Newsguard, a specialist in disinformation and its propagation, has identified 49 news sites whose content is written by text-generating AI systems.
They are 49 and publish in seven different languages. Newsguard’s latest study shows that artificial intelligence has taken hold in the news and information sector. The specialist in false information and their dissemination has counted around fifty sites “entirely or mainly generated by artificial intelligence software”.
These sites produce numerous articles on multiple subjects whether in English, French, Chinese, Portuguese, Tagalog, Thai or Czech. Some exceed 100 posts each day, Newsguard notes. With many advertising placements, they are similar to “content farms”, that is to say, they maximize their advertising revenue by bringing as many people as possible to their content.
Empty or poorly completed legal sections
But this very often involves misleading inciting titles or repeats of already existing articles. Thus, the American media CNN sees its articles rewritten or summarized by AI systems before being published by these fake news sites.
To identify them, Newsguard relied on the language used in the articles. Black point of artificial intelligences, the vocabulary is quickly running out. Another index, pages “About”, supposed to give information on the owners and legal persons in charge for the publications, empty or filled in an approximate way.
“This site was founded on (date) by (your name)”, for example, indicates HistoryFact.inone of the sites identified by Newsguard.
Sometimes it is the content itself that betrays its origin. The disinformation specialist has thus found that articles integrate the classic phrases of generators by artificial intelligence, just like in certain reviews on Amazon.
“I am not able to produce 1500 words… However, I can provide you with a summary of the article”, specifies certain publications of the site BestBudgetUSA.com.
The fake death of Joe Biden
Only problem, these sites can convey false information, taking again the faults of artificial intelligence. This is the case of CelebritiesDeaths.com, which indicated on April 2 that US President Joe Biden had died. L’article consists of a catchphrase and a short body of text where the generation tool used itself denounces the false information.
“I’m sorry, I can’t complete this request as it violates OpenAI’s usage policy on generating misleading content,” the AI model warns.
The tool, presumably ChatGPT or its higher version GPT-4, also adds: “It is unethical to fabricate information about someone’s death, especially someone as important as a president.”
