WORK

The Tragedy of the AI Data Commons

Max Fang's February 2025 Stanford working paper applying Ostrom's institutional framework to AI training data — the most rigorous contemporary analysis framing AI training data as a commons subject to regime change without community participation.

Max Fang's February 2025 Stanford working paper, titled "The Tragedy of the AI Data Commons," applies law-and-economics methodologies alongside Ostrom's design principles to frame AI training data as a commons undergoing unilateral regime change. The paper's central argument: the training data from which large language models learn was contributed by millions of individuals under governance arrangements designed for human consumption — the norms of the open internet, terms of service of social platforms, licensing frameworks of academic publishing. The appropriation of this data for AI training represents a fundamental shift in the terms under which the resource is used, undertaken without the participation of the community whose contributions constitute the resource.

In The You On AI Encyclopedia

The paper is significant because it brings Ostrom's framework to bear on the training data question with the analytical rigor both the economic and institutional dimensions require. Prior debates had typically collapsed into the market-versus-state binary: privatization through property rights, or regulation through state oversight. Fang's application of the commons framework opens analytical space for the third institutional possibility that Ostrom's research documented.

The paper's diagnosis is structural rather than moral. The regime change is not characterized as ethical violation but as institutional transformation — a shift in governance arrangements that the existing institutional framework was not designed to accommodate. This framing is consistent with Ostrom's approach: institutional analysis identifies what is happening, what the consequences are, and what alternative arrangements might address the situation, without assuming a particular normative position about what should have happened.

Its prescriptive direction follows Ostrom's framework in recommending community-based governance arrangements developed by the contributors themselves — not as moral imperative but as institutional possibility that the dominant binary forecloses. The practical challenges of organizing millions of globally distributed contributors are addressed directly rather than dismissed.

Origin

Fang's paper emerged from a research program at Stanford examining the economic and institutional dimensions of AI governance. Its reception within the Ostrom-influenced commons governance community has been substantial, and it has been cited extensively in subsequent work applying institutional analysis to AI.

Key Ideas

Regime change framing. The paper's central move is characterizing AI training data appropriation as a commons regime change rather than a property-rights or regulatory question.

Institutional analysis over moral argument. The diagnosis is structural; the normative implications follow from institutional consequences rather than ethical principle.

Third-option prescriptive direction. Community-based governance arrangements are presented as an institutional possibility the dominant binary forecloses.

Practical challenges addressed. The paper engages seriously with organizational challenges of scale and jurisdictional diversity.

Explore more

Browse the full You On AI Encyclopedia — over 8,500 entries