Wikipedia draws a red line against ai slop
Wikipedia’s editors just slammed the door on machine-written prose. Their reason: artificial intelligence doesn’t quote facts—it paraphrases them until they bleed.
The ban, slipped into an internal policy update last week, amounts to a quiet cultural coup. Volunteer moderators will still use bots to fix typos, but any article whose narrative arc was sketched by a language model faces summary deletion. No grace period, no appeal.
The hallucination tax
Chatbots, the editors note, excel at sounding certain while being wrong. A human plagiarist copies; an algorithm hallucinates, spinning fresh sentences that smell original yet carry hidden mutations—dates slide, contexts melt, nuance evaporates. In an encyclopedia, that drift is poison.
The numbers back the fear. A 2023 Stanford audit of 5,000 AI-generated Wikipedia drafts found 73 % contained at least one “substantial factual warp”: a birth year off by a decade, a company headquarters shifted across borders, a scientific claim attributed to the wrong Nobel laureate. Each error wore perfect grammar, making it almost invisible to casual readers.
Detection tools, meanwhile, lag like drunk bloodhounds. They flag barely 60 % of synthetic text when the topic is niche, and even less when the model has been fine-tuned to imitate academic dryness. The result is an arms race Wikipedia refuses to run.
Youtube’s landfill looms
Editors invoke YouTube as a cautionary tale: search any historical event on the video giant and you’ll drown in low-frame, AI-narrated explainers stitched from stock footage. The platform’s recommendation engine rewards watch time, not veracity, so the slop multiplies. Wikipedia, determined to remain the last major website you can still cite in a courtroom, chooses exile from that economy.
Yet the embargo is porous. No editor patrols the 300-plus language editions with equal vigilance; smaller wikis already choke on machine-translated filler. And seasoned contributors admit they sometimes feed a stubborn paragraph into ChatGPT just to untangle syntax, then rewrite every line. The tool leaves fingerprints only if the writer gets lazy.
What the policy really criminalizes is authorship without accountability. Sign your real name, stake your reputation, and you may polish sentences with silicon assistance. Let the algorithm sign for you—ghostwrite the past—and your draft becomes radioactive.
The stakes rise daily. Google’s search snippets still scrape Wikipedia first; TikTok’s fact-checking partners transcribe it for Gen-Z clips. Each synthetic paragraph that slips through could metastasize across the web before a human notices the tumor.
So the volunteers dig in, rolling back edits at 3 a.m., debating commas on talk pages that read like Talmudic scrolls. They know the ban won’t end AI; it merely postpones the moment when the encyclopedia’s voice becomes a ventriloquized average of everything ever written. That, for now, feels like victory enough.