‘Knowledge is human’: Co-founder Jimmy Wales on why Wikipedia still matters in an AI world | Technology News

On the sidelines of the AI Impression Summit 2026, we sat down with the co-founder of one of many web’s nice public establishments to debate reality, belief, and why, within the age of AI, probably the most radical factor one can do is cite their sources.

Edited excerpts from the dialog with Jimmy Wales, co-founder of Wikipedia.

Wikipedia is constructed on the concept of free human information. Within the age of AI, the place AI fashions generate on the spot solutions, what does ‘free information’ really imply?

Jimmy Wales: Effectively, for us, the idea of free information has all the time meant two issues. Freed from value, which the AIs are; effectively, most of the fashions are free, however the higher ones are normally not. So freed from value, but additionally free within the sense of being freely licensed to be redistributed, reused, and shared by anybody for any objective. That hasn’t modified. Clearly, the knowledge ecosystem is altering -in some methods for good, in some methods for dangerous. 

AI fashions are skilled on Wikipedia at an enormous scale. Does that make Wikipedia, in essence, the spine of the AI financial system or the infrastructure that’s quietly propping it up?

Jimmy Wales: Positively. For a few years I’ve mentioned that at Wikipedia we really feel a extremely robust and heavy duty, as a result of Wikipedia has change into a part of the infrastructure of the world. Now, LLMs are clearly a part of the infrastructure of the AI world. And I believe it’s actually vital that human-curated information, “information is human” is definitely considered one of our themes for this 12 months, stays central. It’s vital in no small half as a result of AIs can have horrible issues with accuracy and details.

I proceed to be amused once I mirror on this: what when you had requested me 25 years in the past, on the founding of Wikipedia, what the primary AIs could be like? I believe all of us would have mentioned, ‘Oh, they’ll be utterly uncreative, very fact-bound, very matter-of-fact.’ It seems they’re actually dangerous at details, they usually’re really fairly inventive. You’ll be able to ask them to brainstorm concepts. Due to that, we expect that for a really very long time, human oversight and curation can be actually, actually vital.

You’ve all the time maintained that AI ought to assist human editors and never change them. The place precisely do you draw that line, and who will get to implement it?

Jimmy Wales: For us, every part is enforced by the group by means of established processes, together with a substantial amount of deliberation, dialogue, and debate. That work is ongoing, significantly as we take into consideration which sorts of instruments are helpful and which aren’t.

I actually loved one latest story. A German Wikipedian named Matthias got here to me and mentioned, “Jimmy, it is best to see what’s been taking place on German Wikipedia.” He had written a bot, not an AI bot, that scanned German Wikipedia for ISBNs, the worldwide identification numbers for books, and checked them in opposition to a database to confirm they have been right.

He defined that when he finds an incorrect ISBN, it’s normally an apparent human error – a cut-and-paste mistake or a single digit typed incorrectly. However this time, he discovered ISBNs that didn’t exist in any respect, and the corresponding e book titles weren’t within the database both. Trying extra carefully, he seen that a number of of those entries got here from the identical editor.

Matthias contacted the particular person and requested, “What is that this? You’re including ISBNs that don’t exist.” It turned out to be a good-faith mistake. The editor mentioned, “I’m so sorry. I’m new to Wikipedia and wished to assist, however I didn’t know what to do, so I believed I’d add references. I requested ChatGPT for them, and it gave me these. I had no thought it might make up ISBN numbers.”

And sure, it could make up ISBN numbers. On this case, the difficulty was innocent as a result of it was caught by a volunteer. It’s a cautionary story: the output can look convincing, nevertheless it isn’t dependable sufficient. The issue was fastened and didn’t trigger lasting hurt. Now, Matthias plans to run his device throughout many extra languages to search for comparable points.

That’s the type of work the group is doing, like determining the best way to use these instruments responsibly.

Proper in the beginning of the 12 months, the web appeared to blow up with Moltbook, a platform the place AI brokers work together with one another. As AI begins citing different AI-generated work, is there an actual hazard that AI-generated information might override human consensus relatively than mirror it?

Jimmy Wales: One of many issues I’d say about that’s that journalism is extra vital than ever, as a result of constructing belief within the data we learn nonetheless very a lot depends on people. You got here right here to see me, you recorded what we mentioned, you’re a human witness, and you may write it down and say, ‘That is what we talked about; that is what occurred.’ Prior to now, to make issues up wholesale was very costly; you simply couldn’t do it simply. However now you possibly can push a button and generate 100 faux interviews with me, all of that are type of believable. So we actually do want human journalists doing this work, and I do fear about it.

