Beware the Ires of Musk

A little over two years ago, I reported on a trip to Singapore I and my colleague Heather Ford took to visit Wikimania, a conference attended by volunteer Wikimedians from around the world. We were there to investigate what they thought about the potential impact of generative AI on Wikipedia as a trusted source of knowledge.

Many were worried, naturally, that the then already-infamous weaknesses of generative AI would lead to Wikipedia being infected by error and bias, or by the puffery to which many chatbots are prone. But the overriding concern was a deeper one, centred on the potential for generative AI to undermine the long-term viability of Wikipedia. If you can get all the information you need from a chatbot, why visit Wikipedia at all?

This worry was magnified once generative AI was incorporated into search engines, such as Google’s AI Overviews. Early data was inconclusive, but recent figures show that traffic to all Wikimedia sites is now taking a hit, with human pageviews declining 8% over the past year. Similarweb reported a drop of 23 percent in visitors specifically to Wikipedia over the 3 years to March 2025. Unique visitors have also been falling.

These declines in traffic mirror those hitting news websites, a matter of strong concern amongst Australian news companies, with Similarweb showing declines of over 30% at some major news sites. At the same time, traffic from AI scrapers has increased markedly. With the cost of bot traffic higher than for human users, this is straining Wikipedia’s digital infrastructure and increasing running costs. Some scrapers are programmed to act like human users to avoid detection. This leaves a bad taste in the mouths of many Wikimedians, who freely contribute their time to build and maintain the encyclopaedia, only for the contributions to be exploited for profit by AI companies.

Elon Musk’s recent announcement of Grokipedia, an AI-driven competitor to Wikipedia, will likely have made this bad taste a whole lot worse. Musk has had it in for Wikipedia for a while, calling it ‘Wokepedia’ for what he perceives as a leftist bias. To the bad taste you can add a large dose of irony. Musk is building Grokipedia in part by piggybacking on Wikipedia content: using Grok to analyse a page to distinguish what is true from what is ‘partially true, or false, or missing’, and then rewriting it. But Grok, like other consumer AI tools, draws on Wikipedia’s vast archives for much of its data.

Grokipedia v0.1 was expected to launch this week, but it has been postponed as, Musk says, more work needs to be done ‘to purge out the propaganda’. When propaganda is a view you disagree with, this may take longer than Musk expects. The project is also unlikely to garner much long-term interest, as various other ideologically driven online ‘encyclopedias’ attest. Musk says Grokipedia is a ‘necessary step towards the xAI goal of understanding the Universe.’ But a true encyclopedia has a purpose that is at once more modest and more ambitious. It does not purport to know the truth in advance, as an ideologist does, but seeks to record what is currently known by relying on the expertise of others. Wikipedia’s open, communal editing process relies on a common commitment to a set of principles that lead over time to greater accuracy and reduced bias. AI can offer – at best – only a simulacrum of that process.