, OSI Executive Director Stefano Maffulli acknowledged the challenges in the development process – such as debates over open-source transparency, but the final result aligned with the project's original goals.
He also stated: 'This is the starting point for an ongoing effort to engage with communities to improve the definition over time.'
OSAID does not require a specific legal mechanism to ensure that model parameters are freely provided to everyone, although this may involve licenses or legal tools. This aspect is expected to become clearer over time as the legal system addresses these open-source AI systems.
