Researchers create AI capable of writing convincing fake news

Updated on 03-Jun-2020
HIGHLIGHTS

Elon-Musk backed OpenAI has just published a paper outlining how they were able to create an AI capable of writing very convincing stories on any topic by scraping the internet, all of which are fake.

Highlights:

  • OpenAI's new AI is capable of writing very convincing fake news.
  • The company has decided to not release the trained model due to its implications.

The team at OpenAI has announced that they have successfully created an AI that is capable of writing complete fictitious stories, all on its own. The company also said that they’re keeping the details of this system to themselves for its potential for abuse; to create and propagate fake news. The goal of their research was to create a system capable of producing natural language, but what they ended up with was an AI capable of writing very impressive fake news.

According to OpenAI, the system is capable of producing the content wholly by itself and only needs a brief. Once the system receives the brief, it scrapes data from over 8 million pages on the internet and does not require any retraining or human intervention to generate the final result. The AI is capable of generating content on any topic without supervision.

For the purpose of the paper, the researchers programmed the system to scrape data only from Reddit. The system would only take information from those links which had a karma score of 3 or higher, meaning, at least three people had deemed the information contained in the link as worthy and accurate. "This can be thought of as a heuristic indicator for whether other users found the link interesting, educational or just funny," the research paper said.

According to Dave Lee of the BBC who had a chance to see the AI in action, the resulting story generated by the AI is mostly coherent but completely false on almost all accounts. The AI uses information that is already published online but arranges it in a manner so that it appears to be unique and trust-worthy.

While many in the AI community do not have a very favourable view of OpenAI’s work, some do note that what the company has done is prompted everyone else to consider the consequences of their work. As a result, Elon Musk-backed OpenAI has decided to withhold the trained model of the new AI and instead, just released a small code sample along with a technical paper for fellow AI researchers.

Related Reads: AI will diversify human thinking, not replace it: Study

Digit NewsDesk

Digit News Desk writes news stories across a range of topics. Getting you news updates on the latest in the world of tech.

Connect On :