So-called content farms are websites that post articles in bulk to generate clicks and ad revenue. These include websites, whose articles are created using artificial intelligence (AI) and published by news bots.
What are AI-generated news sites and bots?
They are programs or algorithms that select news from the Internet according to certain criteria, but they can also generate, for example, current news with AI.
In a recent study, New York-based media ratings firm NewsGuard identified 49 content farms in seven different languages (Chinese, Czech, English, French, Portuguese, Tagalog, and Thai), whose content appears to come “almost entirely” of Large Language Models (LLM), Extensive Language Models, which can produce texts that appear to have been written by people. The best known available LLM application is probably OpenAI’s ChatGPT.
According to NewsGuard, the typical characteristics of the pages that publish AI-generated texts are that they lack information about who the site operators are and authorship of the articles, and they have tons of advertising.
Why are they problematic?
Probably the main goal of most of these sites is to attract users, save staff and earn ad revenue, NewsGuard suspects. Deliberate misinformation is the exception, not the rule.
Felix M. Simon of the Oxford Internet Institute opines that “LLMs have no consciousness, they cannot think. They can only produce strings of words and phrases based on statistical probabilities.”
“And that leads to the massive problem that an LLM produces texts, which sound very plausible, but have nothing to do with reality,” says Ross King, head of the Department of Data Science and Artificial Intelligence at the Austrian Institute of Technology. .
In this context, Gordon Crovitz, co-CEO of NewsGuard, makes a harsh judgment about content farms with AI texts: “The use of AI models, known for inventing facts as if they were news to create websites, is a fraud, because it pretends to be journalism,” he told The standard.
Can AI be used to intentionally produce false news?
Available LLMs have certain rules programmed in, but in fact, it is possible to outwit chatbots to some degree by giving them cryptic instructions.
So-called “fast engineering,” i.e. “instruction manipulation,” is gaining in prominence, says AI researcher Ross King: “I’m pretty sure there will be an illegal market for instructions,” says King.
DW, following NewsGuard’s investigative guidelines, also found pages with the mentioned characteristics in English, Portuguese, French and also in Spanish. Ross King opines in this regard that “these are technologies that used to be available only to researchers and perhaps government actors, and now they are available to the general public.”
How to distinguish AI sites?
Felix M. Simon shares the idea that “there will be more of these pages”, but the communication researcher, however, considers that their importance is overestimated. Simon is optimistic: the flood of AI sites will increase people’s awareness, and many will choose their sources of information more carefully in future.
Experts agree that it is important to strengthen the media skills of users, because there are no reliable AI detectors of videos, photos or texts yet.
Users may ask themselves these questions when finding out if a text or website comes from an LLM: Does the text make a good/serious impression on you? Does the information contradict your own knowledge? Is it a plausible text? Is there information about who is behind the page, who is responsible for the content, authorship of texts and photos? Do the profiles look authentic to you?
AI can create images and profiles that are difficult to unmask. If you can’t find information about such people or images on the web, they probably don’t really exist.
(rmr/ers)