The Case for Public Investment in Open Source AI

Open Source AI as a Public Good

Governments, private companies, and nonprofits should invest heavily in free and open source AI to ensure that the foundational knowledge of the future remains a public commons rather than a proprietary asset. David Siegel, a computer scientist and co-founder of Two Sigma, argues that the current trend toward closed frontier models risks locking down scientific progress and creating an "oracle" system where users must trust results they cannot audit.

The Parallel Between Software and AI

Open source AI is the modern equivalent of the free software movement started by Richard Stallman. In the 1980s, the shift toward open source software proved that shared bodies of knowledge grow stronger than proprietary ones. This transparency led to the creation of essential infrastructure like GCC and GNU/Linux, which power much of the modern internet.

Siegel notes that the same arguments used to oppose open source software—specifically that "security through obscurity" is safer—were proven wrong. Transparency allows a global community to identify and fix vulnerabilities more effectively than a closed system. He argues that the same logic applies to AI: closed models are not inherently safer; they are simply less transparent.

The Risks of Closed Frontier Models

When a small number of companies control the most advanced AI systems, several systemic risks emerge:

• Stagnation of Scientific Progress: If future science relies on AI, locking that AI inside private firms risks locking down scientific discovery itself.
• Lack of Auditability: A closed model's explanation for an answer is a "plausible story assembled after the fact," not a record of computation. Without access to the underlying code and data, users (including doctors, engineers, and judges) cannot truly audit the reasoning process.
• Knowledge Monopolies: Siegel compares closed AI to a library where a few companies decide which books are available and can quietly rewrite the content, controlling the terms of access to human knowledge.

The Distinction Between Open Weights and Open Source

Siegel emphasizes a critical technical distinction: there is a difference between the code that runs a model and the code that built it.

Many models marketed as "open" today provide the weights (the numbers that produce intelligence) but withhold the training code and the datasets used. Siegel argues that providing weights without the training methodology is a "favor, not a commitment," as it provides a compiled result without the recipe. True open source AI requires transparency in both the model weights and the training process.

Proposed Path Forward

To maintain a credible open alternative to proprietary giants, Siegel proposes three specific interventions:

1. Public Compute Grants: Governments should provide compute resources for open research.
2. Philanthropic Support: Increased corporate and nonprofit funding for universities and non-commercial entities.
3. Open-by-Default Mandates: A requirement that any AI developed with public funding must be open source.

Community Perspectives and Counterpoints

Discussion among technical peers highlights several tensions regarding the feasibility and nature of open source AI:

Economic and Structural Challenges

Some argue that profit incentives will always give commercial AI an edge. As one commentator noted, "Goodwill and part-time contributions cannot reliably compete with livelihood and profit incentives." Others suggest that the scale of frontier AI is more akin to a scientific research program (like the Manhattan Project or CERN) than a traditional software project, suggesting that public funding would need to be massive and centralized rather than distributed via small grants.

Alternative Funding and Governance Models

To incentivize open development, some suggest targeted "inducement prizes"—cash rewards for the first open model to hit specific benchmarks on limited hardware (e.g., 16GB-128GB VRAM). Other suggestions include the creation of member-owned cooperatives to ensure public control over the technology.

Skepticism of Government Intervention

Critics of public investment argue that government funding could be wasteful or that the priority should be taxing GPU acquisitions for large firms rather than subsidizing open research. There is also the view that the market will naturally drive the cost of intelligence down, making frontier-scale models unnecessary for most practical applications.