New AI Algorithm Generates And Detects Fake News Articles

In an effort to stop bot-written fake articles from spreading online, a scientist team developed an artificial intelligence algorithm generating the most convincing fake news written by a bot based on only a headline.

GROVER can generate misleading and fake news which are even more believable than humans-written articles, as per research posted on ArXiv preprint server on May 29 and the AI can detect them as well. The researchers wrote in the paper:

That means the new algorithm can apparently discover fake news written by AI better than other current tools. However, the system can also create a lot of fake and misleading information.

In an attempt to prevent the spreading of fake news, scientists from the University of Washington who created GROVER state in their study that they intend to introduce the tool to the general public, which is a big difference from OpenAI team who refused to release its GPT-2’s full version, a similar algorithm.

The algorithm is able to evaluate a news article in more fields than other tools, from the headline of the article to the body, author name, publication name and other details.

However, the attention to detail can also allow other people to use the system to generate their own fake news.

The research shows how easily the system can churn out an article written in the styles of news outlets like The New York Times, TechCrunch, Wired, and The Washington Post wrongly declaring there is a link between vaccines and an autism spectrum disorder. People reading the articles created by GROVER found that they are more believable than the humans-written ones, as per the research.

GROVE, which writes in the style of the science section of the NYT, generated from the headline, the name of the author, as well as the opening of the news article that points out the link between autism and vaccines from the federal government and UC San Diego’s scientists:

In addition, the scientists also show the way GROVER can refine output over time to match a particular publication better. They provide the system a headline about vaccines that cause autism and order it to write in the style of Wired.

The new algorithm uses the headline to create a full news article and after that, it perfects the headline to turn the article into the one that seems to be actually published by Wired.

The researcher team admits that the release of GROVER could pose a risk, however, they believe that the system is the best tool to fight algorithmic propaganda.