From chatbots to search engines, the artificial intelligence (AI) systems that power modern life all seem to have studied the same site: Wikipedia. With over seven million English-language articles and a free-use policy, the site is a goldmine of high-quality training data.
But will the online encyclopedia embrace the technology itself? Speaking to BBC Science Focus on the Instant Genius Podcast, Wikipedia founder Jimmy Wales said that for editing and error spotting, the answer is an unequivocal yes – but he’s less convinced about its usefulness for writing entire articles.
“For me, what’s really exciting about AI is the potential that we might find some ways to use AI to support our community in Wikipedia,” Wales said.
Explaining what some of those ways might be, Wales said he’d been experimenting with a tool to analyse short Wikipedia entries alongside supporting sources to identify missing information or statements not backed up by evidence. “It turns out, it’s pretty good at that,” he said.
Wales added that it’s not just his own tinkering: the Wikimedia Foundation, the non-profit that operates Wikipedia, has created a dedicated Machine Learning Team to develop useful AI tools for the Wikipedia community.
“That kind of work, which is just maintenance for Wikipedia, people do a lot of,” he said. “[These tools] could be very exciting in terms of helping us elevate the quality.”
Asked if he could see Wikipedia articles being drafted in the near future, Wales said he didn’t think it was likely.
“I wouldn’t say absolutely never, but at least not in the short run. The latest models are still, from a Wikipedian standpoint, nowhere near good enough.”
One area the Wikipedia founder thought AI could be of utility is addressing the encyclopedia’s own biases. Studies have, for example, shown that only 20 per cent of Wikipedia biographies are about women. These entries are typically shorter and more likely to focus on family, relationships or appearance.
Responding to those numbers, Wales suggested that “it isn’t hard to imagine having an AI that runs constantly over Wikipedia, that you ask it to look for instances of thus and such type of bias, and call to our attention an entry or something that we can look at.”
But he was also concerned that bias on Wikipedia was being subsumed into large language models (LLMs), of which a majority are trained extensively using the site’s data: “I think that’s something that really needs to be focused on, particularly by the people who are training AI models. They need to give that a lot of thought.”
Even so, Wales insists that few corners of the internet match Wikipedia as a source of training data.
“Thank goodness we don’t have AI models that are trained on Twitter alone. That would be a really weird and angry model,” he says.
“Having training materials that are fact-based, thoughtful, and reflective is actually really important.”
He added: “Broadly, I think it’s a good thing if we have large language models that are more fact-based.”
Jimmy Wales’s new book, The Seven Rules of Trust, is available to purchase now.
Read more: