Hugging Face Pushes for Open-Source Prioritization in Upcoming AI Action Plan
Hugging Face is urging the US government to place a firm emphasis on open-source development in its upcoming AI Action Plan. In a message to the Office of Science and Technology Policy (OSTP), Hugging Face articulated that crafting thoughtful policies can further innovation while keeping AI advancements competitive and aligned with core American values.
With an impressive collection of over 1.5 million public models and a user base of seven million, Hugging Face champions a three-pillar framework for the AI Action Plan:
- Fortifying open-source AI ecosystems: Hugging Face underlines the significance of diverse contributors from various institutions. They argue that bolstering infrastructure like the National AI Research Resource (NAIRR) and investing in open science ignite technical innovation, propelling robust growth.
- Streamlining safe and reliable adoption of AI: The aim here is to spread the benefits of AI technology vertically across sectors. Hugging Face believes that effective, modular AI models require serious investments to encourage widespread adoption, ultimately fostering broader participation across the US economy.
- Enhancing security and standards: Reflecting on decades of practices in cybersecurity, Hugging Face advocates for establishing standards of traceability and disclosure to create a more resilient technology environment.
Open-source: The Heartbeat of AI Development
Hugging Face points out that the foundations of modern AI rest heavily on years of open research, noting that big corporations heavily depend on these contributions. Recent advancements, like OLMO-2 and Olympic-Coder, showcase how open research can lead to developing systems that are on par or even outperform their commercial counterparts - especially in efficiency and domain-specific performance.
Notably, the timeframe to match development has drastically reduced. As they put it, "What once needed over 100 billion parameters just a couple of years back can now be done with models possessing only 2 billion parameters, indicating an accelerated path to parity.”
This trajectory of accessible and collaborative AI signals that open-source approaches hold a crucial role in developing effective AI strategies, supporting both innovation and broader adoption of this technology.
Hugging Face argues that open models, supporting structures, and scientific practices form the bedrock for AI creativity, allowing a vibrant community of researchers and organizations to build upon shared knowledge.
Their platform showcases AI models and datasets from both emerging talents (like startups and universities) and established players (think Microsoft, Google, OpenAI, and Meta). This exemplifies how opening doors to open-source methodologies can speed up progress and democratize AI capabilities.
“The United States needs to assume a leadership role in open-source AI,” asserts Hugging Face, “which can elevate American competitiveness by cultivating an innovation ecosystem that champions a healthy mixture of competition and collaborative innovation.”
Research highlights that open technical systems serve as accelerators for economic impact, potentially turning every dollar invested into enormous value for those leveraging these systems. In fact, around $4 billion invested could yield an astounding $8 trillion for companies utilizing open-source. Furthermore, the absence of open-source contributions could cost countries an average of 2.2% of their GDP.
Back in 2018, open-source effectively fueled between €65 billion and €95 billion of the European GDP, prompting the European Commission to take action for establishing a streamlined process for open-sourcing government software.
Factors Fueling the Shift to Open-source AI
Hugging Face has pinpointed several practical factors driving the move towards commercial adoption of open models:
- Cost efficiency: Creating AI models from scratch can be an expensive venture. Adopting open-source foundations helps in cutting development costs.
- Customization: Organizations can adapt models to suit specific needs rather than being forced into generic solutions.
- Reducing vendor lock-in: Open models offer companies autonomy over their tech stack without tying them to one provider.
- Capability leap: Open models currently rival - and at times even surpass - those developed in closed environments.
These advantages are particularly beneficial for startups and mid-sized businesses, enabling them to tap into advanced tech without incurring hefty infrastructure costs.
Forward-Thinking Policy Recommendations from Hugging Face
Hugging Face underscores the need for policies that support the enhancement and adoption of open AI systems. This includes:
- Boosting research infrastructure: Through the successful implementation of the NAIRR pilot, Hugging Face emphasizes providing researchers with access to computing resources and datasets.
- Dedicating public resources: Ensuring that the public can partake in AI advances through infrastructure can lower innovation barriers for smaller research teams and companies.
- Data access: Building sustainable data ecosystems is essential, especially as more publishers turn towards exclusive licensing deals, potentially limiting quality data access for small developers.
Tackling Barriers to Efficient AI Adoption
Hugging Face acknowledges the considerable obstacles smaller entities face when integrating AI into their operations, particularly the steep costs involved. Despite projections indicating global AI spending reaching $632 billion by 2028, these figures remain daunting for many smaller firms.
Yet, those who adopt open-source AI tools often report better financial outcomes. A recent survey showed that 51% of companies using open-source AI tools realized a positive return on investment, compared to 41% among those opting not to.
As energy constraints loom on the horizon, the necessity for efficient AI adoption becomes even more pressing. It’s projected that data centers' energy consumption could double by 2026, potentially stretching resources thin. While training AI models is resource-heavy, inference tends to consume even more energy. Hence, making AI accessible will require both hardware advancements and scalable software frameworks.
Ultimately, Hugging Face’s outreach to the OSTP aims at establishing an AI Action Plan grounded in open-source principles. By forging ahead with these initiatives, the US can solidify its leadership position, ignite innovation, and ensure that the benefits of AI are enjoyed across all sectors of society and the economy.
