Wikipedia, Media Bias and AI with Jimmy Wales
Wikipedia founder Jimmy Wales talks to Laila Rizvi, Autria Godfrey and Stephen Horn about trust, neutrality and what AI means for public information as Wikipedia marks its 25th anniversary. He argues that large language models are still weak on facts and depend on human-curated knowledge, and that there is a business case for neutral news.
Transcript
When Wikipedia launched, nobody trusted it. Teachers banned it, experts dismissed it. But now, 25 years later, it's one of the most used sources of information on the internet built entirely by volunteers with no ads and no algorithm. Jimmy Wales made that happen. He bet that ordinary people given the right tools could build something more trustworthy than any institution. Well, now, AI is facing the same scepticism, but some in Silicon Valley argue AI could actually be more trustworthy than humans. No ego, no hidden agenda, and no self-interest. Others say though, that's dangerously naive. So as AI gets more capable, will it make public information more trustworthy or less?
This week on Agents of Tech, we talk AI, information, and who we trust to get it right with Wikipedia founder, Jimmy Wales. Alright, Wikipedia has had open information space to itself for 25 years now. But is AI changing the way that people access information and thereby changing the way we trust?
So I think, Autria, that is what people used to call the $64,000 question. And we are sitting here in Washington, DC today, and I would say a politically, Autria, that trust is very scarce on the ground. And I think we trust-
I think you're right on that.
Right. It's a trust when it comes to government trust, when it comes to media trust, trust is scarce. I do think, however, it's a false dichotomy. We all look back to the good old days, right? Where we trusted our neighbours, we trusted our friends, and we trusted other human beings. And I think there's a temptation to harp back to that time and to think everything that's human is good, everything that is machine is bad.
Right.
And I don't think that's the world that we live in now, and I don't think it's the world that we're going to live in, right? So I think what we have to find is a way of building trust within those mechanisms that are going to see us through into the future. So Laila, trust.
I don't know whether we have to just rehaul the whole system because Wikipedia is kind of struggling and they need donations. It's really hard to stand it up. And I mean, it's done really well to survive so far. And we've seen in our previous episodes, in our comments, that people don't trust things which are associated not only with tech, but which are money motivated.
And we live in a capitalist society. So if there's just a commercial interest attached, and unfortunately, it is to all AI is that, you know, you've got ads in ChatGPT coming soon. So that's where people's trust really starts to shake. So I think it is a bit more complicated where there are a lot of factors which impact trust kind of at play here.
That's true. Wikipedia has not been beholden to any corporate sponsors or any companies.
Yeah.
And because it's a community of editors, I think people have put a lot of trust and a lot of faith into what they find on the page. But how does AI change that?
Exactly. But it is also like, people, we really do struggle to trust people that are associated with certain things which are associated with money, like, if they have an agenda, financial agenda, if they have an another agenda, a political agenda.
Exactly.
Right. So I think what we've become quite good at as human beings is navigating trust. And Jimmy uses the example quite often of, in England, you know, you might have a slightly left leaning newspaper, The Guardian, a slightly right wing, The Telegraph. But you can navigate in between those. You can figure out kind of what the facts are, and you can figure out what the opinion is.
And I think from my mind, that's going to be the same moving forward. But the players might be different, the technology might be different, but I think as human beings, we're going to have to do the same job. We're going to have to navigate trust.
Well, we want to navigate to our comment section now, because we're asking the question, is AI a threat to Wikipedia or will we always be reliant on the human component when it comes to seeking trustworthy information? Tell us in the comments below, and then don't forget to like and subscribe while you're there. Let's get into it now. Here's our conversation with Jimmy Wales.
Well, Jimmy, thank you so much for taking the time to speak with us today. We really appreciate it. We mentioned that you have hit a major milestone, congratulations on the 25th anniversary of Wikipedia. When you were first starting out, you weren't all immediately, you know, regarded as the most trustworthy source of information on the internet. How, if you share any sympathy for the scepticism that people have for AI, what is your thoughts on that, and I guess maybe your advice to those AI companies?
