Wikipedia’s volunteer editors have enacted a significant policy update, prohibiting the use of large language models (LLMs) for generating or rewriting article content. This move underscores a growing apprehension within the community regarding the accuracy, verifiability, and sourcing integrity of AI-generated text, which often clashes with the platform’s foundational content policies.
The new guideline explicitly states that text produced by LLMs frequently violates Wikipedia’s core tenets. While the direct use of LLMs for article creation or modification is now banned, the policy carves out an exception for limited AI-assisted copyediting. This allowance is strictly confined to suggesting minor edits to an editor’s own work, with the crucial caveat that no new information should be introduced. Editors utilizing these tools are strongly advised to meticulously review all AI suggestions before implementation.
The policy doesn’t detail specific penalties for AI-generated content violations. However, it aligns with existing disclosure guidelines, indicating that repeated misuse could be construed as “disruptive editing,” potentially leading to account blocking or banning, subject to an appeal process.
Key Takeaways
- Wikipedia now explicitly forbids the use of large language models for generating or rewriting article content.
- Limited AI-assisted copyediting is permissible, provided editors review suggestions and no new information is added.
- The policy addresses concerns about AI-generated text containing inaccuracies, fabricated sources, and a lack of accountability.
- The Wikimedia Foundation emphasizes that editorial policies are determined by volunteer editors, not the foundation itself.
- The decision reflects a broader trend of evaluating and regulating AI integration within decentralized and community-driven platforms.
A spokesperson for the Wikimedia Foundation clarified that editorial policies are community-driven, stating, “Wikipedia’s strength has been and always will be its human-centered, volunteer-driven model.” This stance highlights the platform’s commitment to its established collaborative knowledge-building ethos.
Linguistics professor Emily M. Bender noted that while AI tools like advanced spell checkers or grammar correctors might be acceptable, distinguishing between legitimate editing assistance and content generation by LLMs presents a challenge. She pointed out that LLMs lack the accountability inherent in human contributors, who stand behind their contributions based on belief and responsibility, unlike AI systems.
Joseph Reagle, an associate professor studying Wikipedia’s governance, echoed these concerns, emphasizing Wikipedia’s historical wariness of AI-generated prose due to limitations like “hallucinated” claims and fake sources. He also highlighted the potential impact on Wikipedia’s reputation if AI-generated content becomes prevalent, as it could degrade the overall value and trustworthiness of the platform.
Reagle further elaborated that many LLMs are trained on Wikipedia content, creating a complex relationship. He noted recent trends where search engines and chatbots directly answer queries, reducing traffic to Wikipedia. This dynamic, coupled with the Wikimedia Foundation’s commercial agreements for content use, has led to some friction among editors who are wary of AI services leveraging community-generated data without adequately addressing the potential influx of low-quality AI-generated content.
Despite the prohibition on generative AI use for article content, Wikipedia still permits AI tools for translating articles from other language editions into English, contingent on human verification of the original text. The policy also cautions editors against relying solely on stylistic analysis to identify AI content, advising a focus on policy compliance and the editor’s contribution history instead.
Long-Term Technological Impact on Blockchain and Web3
Wikipedia’s measured approach to AI integration offers valuable insights for the broader blockchain and Web3 ecosystem. As these decentralized technologies mature, they face similar challenges in maintaining data integrity, user trust, and community governance in the face of rapidly advancing AI capabilities. This policy decision suggests a potential future where Web3 platforms might adopt strict guidelines for AI-generated content, prioritizing verifiable data and human oversight. The emphasis on accountability and verifiability aligns with the core principles of blockchain, which seeks to create transparent and trustworthy systems. For Layer 2 scaling solutions and AI integration within blockchain, this precedent could encourage the development of AI tools that augment human capabilities without replacing them, ensuring that innovation serves to enhance, rather than undermine, the integrity of decentralized networks and applications. The focus on human-centered models, even within technologically advanced platforms, reinforces the idea that the ultimate value of Web3 will stem from its ability to empower human users and foster authentic community engagement.
Original article : decrypt.co