The situation, in brief
Wikipedia does not trust artificial intelligence enough to let it play a direct role in editing articles on its platform, cofounder Jimmy Wales told AFP on Monday. The problem of AI “hallucinations” — in which fabricated output is confidently presented — has been reduced with newer AI models but…
Wikipedia does not trust artificial intelligence enough to let it play a direct role in editing articles on its platform, cofounder Jimmy Wales told AFP on Monday.
The problem of AI “hallucinations” — in which fabricated output is confidently presented — has been reduced with newer AI models but remains “very, very bad”, Wales said on the sidelines of a climate action week event in London.
He added, however, that AI agents could prove useful in alerting Wikipedia’s community of millions of editors to certain niche news that would otherwise be missed.
“We would not let it edit directly because you can’t really trust it enough,” he said.
Artificial intelligence platforms, meanwhile, rely on Wikipedia’s content to answer users’ questions.
That has contributed to an overall growth in visitors to the site from AI bots, while human traffic has dropped eight percent.
Wales, who sits on the board of trustees at the Wikimedia Foundation, which operates Wikipedia, described the fall in human traffic as “meaningful” but “not a disaster,” for the online encyclopaedia, which ranks among the 10 most visited websites in the world.
The site, created in 2001, depends on donations from users so its business model does not directly rely on traffic.
Wales encouraged AI companies to “pay their fair share”, because “hammering us with millions of requests costs real money,” in the cost of running servers.
Wikipedia has already been “very successful” in signing agreements with several tech giants, the founder said.
“We’re starting to block the ones who aren’t behaving themselves, but we’ll see how that goes.” — Agence France Presse