Well, yeah, sure. I mean, certainly there is a, you know, a cycle that happens, you know, when the quality is low and something's very new, then people obviously are sceptical and quite reasonably so, and certainly when we look at large language models today, you know, in thinking about Wikipedia and can it help us and so on and so forth, it's pretty bad. And, you know, it doesn't seem that bad when you first start playing around with it. But the deeper you go, the more you realise it's very, very superficial in its knowledge. And, you know, if we needed help anywhere, it'd be on more obscure subjects, and that's actually where it completely breaks down. So I think they need to really focus on improving quality. They need to focus on the hallucination problem.
From my perspective, you know, what I'm trying to think of is LLMs aren't going away, you know, I mean, we are not going to go back to the same world that you started Wikipedia in 25 years ago, you know, we're moving forward and in all kinds of fronts, you know, technology, be it AI or whatever, you know, is part of our lives and is moving very quickly. I think one of the examples you give is when you're trying to maybe fix something, find something with Wikipedia. But we seem to be at this point at the moment, human's good, machine's bad, and I'm trying to figure out, you know, how we move forward.
You know, what I'm interested in is can, you know, large language models in terms of helping us with our work, can they make suggestions that are not a complete waste of time? In other words, could they run through Wikipedia looking for things easy, low hanging fruit, things that people could do? And if it wastes our time and suggest things that make no sense or aren't valid, that's not helpful. But if it's useful, that could be interesting. And so, you know, I think there's that.
And I think, you know, in our lives, I mean, what I always say is like, if you ask me 25 years ago, if you asked any of us 25 years ago, gee, what do you think the first AI to talk to the public will be like? I think we would've all said, oh, it will be very uncreative, it will be very fact bound, very, you know, well, it's kind of the opposite. They're pretty terrible at facts, but they're kind of creative. You know, you want to brainstorm about your holiday, it's actually pretty good, but if you want to make sure that something's actually true, well, no, they're a little thin on understanding what's true and not true.
On that point, Jimmy, I know, I guess the strengths of Wikipedia were the fact that you gave a framework for people to use to be a trustworthy source of information. But AI and specifically LLMs are actually quite good now, if you give them the correct framework, like if you take them raw out of the packaging, they're pretty abysmal, like, they can hallucinate. But if you compare them, like for like, as in Geoff Hinton made the argument that people also hallucinate, like, you had the Watergate scandal where things were completely misremembered because it was a long time ago and it seemed plausible, and that's kind of what AIs also do, LLMs do.
So if you are giving it the correct scaffolding, whether that's in the case of them coding or them doing a specific task, actually they perform equally as well as human beings. So is that something that you might consider?
Yeah, I mean, I think we have to keep an active mind about the whole thing. I mean, certainly you can improve the outcomes from a large language model. If you give it a good framework, if you feed it the information it needs, you know, that can be very helpful. Certainly if you just say to any large language model, you know, write a Wikipedia style article about some topic, maybe a fairly obscure topic, it's going to put something that looks pretty plausible. But that if you're experienced Wikipedia and if you dig in, you're like, actually this is not very good, and that's a problem.
But if you say, look, here's the task, you know, we've got this link has gone dead, and it was supporting this statement, can you go and find some sources and read them and see if you can find a source that looks like it would support this, and then tell me that link and quote to me what it says exactly and so on. Actually, it can do that, it can do it not perfectly, but pretty well. And I think that sort of thing is very interesting in terms of how do we use this technology?
And I think that's going to apply across all kinds of fields. I mean, certainly, you know, when we think about things that, you know, you hear these reports about, ooh, breakthrough in mathematics by large language model, and usually those claims evaporate within a couple of days because it was some computer person who did it, and then the real mathematicians come in and go, no, no, no, no. Like, actually, it's a lot more subtle than that.
