Industry Leaders Support Defining Open-Source AI

30.10.2024

The Open Source Initiative (OSI) recently introduced a framework defining open-source AI, unveiling it as the Open Source AI Definition (OSAID) at the All Things Open conference. This definition represents years of research, international collaboration, and a community design process, offering a detailed outline to assess whether AI systems can be truly considered open-source.

The OSI, respected globally as the authority on open-source definitions, developed this framework with input from industry stakeholders. It establishes that open-source AI must adhere to four key freedoms: unrestricted usage, transparency for study, modifiability, and shareability. For an AI to qualify, users must have access to essential components like data information, source code, and model parameters to make modifications.

A notable requirement of this framework is transparency regarding training data. Mozilla’s Ayah Bdeir, who leads AI strategy, acknowledges that while the standard may not be perfect, it strikes a balance between idealism and practicality, enabling skilled users to recreate similar systems with disclosed data.

Endorsed by organizations like the Digital Public Goods Alliance and EleutherAI Institute, the OSAID highlights the importance of access to data and code, ensuring transparency and fostering research and innovation in the AI field. OSI Executive Director Stefano Maffulli noted that while the process faced challenges, the resulting framework aligns with the project’s objectives and serves as a foundation for continued development.

Although the OSAID does not specify a particular legal mechanism, it opens the door for potential licenses and legal tools to ensure accessibility, setting a new standard for open-source AI as the industry evolves.