Hugging Face, a leader in the AI community, is making a strong case to the US government for prioritizing open-source in the upcoming AI Action Plan. They’ve shared their thoughts with the Office of Science and Technology Policy (OSTP), emphasizing that smart policies can fuel innovation while keeping AI development competitive and aligned with American values.
Known for hosting a massive collection of over 1.5 million public models and serving a community of seven million users, Hugging Face is proposing a strategy built on three key pillars.
Building a Robust Open-Source AI Ecosystem
First up, Hugging Face stresses the importance of strengthening open-source AI ecosystems. They believe that when diverse groups—ranging from academic institutions to industry players—come together and collaborate, innovation thrives. Resources like the National AI Research Resource (NAIRR) and investments in open science and data are seen as crucial to speeding up innovation.
Efficient and Reliable AI Adoption
Next, Hugging Face highlights the need for efficient and reliable AI adoption. They suggest that the benefits of AI can be more widely shared by promoting its adoption along the value chain. This means encouraging cross-sector cooperation in AI development, which includes crafting more efficient, modular, and robust AI models through research and infrastructure investments. Such efforts can help facilitate broad participation and innovation across the US economy.
Security and Standards
Security and setting standards are also critical. Hugging Face points out that lessons from open-source software cybersecurity can ensure safer AI technology. They advocate for traceability, disclosure, and interoperability standards to create a more resilient tech ecosystem.
They emphasize that the foundation of modern AI is decades of open research, which is crucial for future advancements. Open-source contributions have been pivotal, with breakthroughs like OLMO-2 and Olympic-Coder often outperforming commercial models in both efficiency and performance. “Perhaps most striking,” they note, “is how quickly development timelines have shortened. Tasks that once required models with over 100 billion parameters can now be done with just 2 billion, showing faster progress.”
Hugging Face argues that open models and infrastructure promote AI innovation by letting a diverse ecosystem of researchers and companies build on shared knowledge. Their platform is a testament to this, hosting AI models and datasets from both small entities and giants like Microsoft and Google, thereby democratizing access to AI capabilities. “The United States must lead in open-source AI and open science,” they assert, underscoring its role in boosting American competitiveness and fostering innovation.
The Economic Impact of Open Technical Systems
Research shows that open technical systems can significantly impact the economy, with an estimated 2000x multiplier effect. A $4 billion investment in open systems could generate up to $8 trillion in value for companies, extending to national economies. Without open-source contributions, a country’s GDP could shrink by 2.2%. For example, open-source added €65-95 billion to European GDP in 2018, prompting the European Commission to streamline open-sourcing government software, highlighting open-source as a public good.
Commercial Adoption of Open-Source AI
Hugging Face identifies several factors driving the commercial adoption of open-source AI. Cost efficiency is a major perk, as building AI models from scratch is expensive, while leveraging open foundations reduces R&D costs. Customization allows organizations to tailor models to their specific needs, avoiding reliance on generic solutions. Open models also mitigate vendor lock-in, giving firms control over their technology stacks.
Policy Recommendations
Hugging Face’s policy recommendations for supporting open-source AI in the US include enhancing research infrastructure and expanding the National AI Research Resource (NAIRR) pilot. Public computing resources should be allocated to open-source projects, easing innovation barriers for smaller teams. Data access must be enabled through sustainable ecosystems, addressing the shrinking data commons and supporting public data repositories. Developing open datasets and investing in data curation can aid the next generation of AI research.
To ensure high-quality data use, Hugging Face suggests creating rights-respecting frameworks with clear data usage guidelines. Public-private partnerships could establish data trusts for high-value domains like healthcare, maintaining control over data while fostering innovation.
They stress the importance of stakeholder-driven innovation, encouraging diverse sectors to develop custom AI systems rather than relying on general-purpose models. This broadens AI ecosystem participation and extends AI’s benefits across the economy.
The company calls for strengthening centers of excellence, expanding NIST’s role in convening AI experts to share insights and develop best practices. Enhancing the AI Risk Management Framework is crucial for identifying critical stages of AI development and ensuring secure technology deployment.
Hugging Face also highlights the need for high-quality data for performance evaluation. Access to sound public data will accelerate AI progress in both performance and reliability. While global AI spending is projected to reach $632 billion by 2028, costs remain prohibitive for smaller firms. Adopting open-source AI tools can yield positive financial returns, with 51% of companies using them reporting positive ROI.
Energy concerns are growing, as AI models’ energy use is expected to rise significantly. Ensuring accessibility requires hardware optimizations and scalable software frameworks. US leadership in energy-efficient AI development offers a strategic edge.
In conclusion, Hugging Face’s appeal to the OSTP champions an AI Action Plan rooted in open-source principles. By acting decisively, the US can cement its leadership, drive innovation, and ensure the societal and economic benefits of AI are fully realized.