But at the same time, that doesn't mean that mathematicians should just ignore the technology. I mean, maybe there are ways that it can be used and certainly maybe not large language model, but maybe trained models that understand theorems and can do sort of mathematician type of work and reasoning might be very useful, even if it can't quite be at that top level because it doesn't really sort of know about reality and so forth. If it's useful, then I think broadly we should be looking at how to use it to, you know, improve our understanding of the world.
Well, so on that note, do you worry at all about LLMs replacing Wikipedia? Or do you think that there's always going to be that need for the human component when it comes to seeking truthful information or in the future, is there-
Well, I can't predict what the world will look like in 25 years, obviously. So it's hard to say. It's very fast moving technology at the moment, but certainly for the foreseeable future, I mean, one of the things we have to understand is that the only reason the large language models are any good at all is 'cause they've read Wikipedia, you know, like, I don't think, imagine a large language model that was trained only on Twitter or X, you know, I think it would be very angry model and very confused about the world.
And so that human curation of knowledge, and of course I would say because they've read Wikipedia, but also by the way, they've read all the academic journals, they've read major books and all of this, all of that good quality human knowledge is really necessary to build a model that is echoing that knowledge and is sort of producing it in a new kind of way. Without that, they would be completely lost. And so I think that we are still a long way away from all being replaceable, and I don't even know what that world looks like. I mean, maybe we're going to find out if we live another 25 years.
But these, I mean, there are claims that these models are heading towards super intelligence in a couple of years, as in, so that suggests that really isn't that far away, as in, do you think that that's just, that doesn't stand up? And do you not think that, what is the major obstacle that's not allowing these models to do the entire job that Wikipedia kind of does?
So I'm a tech geek and I'm keenly interested in all this technology, but I'm not an expert, I'm not an AI researcher, so I listen to various people who caution that it's not really that close. I mean, there's a lot of hype. So Gary Marcus is one, Demis Hassabis, they're both friends and, you know, they say actually, we need some more fundamental breakthroughs, like, Gary Marcus in particular seems to be pretty sure that we're hitting a plateau, that the models have not been improving in the last year or two at a rate that they were, and that he thinks we've kind of capped out on what we can do with the current technology.
Now, they might be wrong, you know, things have been moving very, very quickly. So I suspect it's one of these things, it's a classic sort of thing in tech. It's an old saying, people tend to overestimate the short run and underestimate the long run. And we might be sort of in that hype saying, oh wow, like this is incredible. If you compare to five years ago and look at the progress, it's amazing. They're going to be so much smarter than us in five more years. Well, maybe not, but what about in 25 years or 50 years or a hundred years? I don't know. Like, I think it's going to be pretty big.
All right. Well, that brings us to our first break. So we're going to coming back with Jimmy Wales and we're asking the question, can knowledge or any information platform be truly neutral? Jimmy's take on that when we come back.
So Jimmy, in your new book, The Seven Rules of Trust, you argue that neutrality was never about pretending that values do not exist, but about process, transparency, and accountability. So my first question back for you is, when it comes to today's most contested political, religious, and cultural topics, do you really feel that that kind of neutrality is achievable?
Yeah, I do. I mean, obviously it will always be imperfect. It's always a very human sort of discourse and dialogue. I think everybody understands what it means to be biassed, and we can think of any sort of media organisation or publication that we think is really biassed, and then think of the opposite of that, think of reducing that bias, think of trying to take into account different viewpoints, trying to step back a little bit, instead of taking a side into debate, explain what the debate is.
The classic example I always give is, if you think about a controversial topic like abortion, and you can imagine a kind and thoughtful Catholic priest and a kind and thoughtful Planned Parenthood activist, they're never going to agree. But because I've said kind and thoughtful, they can probably present the issue and say, oh, well, the Catholic church position on abortion is thus and such, and the Pope said this, and critics responded that. And that's really useful. It's really useful to make that attempt to make that try to say, actually, let's just try to explain the debate. Let's try to understand it first and not, you know, sort of take one side or the other, even though personally we might actually say, yeah, I'm actually, I do have a side, I have a perspective on this, but I think we can do that.
