Around 5% of New Wikipedia Articles in August Were AI Generated [Video]

Research from Princeton University, authored by Creston Brooks, Samuel Eggert, and Denis Peskoff, suggests the growing need of AI-generated content in Wikipedia and its implications on content quality, accountability, and bias amplification. GPTZero (a proprietary AI detector) and Binoculars (an open-source alternative) were employed to measure the extent of AI generated content in Wikipedia.

It was concluded that there was a marked increase in AI-generated content in recent pages compared to those from before the release of GPT-3.5. 

The upside of these models is that they boost productivity, but unchecked reuse of AI-generated content for training purposes can degrade model performance, and even impact quality. 

Types of AI-Generated Content on Wikipedia

Researchers found that 4.36% of 2,909 English Wikipedia articles from August 2024 contained significant AI-generated content. GPTZero flagged 156 articles, Binoculars 96, with an overlap of 45 articles between the two tools. The flagged content was generally of lower quality, featuring fewer references and …