AI is fast changing the way information is created and shared, but speed and scale don’t necessarily mean credibility. Jimmy Wales , who co-founded Wikipedia — harnessing the wisdom of the crowd to upend traditional encyclopaedias like Britannica and become the internet’s default reference point — now finds it at the centre of a new debate on trust, including attacks from Elon Musk over alleged bias. Speaking to Rohit Saran and Saikat Dasgupta on the sidelines of the AI Impact Summit in Delhi, Wales reflects on the opportunities and risks AI presents, why neutrality is non-negotiable, and why trust is more important than everFor centuries, humanity has found itself caught between the promise of tomorrow and its perils when talking about the future. Socrates died a worried man because he thought writing will kill the quest for knowledge. It’s the same now with AI. How do you see this?■ John Philip Sousa (one of US’s most celebrated composers) believed people wouldn’t sing anymore when music began to be recorded. Chess, as a game, is more popular than ever before, even though the best chess player in the world is no longer human. It hasn’t stopped people from saying, ‘Oh, but we really enjoy playing chess!’ About AI, since the technology is so new, and so accessible, I don’t know. You ask a computer a question and it can answer, that’s incredible. But we also know it’s flawed. And then, there is this wild conjecture that it’s going to destroy all jobs, or no one will need to work anymore because we’re going to become vastly wealthy. Probably the answer, as ever, is somewhere in the middle. AI is obviously going to have a huge impact. What it will be is so hard to predict right now.Broadly speaking, internet has made information a ‘commodity’. Do you think AI will make intelligence a commodity?■ All I can say is that right now, when we look at large language models — and I use them a lot, I’m a programmer, but not a very good one because I enjoy making things — it’s incredibly helpful and very fun. But it also makes things up and hallucinates. What I’m most interested in right now is, are there ways we can use this technology to support the community? Are there things that I can do pretty well? A lot of Wikipedia discussions are really long. You can get AI to summarise it. But here’s a key point, I want to read the original. It’s very useful. Another example would be you load up the Wikipedia article and all the sources and ask if there’s anything in the sources that should be on Wikipedia, but isn’t, or is there anything that’s not supported by sources. I did this and I think it is potentially useful for the community. Then say I want to write about a Bollywood movie that’s not globally famous in Wikipedia and I want to just get some basic facts about it. But I can’t read Hindi. Maybe the AI could help me a bit.Wiki powered a huge number of Google search answers and is now emerging as a source layer for AI. Are LLMs both a threat and an opportunity to Wiki’s future?■ We’re really about that human element, the human curated knowledge, the judgement. That machine translation might be a grammatical translation, but if you think about the cultural context of the reader, what they need to understand, what they are likely to know and what you need to explain to them, that goes beyond just the text. My example is, who’s the most famous cricket player? Today, it could be Virat Kohli in India. But if you’re writing for a global audience, you need to add a little text to explain who he is and put it in context. Machine translation can’t do that. But a human can.You’re making the case really for human moderation of information. Do you think that is where eventually the biggest problem with AI-curated information will lie, a wall that it cannot breach?■ So, Gary Marcus is an AI researcher who has become known as kind of an AI sceptic, although I would say he’s not really an AI sceptic, but thinks that large language models have already hit a sort of wall — that we’re not seeing improvement on a lot of key issues like hallucinations. He thinks there needs to be some more fundamental breakthroughs. For some time, scaling seemed to make all the difference. But there are other experts of equal fame who disagree with them. Just watching that, I think, maybe we’re going to have a little bit of a break for a few years until there’s more breakthroughs where it’s like ‘okay, we’ve got this amazing tool but maybe we’re not that close to the next steps’.Like Google search in the past, AI companies have a frenemy relationship with news media — New York Times has sued OpenAI, for example. If AI systems increasingly cite original sources, should they be required to link back and share revenue or traffic?■ I think we’re going to have a big fight on copyrights. It’s going to be across legislatures and courts, rethinking how copyright law is structured. My concern is, we want to be careful about overreach. One of the classic principles of copyright law has always been that you can’t copyright facts. Some scientific publishers may be very excited to be able to say, you can’t use the facts unless you pay. And that’s a disaster. We don’t want to go there. That damages Wikipedia and our ability to say that on this date, this happened. Here’s the sources, and it’s five different newspapers. Also, newspapers don’t want to go there. One of the bigger, deeper issues is that local journalism has been wrecked. And that happened long before AI. For society, that’s a huge problem. I’m from Huntsville, Alabama, which is not a huge city, but it’s not tiny — 250,000 people. When I was a kid, I was a paper boy. I rode my bike and threw the papers (into houses). As a Wikipedian, the value of this is if I want to write about the history of Huntsville or the 1978 mayoral election, I’ve got a lot of good content to work with. But if I want to write about the most recent election? Very thin content. Because there’s just one afternoon paper now that’s published three times a week and from 100 miles away. And that means the first draft of history, which is journalism, isn’t being written. So, the second draft of history, which is Wikipedia, becomes much harder.How do we solve that?