And I think people are better at that than maybe social media algorithms have trained us to believe, you know, that sort of, we have to be yelling at each other in each other's throats. But actually a lot of people are really kind of eager to say, oh, let's slow down a little, let's think this through. Let's be fair to the other side. Let's think about, you know, what is the strongest argument on either side.
But to expand on that, one of the themes in your book is building that individual level trust and you stress that people tend to trust individuals over say, abstract entities a little bit more. But then how do you account for human bias there that I feel like you're essentially saying you can't truly leave it at the door no matter how hard you try.
Yeah, I mean, of course. I mean, I think, you know, there are many different aspects of this. You can get pretty good at it, I think, and say like, actually, I'm going to talk about a certain issue in such a fashion that you actually have no idea what my personal opinion is, but you're going to walk away understanding that issue.
There are many ideas that people have at various points in history that are just unquestioned, and you don't think about it as bias. You just never have thought of it in any other way. And that's of course going to be, you know, there's no magic answer. We can't jump outside of reality. We can't sort of take a God-like perspective, but we can be aware that oh, okay, actually on certain issues, there are multiple viewpoints, they're not all right, that's not what I'm saying. But they're legitimately sort of contenders for the truth that need to be understood and explained even if we take one side or the other.
But how do you build a system that incentivizes that, as opposed to being in your own echo chamber and also just trying to access information that satisfies you, but satisfies you in a way that doesn't cost you anything, is really convenient for you, because that's unfortunately the world that we live in, where people will go to ChatGPT because it's really easy and they don't really care about the quality of the information as long as they kind of get what they want.
They're searching for.
They're searching for, or even on social media like that, people are happy with AI algorithms serving them up what they want to see. So how do you rehaul that in the world we currently live in?
Well, that's a fantastic question, and you're obviously a very astute, oh, sorry, I was channelling ChatGPT who just, to praise, but it is a good question. I'm just teasing. I'm having fun, but, you know.
Wow, that really was a dagger.
You make a very good point though.
People do actually, like, they actually are finding that sycophancy annoying, like it's one of the complaints people have is like, I actually, I mean, I always talk about like, I'm a programmer, but not a very good one. I don't programme professionally. I just, as a hobby, I enjoy doing it. I know I'm not very good, but, you know, I get an error message, I go and talk to one of the AIs about it, and whenever it says, oh, that's a really good question, I'm like, it is not a really good question, like, don't coddle me, it's a beginner's question. Just answer it. You know, I don't need you to suck up to me.
So I think people do, they don't want to be misled. I mean, at times, of course people can want to be misled a little bit, but broadly, you know, the research shows, and this is one of the interesting things, is that, you know, publications that simply pander to our biases, we might like them on one level, but we don't trust them because we sort of realise like, actually you are campaigning for this politician or that sort of perspective in the world. I actually do want to hear the challenges to that viewpoint. Even if I agree with you, I'd like to, you know, have you report on the other side in a fair way so that I make sure that my ideas are actually true and that, you know, I'm digesting things. So I think there is demand for neutrality. I think people do want it.
You say that, you say that, Jimmy, I'm going to push you a little bit and you're going to say, this is not a good question for that reason, you know, but I've been in the media most of my working life and you know, and I can remember the times you are talking about that business model's virtually gone now, you know, and the world that you are talking about where we all get together and freely discuss and debate and see each other's point of view.
Have you looked at the political situation in, you know, the UK or the US that I would argue is not there anymore. And I feel that the modern world is moving away from the position that you are a champion for, and I'm sort of interested in your view on that, but also what do we do about that as opposed to thinking back 25 years and saying, I prefer those days.
Well, my view is we have to fight for it. I mean, we have to be really, really passionate about it. And I think it's really important.