However not a lot within the direct sense of an issue for Wikipedia, as a result of Wikipedia is all the time about dependable sources. You want a supply, and also you want high quality journalism. I fear once I see information tales about information organisations beginning to use AI. I’m like, okay, nice, however let’s be actually cautious. There are specific kinds of makes use of the place I believe that’s most likely high quality, and others the place I’m like, ‘I don’t assume that’s a good suggestion.’ I imply, everyone’s seen, I don’t know if it’s actual or not, however there’s a screenshot floating round of a information article the place the final sentence says one thing like, “For those who’d like to listen to extra, would you want me to put in writing one other essay for one more newspaper?” And it’s apparent that some journalist someplace reduce and pasted from a ChatGPT response on a deadline. Clearly that’s a comic story, and little issues like that can occur. However, we actually do want to verify our details are details.

Newsrooms are already utilizing AI regardless of hallucination dangers. Do you assume one thing like Wikipedia could be referred to as the final line of defence in opposition to AI-driven misinformation?

Jimmy Wales: To some extent, sure. Journalists are the primary line of defence, and communities like Wikipedia are the final. We’ve had an issue for years with genuinely faux information: web sites that look believable however aren’t actual. That drawback is simply intensifying now that huge quantities of content material could be generated simply.

The Wikipedia group is obsessively centered on sources and high quality. Editors always debate whether or not a supply is dependable, they usually already know the key newspapers and establishments. That makes it arduous to idiot Wikipedians, even when it’s straightforward to idiot individuals on social media.

One instance I usually give is a faux information website referred to as the Denver Guardian. It sounds actual, however Wikipedians instantly questioned it, checked, and found it was utterly fabricated. That’s the final line of defence: slowing down and asking whether or not one thing is definitely actual. I hope the general public is creating the identical intuition. Simply because one thing spreads rapidly on-line doesn’t make it true. Validation nonetheless issues.

How do you persuade youthful customers that human-managed information issues when AI feels smarter, sooner, and extra assured?

Jimmy Wales: I believe the very first thing could be to level to reliability – to say, really, AI could make up utterly loopy issues. I all the time get pleasure from asking AI about my spouse, as a result of she’s not a well-known particular person, however she’s recognized a bit, so she’s the proper instance. The AI thinks it is aware of who she is, so it simply makes issues up, they usually’re normally believable. That’s all the time type of amusing but additionally a warning that AI will get issues unsuitable loads.

After which there’s the factor of judgement, which is perhaps the second factor. There’s a distinction between indiscriminately spewing out a bunch of details and the judgement of what’s vital and what you have to know. I discuss this even round machine translation – it’s nice that it’s gotten loads higher, however you have to take into consideration the context of your reader. Who’s studying this? What do they should know? In some languages, you may not want to clarify who probably the most well-known cricket participant is. For those who wrote about him in Hindi, you may not want to clarify who he’s. However in English, you’d higher clarify it, as a result of quite a lot of your readers may not observe cricket. Did you translate the textual content appropriately? Perhaps. However did you translate the information context? You actually do want that richness of human expertise to try this effectively.

How do you view one thing like Grokipedia? I wouldn’t describe it as a rival or a peer to Wikipedia, however these sorts of engines are rising and are sometimes talked about alongside it, whereas additionally carrying a subtext of shaping or subduing sure political contexts. What’s your perspective on this?

Jimmy Wales: I take advantage of massive language fashions loads. I’m very fascinated by them as a result of I really like expertise, and I’m all the time experimenting. And I do know that they’re nowhere close to adequate to put in writing an encyclopedia. I believe when individuals first begin utilizing them, they’re like, ‘Oh, that is unbelievable; that is wonderful.’ After which the extra you ask about an obscure matter, the more severe the hallucination drawback will get. So the standard is simply not adequate. I’m not significantly frightened about that. It’s an fascinating experiment, nevertheless it’s far too quickly.

As an extension of that, when now we have glossy, AI-powered information methods that promise velocity, persona, and synthesis, why ought to customers nonetheless belief a slower, human-driven encyclopedia?

