Public AI Network 2024 white paper
Citation
Public AI Network. “Public AI: Infrastructure for the Common Good.” White paper, August 10, 2024; updated October 2024. DOI 10.5281/zenodo.13914560. Record: https://zenodo.org/records/13914560 (record page saved as raw/shared/2026-10-29 Public AI Network 2024 white paper record.md). PDF from https://publicknowledge.org/wp-content/uploads/2024/10/2024-Public-AI-White-Paper.pdf, because Zenodo serves the file in a form the download script does not read. CC BY 4.0. Read 2026-10-29.
Bottom line
- The Public AI Network argues for public AI infrastructure with three features: public access, public accountability and permanent public goods.
- It asks states to fund public options at each layer of the AI stack, starting with compute. It names CalCompute and Empire AI as local public compute efforts.
Background
The paper was first released in August 2024 for the “Towards a More Public AI” event at the Library of Congress. The October 2024 revision renamed the third feature from “Permanently Public” to “Permanent Public Goods” (front matter, Version History).
Key claims
- Three features: “Public Access”, “Public Accountability” and “Permanent Public Goods”. (Executive summary)
- “Public AI earns trust by ensuring ultimate control of development rests with the public, giving everyone a chance to participate in shaping the future.” (Executive summary, Public Accountability)
- “Public AI is funded and operated in a way to maintain the public goods it produces permanently, enabling innovators to safely build atop its firm foundation.” (Executive summary, Permanent Public Goods)
- “States have a leading role to play in creating Public AI—through ‘public investment’ and an empowered ‘public sector’”. (note 2)
- Public options: “Public investment can also lower barriers to entry by producing and supporting public options in each of these horizontal layers, starting with the most expensive components that are barriers to entry”. (Horizontal platforms)
- Compute: “Public compute initiatives that aim to protect and provide the significant computational capacity that researchers, academia, and small firms need to build competitive AI models.” (Horizontal platforms)
- CalCompute: “Current examples of such initiatives include national efforts like Canada’s AI Compute Access program and the National AI Research Resource (NAIRR) in the US, as well as more local efforts like CalCompute and Empire AI, designed to support local economic development.” (Horizontal platforms)
- Funding risk: public model builders “in the absence of public compute must release their work as open source in order to make their work publicly accessible, but now find that it leaves them dependent on external funding—while hosting providers profit from their models.” (Permanent Public Goods)
- “For decades, states have systematically hollowed out the public sector’s technical” capacity. (Public sector section; the sentence breaks in the copy)
Evidence and method
A policy white paper. It argues from cases and cited literature, with no new data.
Limits
- It is on Zenodo with a DOI.
- It names CalCompute as an example, so it does not assess CalCompute.
- Only the acknowledgements list the authors, and the copy breaks them up. The paper asks readers to cite the Public AI Network.