I mean, I live here in the UK, the BBC's been going through a bit of a crisis lately about, you know, issues in this area. I think it's the right thing for them to do, to have a crisis and to sort of really reflect and go hard on like, okay, how do we get here? What did we mess up? How do, and you know, to be fair to the BBC, it's not, you know, it's not gone crazy woke left, like the kinds of bias that we've seen in the BBC are so minimal compared to most publications, but it's their job. It's the sort of central premise of the BBC in many, many ways.
And so I just think we have to support that kind of work and really champion it and say, you know what, actually, by the way, although I enjoy reading about how fantastic electric cars are in The Guardian, I don't always trust it. So I turn to The Telegraph to hear how terrible they are, and I'd really rather just get informed.
But that's a luxury, as in, we might have the luxury to do that, but most people don't. And that comes back to, again, the ease and convenience of things like AI or just one news outlet. So is that really a realistic solution?
Well, I don't think it's a luxury. I mean, I think the luxury is wallowing in ignorance. And so I think, you know, I think, I mean, I think there are some analogies, you know, and they're not perfect analogies, but, you know, analogies to junk food, you know, junk food's easy. It tastes really good right now. I mean, honestly, for myself, if you put a bag of crisps in front of me, truffle crisps in particular, I'm just going to eat them. So I don't buy them, I don't like to have them around because I know I have, I actually do have a higher order sort of brain. I'm not, you know, a simplistic being.
But I'm going to come back to you with that analogy with the fact that like, it's argued that junk food or cheaper food or fast food is usually consumed by people who are poorer, who don't, who have less time, can't afford, again, the luxury of like nicer, healthier food, like, it's not as in, it's usually those people that it's targeted towards or that suffer the most. So like, I don't know.
There is certainly a lot of truth in that. I mean, certainly people who aren't sort of media literate and who have strong opinions and like to hear those opinions, you know, and they enjoy the battleground mentality of social media. I think it's a problem. I think it's not a great way to live life. And I would counsel people to say no, step back and think a little bit more about that.
I do think, you know, in the longer term, one of the things that needs to be really prioritised in education is media competence. There's really saying, you know what, actually, you know, here's some examples from the media on either side. Let's work through like what's wrong with this approach and what would be a better approach? And how would you arm yourself to make sure you know the best? I mean, people need to be taught this.
Well, arguably there is no business model in media for neutrality. And also arguably the way that people get information has changed. We now live in a push world and not a pull world. This stuff comes out to you through social media. So within that world, how do we do the things that you are talking about where we aren't going backwards because of that there is no, you know, economic model for it? The BBC is taxpayer funded, you know, Wikipedia is a charity. There is no business model for neutrality.
Well, I think there is, and I think, you know, the popularity of Wikipedia is testament to that. It's more popular than the top 10 newspapers combined. So I think there is demand for it. And I think that when we look at the media today and we look at the Edelman Trust Barometer numbers and see the incredible decline of trust in journalism, I mean, it's rivalling how low the trust is in politicians. And that's quite bad. I do think that there is room to say, you know what, actually people are getting fed up with this.
Now, it's easy for me to say that, right? It's much harder to actually get in the trenches and do it. So I have respect for, you know, even when I, my comparison that I love to make between The Guardian and The Telegraph, which would mean more to UK listeners than US, but it's, you know, they're both good quality newspapers. One is centre left, one is centre right. And you know, that's okay.
I actually, you know, I wouldn't say they both need to turn themselves into neutral point of view, Wikipedia, BBC style, but they should really kind of think about reining it in a bit and really, you know, be very careful to say, look, actually let's remember that even if our audience loves to click on that headline, they lose trust in us if we don't also acknowledge, you know, the other side and so on and so forth. So I think we can improve without sort of asking for everything to become completely neutral.
