Wikipedia at 25: Jimmy Wales on AI Hallucination and why he trusts humans over algorithms

When Jimmy Wales first dreamed up the idea that would one day become Wikipedia, few believed it would work. “Most people thought it was a disaster waiting to happen,” he recalls with a laugh. “A joke.” Twenty-five years later, that so-called joke is now the backbone of the internet. For an entire generation, Wikipedia has become the default reference desk of human knowledge — a digital Library maintained by an army of anonymous volunteers. But the information landscape around it has dramatically changed. Generative AI has redefined how people search, summarize, and consume knowledge — often skipping the encyclopaedia altogether in favour of chatbots that sound fluent but can’t always tell fact from fiction.

So, has Wikipedia — once the great disruptor — slipped into the same ‘terrible idea’ trap it once overcame? Is it still relevant in an age where algorithms answer faster than humans can fact-check? In a wide-ranging conversation, Jimmy Wales reflects on Wikipedia’s 25-year journey, the rise of AI Hallucinations, the growing challenge of political disinformation in India, and why he still believes Wikipedia is ‘healthy food’ for a mind fed on a steady diet of digital junk. Excerpts from the interview:

I want to start with something fundamental. If a term — say “Meconium” — appears in a book, can I cite that in the corresponding Wikipedia entry?

You can try, but it’ll probably get removed. The community would argue that’s not encyclopaedic. Just because a word appears in a book doesn’t make the book a credible authority on the topic. It’s like insisting that Taylor Swift deserves a section on the Wikipedia page for Buenos Aires simply because she performed there. It’s a fact, yes — but is it significant to the city’s overall story? Probably not. Wikipedia isn’t a dumping ground of disconnected facts; it’s about curating knowledge. The key question is always: What does the reader actually need to know — and is there a reliable source to support it?

Wikipedia still runs on volunteer editors, but the internet feels more hostile now than when you started. In India especially, politics seems to have become a ‘troll army’. During elections, political wiki pages get vandalized constantly. How do you keep up?

People look back at the early internet with nostalgia, but we had trolls even on Usenet. You don’t need algorithms to bring out the worst in people — we’ve always been pretty capable of that ourselves. What makes Wikipedia resilient is speed. If someone vandalizes a page, a human editor can fix it within minutes.

Also read: How Artificial Intelligence is shaping the future of Indian filmmaking

The real danger now isn’t the vandalism itself — it’s when AI systems like ChatGPT, Gemini, or Perplexity scrape the page during that short window before it’s fixed. A human community can self-correct; an AI model cannot. Once the wrong fact is baked in, it stays wrong until the next training cycle.

You’ve mentioned that large models scrape Wikipedia revisions, mistakes included. Can that be fixed?

Yes, and that’s exactly what we’re working on. We’re collaborating with companies like Mistral and Perplexity to give them access to a trusted feed instead of them scraping whatever’s live. The idea is to provide a version that’s been stable for a few days or vetted by senior editors. That way, we filter out vandalism and reduce the noise that seeps into the AI ecosystem.

When I search Wikipedia, I get the same answer as everyone else. But chatbots tailor answers to each user — more personal, more flattering. Doesn’t that give them an advantage?

To me, that’s a flaw, not a feature. You’re being told what you want to hear, not what’s true. I use AI tools for coding sometimes, and when I ask a bad question, they’ll respond: “That’s a great question!” I know it isn’t. It’s like having a friend who compliments everything you say. At some point, you want a friend who’ll tell you when you’re wrong or when your shirt looks terrible. Humans crave truth, even when it’s uncomfortable.

But don’t most people prefer comfort — even if it means believing fake news?

In the moment, yes. But there’s a big difference between what feels good now and what we actually want for ourselves long-term. Information works like food. If you leave a bowl of chips in front of me, I’ll eat it. But when I talk about my diet, I’ll tell you I don’t want junk food at home. The same goes for content. We know viral videos feel like “snacks,” but we also know we should feed our minds something more substantial. Wikipedia is meant to be that healthy option — something nourishing, not just tasty.

India’s linguistic diversity is massive. Hindi and Tamil Wikipedias thrive, but others lag. How do you see this?

I love it. The growth of a language Wikipedia depends not just on the number of speakers, but also on broadband access — and honestly, on a handful of passionate volunteers willing to build something from scratch. We do see interesting dialect issues, though. For example, Tamil Wikipedia gets a lot of contributions from the Sri Lankan Tamil diaspora, so certain words may sound unfamiliar to Indian readers.

Also read: 14-hour workday in oppressive conditions: Expert shares dark truths of gig work

Portuguese faces the same challenge between Brazil and Portugal. When dialects overlap, we usually maintain one version and encourage contributors to write in language that’s accessible across regions.

With video content everywhere — TikTok, Reels, YouTube — are we approaching a post-text world?

It’s a concern, but I’m not pessimistic. Text still transfers knowledge faster and more precisely than anything else. I often watch tech videos while I’m on the treadmill, but halfway through, I’ll open the transcript just to skim. Reading takes me three minutes instead of ten. Text lets you control the pace — stop, scan, re-read. And despite what older generations believe, young people are consuming long, complex content too: podcasts, deep-dive series, investigative journalism. Text isn’t dying; it’s just evolving.

Wikipedia turns 25. You’ve launched WikiCommons and Wikibooks. What’s the next big leap?

We’re not chasing radical change for the sake of it. Wikipedia works — and that’s rare enough on the internet. The goal isn’t to pivot to video or reinvent ourselves. It’s to keep improving the thing we already do best: provide a stable, reliable reference point for knowledge in a shifting digital world.