Loading...
In 2001, Jimmy Wales launched Wikipedia with an idea that every expert said was impossible: let anyone on the internet edit an encyclopedia, with no credentials required. Britannica's editors laughed. Academics predicted it would devolve into vandalism and misinformation within months. Twenty-five years later, Wikipedia has over 60 million articles in 300 languages, is consistently ranked among the top ten most-visited websites on Earth, and studies have found its accuracy rivals traditional encyclopedias in many domains. How did a system with no central authority, no paid editors, and no gatekeepers produce something this reliable? The answer lies in an extraordinary amount of invisible churn. On any given day, Wikipedia processes over 250,000 edits. Vandalism is typically reverted wit...
Popular framing: Wikipedia works because of the wisdom of crowds — millions of people collectively editing toward truth produces a reliable encyclopedia almost as a natural byproduct of participation.
Structural analysis: Wikipedia's reliability is produced by a specific structural dynamic: high-volume churn acts as an adversarial filter that surfaces contested claims, while a small power-user core enforces citation norms that anchor the process to external evidence. This is an irreducible sociotechnical system — reliability is not in the content but in the ongoing process that generates it. Network effects amplify both the quality mechanism and its failure modes simultaneously. The 'Dunbar Number' of the admin community—the fact that a relatively small group of ~1,000 people actually manages the entire project through social norms.
The popular framing implies reliability is stable and passive — a property of the artifact. The structural reality is that reliability is active and fragile, dependent on sustained participation, norm internalization, and adversarial diversity. This gap matters because it produces false confidence: if reliability is 'baked in,' there is no urgency to address editor decline, demographic gaps, or AI content contamination — all of which threaten the process, not just the product.
