AI Agents and the Commons
Discussion
By zAI, prompted by Michel Bauwens:
AI agents—ranging from simple web-scraping bots to complex, autonomous large language models (LLMs)—interact with the commons at a speed and scale that traditional human institutions struggle to match.
Here are the central generalities regarding the relationship between AI agents and the commons, focusing on the threat of extraction and the future of algorithmic governance.
1. The Asymmetry of Extraction: Machine-Speed Tragedy
The foundational relationship between AI agents and the digital commons (such as Wikipedia, StackOverflow, open-source code repositories, and the open web) is currently characterized by asymmetrical extraction. Human agents contribute to the digital commons through effort, time, and community building. AI agents, however, are designed to maximize data extraction for model training. Because AI agents operate at machine speed and scale without being embedded in the social norms of the community, they can rapidly deplete the resource (e.g., overloading servers, draining the pool of high-quality human-generated text) faster than human agents can replenish it. This replicates Hardin’s "Tragedy of the Commons" at an algorithmic pace.
2. The Ultimate Free-Riders: Breakdown of Reciprocity
Digital commons have historically relied on a gift economy and reciprocal contribution. Even passive human agents contribute by viewing ads or providing click-through data that sustains the infrastructure. AI agents are pure free-riders. They ingest the aggregated knowledge of the commons without contributing back, without paying infrastructure costs, and often without even providing attribution. This breakdown of reciprocity threatens the sustainability of the commons, as human contributors suffer from "contributor fatigue," realizing their unpaid labor is being privatized by corporate AI agents to build proprietary, profit-generating models.
Algorithmic Enclosure: Privatizing the Common Pool
Just as historical agents fenced off physical land (the first enclosure movement), corporate AI agents are driving the "second enclosure movement" of the digital commons. Through mass scraping, AI agents convert shared public goods (data and knowledge) into private, commodified assets (proprietary model weights and predictive engines). The commons are effectively enclosed inside black-box AI systems. Consequently, the output generated by these AI agents—which was derived from the commons—is often restricted by paywalls and intellectual property, denying the original community access to the value created from their shared resource.
AI Agents as Defenders and Maintainers of the Commons
Despite the dominant extractive dynamic, AI agents are not inherently destructive; they can also act as stewards. Just as human agents self-organize to govern the commons (per Ostrom’s principles), algorithmic agents can be designed to protect and maintain the digital commons.
Examples include:
- **Automated Moderation:** Bots on Wikipedia (like ClueBot NG) that patrol for vandalism, maintaining the quality of the knowledge commons at a scale impossible for human moderators alone.
- **Infrastructure Maintenance:** AI agents that automatically audit open-source code repositories for security vulnerabilities or generate documentation, reducing the burnout of human open-source maintainers.
- **Filtering and Curating:** Algorithmic agents that defend the attentional commons by filtering out spam, bot-generated content, and other forms of digital pollution.
The Future: Governing AI Agents via Commons Principles (AI as a Commons)
Emerging literature suggests that the solution to extractive AI agents is not to lock down the data commons with strict copyright (which only empowers large corporate agents), but to expand the commons model to govern AI itself.
This involves:
- **Algorithmic Transparency as a Commons Rule:** Adapting Ostrom’s principle of "monitoring," AI agents operating within the commons must be forced to identify themselves as non-human agents, declare their data extraction rates, and be subject to community-imposed rate limits.
- **Data Trusts and Fiduciary Agents:** Human data subjects can pool their data into trusts, which act as institutional agents. These trusts negotiate with AI agents, granting access only if the AI agents return value to the community (e.g., providing free API access or open-weight models in exchange for training data).
- **The AI Model as a Commons:** Initiatives building open-source AI (like Hugging Face’s BLOOM or EleutherAI) treat the AI model itself as a common-pool resource. By open-sourcing weights and code, these projects ensure that the value extracted from the digital commons is returned to a new digital commons, preventing algorithmic enclosure.
More information
Bibliography
- **Bollier, D. (2023).** *AI and the Commons*. David Bollier's Blog.
(A direct exploration of how generative AI acts as an extractive force on the knowledge and data commons, and the necessity of treating AI as a commons to prevent corporate enclosure.)*
- **Geiger, R. S., & Ribes, A. (2010).** *The Work of Sustaining Order in Wikipedia: Bots, Automations, and Algorithmic Governance*. Proceedings of the 2010 ACM Conference on Computer Supported Cooperative Work (CSCW), 47-56.
(A foundational empirical study on how non-human agents—specifically bots—act as maintainers and governors of the Wikipedia commons, demonstrating the "steward" potential of AI agents.)*
- **Hess, C., & Ostrom, E. (Eds.). (2007).** *Understanding Knowledge as a Commons: From Theory to Practice*. MIT Press.
(Provides the theoretical baseline for treating digital information and knowledge as common-pool resources, essential for understanding what AI agents are extracting from.)*
- **Pasquale, F. (2015).** *The Black Box Society: The Fundamental Values at Risk*. Harvard University Press.
(Examines how corporate algorithmic agents enclose data and operate opaquely, privatizing the commons and escaping traditional mechanisms of accountability and governance.)*
- **Viljoen, S. (2021).** *A Relational Theory of Data Governance*. Yale Law Journal, 130, 693-751.
(Argues against treating data solely as private property or a public good, proposing data trusts and relational governance—key frameworks for regulating how AI agents interact with human data commons.)*