Advancing Open Source AI: Definitions, Standards, and Global Implementation for a Sustainable Future
- 13 November 2024
By Amreen Taneja, DPG Standards Lead at DPGA, Stefano Maffulli, Executive Director at Open Source Initiative (OSI), and David Manset, Senior Coordinator Open Source Ecosystem Enabler (OSEE) at ITU/BDT
Open source AI is evolving rapidly, with global efforts to establish clear governance, standards, and implementation strategies. This blog explores three interconnected initiatives shaping the future of open source AI: the Open Source Initiative’s (OSI) ongoing Open Source AI Definition (OSAID in version 1.0); the co-design process to extend Digital Public Goods (DPG) Standard to better define open source AI; and the ITU’s Open Source Program Office (OSPO) work on a reference implementation for the global south. By aligning these efforts, the international community can build an inclusive, ethical, and impactful AI ecosystem that serves both innovation and public good.
The OSI’s Open Source AI Definition (version 1.0): A Collaborative Approach
The OSI, recognized for setting global open source standards, has released version 1.0 of its Open Source AI Definition (OSAID). This initiative aims to provide clarity and structure around what constitutes open source AI, establishing principles for AI projects that emphasize transparency, openness in development, and adherence to open source licenses.
The version 1.0 was released on October 28 and actively seeks community support. The goal is to ensure the definition remains inclusive and adaptable to the needs of diverse stakeholders, particularly from the global south, where different socio-economic and technological contexts must be considered. Whereas the co-design process ended, OSI’s collaborative approach invites developers, policymakers, organizations, and local communities to contribute to the next phase in endorsing and developing operating manuals to evaluate AI systems. By integrating a wide range of perspectives, OSI aims to build a framework that supports innovation while aligning with the values of inclusivity and responsible AI development. More information can be found here.
Extending the DPG Standard for Open Source AI: Toward a Unified Framework
In parallel, the Digital Public Goods Alliance (DPGA), is updating the DPG Standard to better account for open source AI systems. DPGs are open source solutions designed to address critical development needs, particularly in low and middle-income countries. Recognizing AI’s transformative potential for achieving the Sustainable Development Goals (SDGs), this work will help ensure that AI solutions within the DPG framework adhere to the principles of openness, inclusivity, and responsibility.
The Community of Practice (CoP) on AI as DPGs, an expert group facilitated by the DPGA Secretariat and UNICEF, has been actively engaged in creating recommendations for updating the DPG Standard. Their recommendations include incorporating specific criteria for AI transparency, data governance, and ethical AI practices, ensuring that AI-based DPGs are developed responsibly and inclusively. The CoP has debated various approaches, such as tiered openness levels for AI components (i.e., models, data, and code), but ultimately focused on a binary approach to distinguish open source from proprietary AI.
The CoP’s recommendations extend beyond defining openness. They also include “do no harm” requirements, such as mandatory risk assessments for AI models, evidence of testing for bias and fairness, and environmental impact evaluations using tools like CodeCarbon. These measures aim to align AI development with global ethical standards, ensuring AI technologies prioritize public interest, societal benefits and sustainability. This effort emphasizes the importance of transparency, ethical governance, and inclusivity, inviting further participation from the international community to refine and implement the DPG Standard.
The AI CoP recommendations are currently under assessment by the DPG Standard Council for an independent review process that adheres to DPG Standard Governance guidelines. This process rigorously evaluates proposed changes for technical feasibility and alignment with DPGA’s strategic goals. Following this review, updates from the Council will be published on GitHub for public comment, allowing stakeholders to provide feedback. After considering these suggestions, the Standard Council will finalize the updates to the DPG Standard. Once finalized, the updated DPG Standard will be implemented, ensuring stakeholders have access to the latest version for reference and adoption.
One such fundamental change proposal, which is currently open for public comment on GitHub, introduces mandatory components for AI systems seeking DPG recognition, specifically, to continue requiring open training data for AI systems to be considered DPGs. This proposal remains open for public comment until December 4, 2024, after which community input will be incorporated into the final decision-making process. Once finalized, this updated standard will help ensure that AI DPGs adhere to principles of openness and transparency, further strengthening the impact and accessibility of AI for public good. More information can be found here.
ITU OSPO’s Reference Implementation for the Global South: Building Practical Solutions
Complementing the efforts of OSI and DPGA, the ITU’s Open Source Program Office (OSPO) is developing a reference implementation for open source AI specifically designed for the Global South. This initiative is part of ITU’s broader mandate to enhance digital capacity in low and middle-income countries, making AI technologies more accessible, scalable, and impactful for communities in need.
The ITU OSPO’s reference implementation aims to provide a practical blueprint for deploying AI systems in diverse sectors such as healthcare, agriculture, education, and infrastructure. It prioritizes affordability, adaptability, and scalability, ensuring that AI solutions are tailored to meet the specific needs and challenges of these regions. By focusing on open source principles, the ITU’s initiative not only democratizes access to AI technology but also ensures that these solutions are developed with transparency and ethical considerations.
The working group behind this initiative includes 50+ experts from 10+ international organizations, UN agencies, and local communities, reflecting a collective effort to build AI systems that serve as DPGs. The initiative is closely aligned with the OSI’s Open Source AI Definition and DPGA’s updated standards, ensuring consistency and coherence in the AI governance framework. This interconnected approach allows for a harmonized strategy, enabling countries in the global south to adopt and adapt AI technologies effectively, contributing to a more inclusive global AI ecosystem.
To learn more about the initiative, watch this AI for Good Webinar here
Join the Movement for Inclusive Open Source AI
As these initiatives progress, it is crucial for the global community to engage and contribute. Whether you are a developer, policymaker, academic, or member of a local community, your input is valuable in shaping a future where AI technology benefits all, particularly those in under-resourced and underrepresented regions.
The OSI’s OS AI Definition remains open for public comment, inviting suggestions that can make the framework robust and applicable across different contexts. The DPGA’s extension of the DPG standard welcomes diverse perspectives to ensure that AI development aligns with ethical and sustainable practices globally. Additionally, the ITU OSPO’s reference implementation seeks collaborative efforts to build practical AI solutions that respond to the unique needs of the global south.
A Unified Path Toward Ethical and Inclusive AI
The efforts led by the OSI, DPGA, and ITU are interconnected pieces of a larger movement toward a unified and inclusive open source AI ecosystem. By defining clear standards, developing ethical frameworks, and creating practical implementations, these organizations are paving the way for AI that not only supports innovation but also aligns with the values of social good, sustainability, and accessibility.
The path forward is collaborative. By engaging in these processes and contributing expertise, stakeholders from all sectors can help build an AI ecosystem that serves humanity equitably, ensuring that AI technologies empower societies to address their most pressing development needs. Now is the time to get involved and help shape the future of open source AI for the betterment of all.