■ I wish I had the answer to that. In some cases, maybe AI can help, if there’s some way to make it possible for one or two journalists to do more in a useful way that could be good. The changes in the information ecosystem have a lot of positives, obviously, but also some negatives. So go ahead and test.Wikipedia has had this debate between deletionists and inclusionists. Which side are you on?■ It has helped strengthen Wikipedia that we have this active, intellectual dialogue. I always say that I’m an eventualist, which is, we’ll probably get it wrong a lot but we’ll get it right eventually. The health of the Wikipedia community is important for us. Is the community having active discussions, having fun, behaving well? Are we doing things in a thoughtful way? I’m very comfortable with the debates as long as they don’t just become angry screaming matches. Tell us about your India community and volunteer group. There’s a perception or misperception, you tell us, that India pages are not as rigorous. ■ I find the Indian Wikipedia community to be very similar to everywhere in the world. A lot of nerds, not necessarily professionals in this field. There’s this guy in the global community with the nickname Hurricane Hank, who is a weather expert but not a professional meteorologist. The community has more men than women, unfortunately. That’s always something that we talk about. We want to improve that around the world. This is my third trip to India in a month and a half. On one of the trips, I was in Kerala, and I met with the local Wikipedia group. Among them was a couple, both Wikipedia editors, who brought their kids. On the second part of your question, I’ve not heard that about India pages. I do think it would be likely for the smaller language versions of Wikipedia. Obviously, those pages are typically going to be shorter and less filled in and less rigorous because there are fewer people doing it.When people say, ‘Wikipedia is broken because it’s biased’, how do you react to that? And what’s your take on an AI-first competitor like Grokipedia?■ See, Wikipedia is a source of knowledge, and sources are transparent things. One of the things that Elon (Musk) said is that Wikipedia just reflects mainstream propaganda. And I’m like, that’s really weird. Wikipedia reflects what reliable sources say. We cannot take one strange side, and say, ‘uh, we’re going to fight against all of scientific wisdom’, right? But we should reflect a debate if it’s a legitimate debate. Do we have biases? Well, of course, we’re human beings. So, we have to be really careful about that. Being neutral is one of the core values in the community. There’s no dispute about that. But do we always get it right? Maybe not. An old saying that I love is if you ask a fish about water, the fish will say, ‘What water?’ They live in it. They have no idea about it. And so oftentimes our biases are just there because we don’t know.How important is tone neutrality for credibility? ■ Amartya Sen makes a comment in the introduction of his book, ‘10 Indians, 12 opinions’. That’s actually all humans — 10 humans, 12 opinions. Tone neutrality is very important for Wikipedia and for newspapers as well. I live in London and read two papers, Guardian and Telegraph. Guardian is a sort of centre left, Telegraph centre right. Both quality newspapers. I have an electric car, not a Tesla. I love electric cars and so I read a lot about them. If you cut out the headlines from those two papers, I can probably filter the information 90% accurately, because Guardian loves electric cars and Telegraph hates them. But because of this tone, I also trust both of them less because it feels like they are both campaigning. That’s a problem because it can reduce trust, not just with people who disagree, but even with people who agree with the tone.You launched WikiTribune to address neutrality in public discourse. Why did you not continue?■ Tribune was an experiment to see if there was a way for journalists and community members to collaborate. What journalists can do, like you have come here in the middle of Delhi to chat with me, or go and report on something, or attend a press conference, or talk to a politician, that’s almost impossible to do as a volunteer. So, we explored some good collaborations. And then we would look at traffic stats every day. We had one story that had a really clickbait headline that I didn’t like. To make this into a commercial success, we needed more clickbait headlines. I didn’t want to do that. That’s how I realised the problem is not with journalism but that model, the broader ecosystem. The newspapers always love a good, juicy headline. There’s nothing wrong with that. But if algorithms are feeding people content to keep them around as long as possible, then that encourages more of the same kind of behaviour and so on. So it kind of changed my focus and was like, ok, good experiment.