Having conversations on AI’s role in Wikipedia, says founder Jimmy Wales

Jimmy Wales discusses AI's potential benefits and challenges for Wikipedia, emphasizing the importance of accurate translations and attribution.

Wikipedia is exploring roles Artificial Intelligence (AI) systems will play in the volunteer-run online encyclopedia’s future, the site’s founder Jimmy Wales told The Hindu in an interview recently.

“We are having a lot of conversations these days about AI, so that’s a big part of the intellectual work we are doing these days, thinking and talking with people in the community for what that means for the future,” Mr. Wales said.

Wikipedia’s traffic, at least from human readers, has gone down over the past few months by around 8%. The dip has been due to AI models, such as those built into Google Search, which repackage precise queries from users with answers sourced from Wikipedia and other sources. Since the site is ad-free, the fall in human traffic does not impact it as it would a news website, Mr. Wales said, adding that the Wikimedia Foundation has taken steps to share articles with AI developers in a responsible way.

Mr. Wales said that even before large language models (LLM) were introduced, “Google started getting smarter and smarter, and then it could tell you how old Tom Cruise is” without users having to visit the Wikipedia page. Now with LLMs, “quick questions” are being answered without driving traffic to Wikipedia.

He indicated that if the source of information remains unknown to the user, it may affect donations to the site. “For us, there is a difference,” he said. “Obviously, if people don’t know the information came from Wikipedia, they don’t remember, ‘oh, I’d like to donate to support Wikipedia’, that would be a problem. So, attribution is really important. Exact page views are not really something we obsess about.”

Last year, Wikimedia Enterprise put out an AI-friendly dataset of its English and French articles on the platform Kaggle so that LLM developers do not just scrape the data from its site. Some volunteers have expressed discomfort that their contributions are being part of an LLM’s training data. On this, Mr. Wales said, “I’m sure there are some [complaints] because obviously it’s a large, noisy community.”