Wikipedia’s global reach hinges on its multilingual content, yet a growing reliance on AI-generated translations threatens the platform’s accuracy. The Open Knowledge Association (OKA) has been employing large language models like Google Gemini and ChatGPT to accelerate translations, but routine editorial reviews are uncovering factual errors and incorrect citations—common pitfalls for generative AI systems.
These ‘hallucinations,’ as they’re known in the industry, extend beyond basic misinformation. LLMs struggle with documentation, a flaw that has already caused significant issues in legal contexts. Wikipedia’s editors, who operate under strict scrutiny themselves, are now faced with a difficult tradeoff: speed and scale versus precision.
OKA’s model offers translators a monthly stipend of around $400, a modest but meaningful sum for many contributors based in the Global South. The organization previously used Grok, Elon Musk’s AI model tied to eX-Twitter, but has since shifted its approach. However, the commercial nature of this translation work introduces new risks. Unlike unpaid Wikipedia editors, who must disclose their roles and face additional scrutiny, OKA translators are subject to bans after just five documented errors—with their previous contributions potentially wiped unless a intervenes.
Despite these challenges, Wikipedia remains dependent on AI-driven translations for languages with fewer speakers. English dominates the platform, comprising over half of its content, leaving many other languages underserved. The pressure to fill this gap is intense, but the current approach raises questions about whether speed can be maintained without compromising accuracy.
The situation underscores a broader tension in digital publishing: how to leverage AI for efficiency while mitigating its inherent risks. For power users and editors, the stakes are clear—balancing rapid content expansion with the integrity of information remains an ongoing struggle.
