Study: ChatGPT misattributes, misrepresents news publisher content

A study by Columbia University's Tow Center for Digital Journalism revealed significant issues with ChatGPT's ability to accurately attribute and represent news publisher content, calling into queation OpenAI's commitment to responsible AI development in journalism . What happened Based on an evaluation of ChatGPT 's search function by testing 200 citations from 20 different publishers, researchers found that ChatGPT misattributed or misrepresented news content 76.5 percent of the time, with incorrect or partially incorrect source information in 153 instances. The citations used were from publishers that have agreements with OpenAI and those that do not allow their content to be accessed by it. The researchers also noted that ChatGPT often presented false sources confidently and fails to acknowledge its inaccuracies in most instances. Why it happened The misattribution issue is seen to stem from ChatGPT's design, which prioritises generating responses over ensuring accuracy. The result is that the AI fabricates sources rather than admit it is unable to find an answer. Furthermore, the system sometimes cites copied or syndicated articles instead of original sources, particularly when it is unable to access certain publishers' content due to restrictions. The study highlights that even when publishers allow OpenAI to crawl their sites, there is no guarantee of accurate representation, raising concerns about brand visibility and control over their material. What it means The study challenges OpenAI's commitment to responsible AI development in journalism, suggesting that without significant improvements in citation accuracy, publishers may be hesitant to collaborate with AI technologies such as ChatGPT in the future. The findings pose serious implications for news publishers, as they risk having their content misrepresented, which could undermine their credibility and reputation. System 🤖 ChatGPT Developer: OpenAI Country: USA Sector: Media/entertainment/sports/arts Purpose: Generate text Technology: Generative AI; Machine learning Issue: Accuracy/reliability; Mis /disinformation; Representation; Transparency

Date it happened

2024-11-01

Organisation involved

OpenAI

Product, system or model

ChatGPT

This is a record of what was reported, not a finding that anyone broke the law. If it names your organisation and you believe it is wrong, the corrections process is free and open to everyone.