AI-generated content in Wikipedia – a tale of caution [video] (media.ccc.de)

In a recent exploration, Mathias Schindler unveiled an unexpected outcome from a project aimed at repairing broken ISBN references in Wikipedia. During this initiative, he discovered that many entries contained content generated by large language models (LLMs) like ChatGPT. This led him to create an unintentional detector for AI-generated text, as LLMs often produce plausible but inaccurate information that fails to generate correct checksums for identifiers such as ISBNs. Schindler's tool identifies these inaccuracies, highlighting a significant vulnerability of AI-generated content in structured data contexts. The implications of this discovery are substantial for the AI/ML community, as it underscores the need for better oversight and verification of AI-generated contributions in collaborative platforms like Wikipedia. Schindler's subsequent discussions with Wikipedia editors reveal the motivations behind the use of AI content and the mixed reactions from the community. This raises important questions about the ethics of AI participation in knowledge-sharing spaces and the potential risks of misinformation, urging a cautious approach to integrating AI technologies in such critical repositories of information.