Alright, Jimmy, before we let you go, we have a couple of rapid fire questions. This is what we ask all of our guests to get varying perspectives on these same questions. My first one is for you, where do you think people will draw the line on allowing AI into their personal lives?
Well, you know, rapid fire, so I got to keep it short. There's this new OpenClaw thing where you can download this software, give it complete control of your computer and let it do all kinds of things. I think it's insane.
Blows your mind.
I think it's a cool innovation and I'm actually testing and playing with it, but not on my personal computer, you know, I'm not going to let it read all my emails and start writing to people. That's crazy.
I mean, you even talk about mass surveillance.
So I think we'll draw a distinction between, you know, personal, private information, and getting things done. Even, you know, if I have an AI on my computer that's giving me suggestions, but has no ability to talk to anyone else, okay, that sounds a little better, but I think this is something that we're going to be all exploring in the next few years.
Good.
So, okay, so my one is, what's something that's universally accepted in your kind of field that you disagree with?
Maybe that news media has to be biassed to be financially successful. I agree that I'm a minority viewpoint there. I think a lot of people are grappling with that issue.
Okay. My question is, what can you, you said before that sometimes we underestimate the future. What can you see in the future that we are not talking about today?
Well, I mean, I think we do, we talk, we're talking a lot about AI in the form of large language models, but there's a huge amount of other stuff going on. I mean, most prominently we see things like, you know, the AlphaFold, the sort of advances in biology that are really, really substantial. You know, drug discovery. We see things like driverless cars are coming very, very quickly. I think there'll be a lot of things like that that are going to be really transformative in a really positive way for our lives. And so while we're worried about, you know, things and large language models feel particularly worrisome because we can understand them and interact with them, I think there's a lot more that's going to come, that's going to be really pretty amazing.
Thank you.
Jimmy, to put Wikipedia to the test, I looked at your personal Wikipedia page and it says that you like to go by Jimbo. True or false?
Online, I go by Jimbo. In real life, people just call me Jimmy.
Alright.
The funny one is, if people call me James, I'm like, no, that's really weird because my actual name on my birth certificate is Jimmy 'cause I'm from Alabama.
Awesome. All right. So it is true, verified by the co-founder himself. Thank you so much for your time today. We appreciate it.
Lovely. This was great fun. Thank you.
Thank you.
I was a little bit sceptical when we first started speaking to him that there is a future where both AI and Wikipedia can coexist, but he really made me believe that the pendulum will swing back to people seeking out human generated content versus machine generated content. Am I on an island there?
I think I've got a nice bit of swamp land in Florida with your name.
You don't think so? You think that we're headed toward machine?
Look, I love talking to Jimmy. Jimmy was just such a nice guy and had so many interesting things to say and you could talk to him all day. I just feel we've moved on. I just feel we, our world of consuming news media has moved on to a place away from that. And I don't think we are likely to go back much as I would rather we did.
I think it's unfortunate because the things which are commercially very reliable, such as ChatGPT, are not to be trusted.
No.
But things which are not commercially viable should be trusted.
Right.
So it's a bit of a Catch-22. You're not going to win.
I think as long as the hallucinations still exist with LLMs, you're going to have people who are truly seeking out a factually correct information.
And like the hallucination rate is massively, I think overstated, like, it's not that bad. But I do agree with him on the, I don't like ChatGPT saying, oh, you're so wonderful, Laila. I like Claude who says that you're, you know, you're a bit delusional. Like, I want a bit of-
You don't want to be in an echo chamber.
But that's just like a preference in terms of AI tone.
Listen, I think you're right, Autria, I think for people who want to seek out the truth or impartial information or whatever, that's absolutely right. What I'm not sure of is that those people are these days in the majority.
I agree with you.
I agree.
Two things can exist at the same time. Well, a huge thank you to Jimmy Wales for joining us today. Trust in public information is collapsing. Do you think AI makes that better or worse? Drop your comments below, and then make sure to like and subscribe. Thanks for joining us today. We'll see you next time on Agents of Tech.