Jimmy Wales: Effectively, Wikipedia is de facto, actually quick. I all the time say, when you hear on the radio that some well-known particular person has died and also you go as quick as you possibly can to Wikipedia to be the primary particular person to replace it – it’s already there. So we’re fairly quick. However there are additionally instances the place we’re sluggish, and I believe these are fascinating as effectively.

One instance I used to be actually happy with: when Michael Jackson died, it took about 10 to 12 hours earlier than the group would enable it on Wikipedia, as a result of the primary report got here from TMZ – a star tabloid, which is okay, however it’s a tabloid and it’s made errors previously, and it wasn’t being confirmed by any main information supply. The group handled it as, ‘Effectively, it’s on TMZ, and there are quite a lot of rumours on-line, nevertheless it hasn’t been confirmed by any critical information organisation, so we shouldn’t have it.’ I believe that’s an excellent factor. In some instances, we needs to be just a little bit slower, as a result of we don’t know but. I don’t assume AI is essentially going to be sooner. It’s an excellent query.

India is considered one of Wikipedia’s largest audiences and editor bases. How crucial is India to the way forward for free information on-line?

Jimmy Wales: I believe it’s enormous. As a aspect word, that is my third journey to India this 12 months in a month and a half. I really like coming to India, and I really like the power. Once I was in Kerala, I met with the native Wikipedia group, and it was simply very enjoyable. Their pleasure about working with faculties and with libraries and the totally different tasks they’re engaged on, and that type of power is unbelievable.

Clearly the Indian IT sector is essential to the financial system, and the curiosity in AI is nice. I believe it’s nice that the AI Summit is right here; it actually highlights that. And for us, for Wikipedia, the multilingual nature of every part we do is so vital. One of many nice aspect advantages of LLM expertise is that machine translation has gotten a lot better than it was. It was utterly ineffective, then type of okay however amusing. It acquired first rate first between sure language pairs that have been economically vital and English-Spanish was one of many first to be first rate. Now it’s change into higher throughout every kind of languages, and that’s actually thrilling for our group. Folks need to have the ability to draw information from totally different locations and combine it. For a rustic like India with an enormous variety of languages, now we have many communities working in each English and their mom tongue and different languages.

Is there any concern that AI fashions, that are largely skilled on English content material, might find yourself marginalising non-English languages?

Jimmy Wales: Sure, undoubtedly. It’s one of many issues that’s actually fascinating. I simply noticed information about an organization right here saying an AI skilled on eight Indic languages, and it appears like an important programme. I can’t actually touch upon it intimately, however I believe that’s massively vital.

The issue of bias in AI is similar as the issue of bias in people. I may need sure cultural biases I don’t even find out about, as a result of I’m an English speaker who grew up in America and lives within the UK. However the excellent news is I can’t trigger a lot hurt with my very own biases, as a result of I can’t come into your language and begin writing about issues I do know nothing about. If I have been an AI, really fairly good at writing in all languages, I’d herald every kind of bizarre assumptions and issues which are simply unsuitable. We all know this type of bias is an issue. I believe it’s actually vital for individuals working in AI to concentrate on: what’s the physique of labor we’re coaching on? Is it complete? I believe it’s massively vital.

For those who have been to begin Wikipedia at present, in an AI-first web period, what would you design in another way to guard belief?

Jimmy Wales: I all the time say I’m a pathological optimist, so I are likely to assume every part is okay, and I don’t imagine I’d change very a lot. But when we have been really ranging from scratch at present, we’d have to spend so much extra time discussing when not to make use of AI, and there could be an actual temptation we’d want to withstand.

The perfect parallel I can consider, and I don’t assume I’ve ever mentioned this earlier than, comes from Wikipedia’s early days. The 1911 version of Encyclopaedia Britannica was within the public area and legally out there for anybody to make use of. Somebody prompt, “Why don’t we simply copy all these articles to get began?” Folks really started doing that, however we rapidly stopped.

After we seemed carefully, the content material merely wasn’t excellent. You would possibly assume, for instance, that an article on Aristotle wouldn’t want updating – he died 2,000 years in the past. However in actuality, an important deal had modified: scholarship had superior, archaeological analysis had progressed, and our understanding had advanced. We realised that copying that materials wasn’t useful.

I believe the identical factor would occur with AI at present. You’d begin utilizing it after which realise it comprises too many errors to be genuinely helpful. We’d doubtless undergo that studying course of, however I don’t assume we’d essentially change how we function. The basics of information are all the time going to be the identical.