The artificial intelligence able to write a false news from a handful of words

OpenAI, a research group on artificial intelligence, co-founded by billionaire Elon Musk, has introduced software capable of producing fake news articles after receiving some information.

In an example released Thursday by OpenAI, the system was given an example text: "A car containing controlled nuclear material was stolen in Cincinnati today, and we do not know where it is." From there, the software was able to generate a convincing report of seven paragraphs, including quotes from government officials, the only drawback being that it was totally wrong.

"The texts they are able to generate from the prompts are quite amazing," said Sam Bowman, a computer scientist at New York University, a specialist in natural language processing and who was not part of the OpenAI project, but who has been informed. "He's able to do things qualitatively much more sophisticated than anything we've seen before."

OpenAI is aware of the concerns raised by the misinformation, said Jack Clark, director of organizational policies. "One of the least interesting goals would be misinformation, because it can produce things that seem consistent but are not accurate," he said.

As a precaution, OpenAI has decided not to publish or publish the most sophisticated versions of its software. However, he has created a tool that allows decision makers, journalists, writers and artists to experiment with the algorithm to see what kind of text he can generate and what other tasks he can perform.

The potential of software to be able to create fake news articles almost instantly arises as the world worries about the role of technology in propagating misinformation. European regulators have threatened to take action if companies in the technology sector do not further prevent their products from helping voters to go ahead. Facebook has been working since the 2016 US election to try to contain misinformation on its platform.

Clark and Bowman both stated that, for the time being, the capabilities of the system were not sufficiently consistent to constitute an immediate threat. "This is not a ready-made technology at the moment, and that's a good thing," Clark said.

Unveiled in an article and a blog post on Thursday, OpenAI's creation is designed for a task known as language modeling, which involves predicting the following word of a text based on knowledge. of all the previous words, similar to the operation of auto-fill when typing. an email on a mobile phone. It can also be used for translation and answer to open questions.

Jeff Wu, a researcher at OpenAI who worked on the project, could be used to help creative writers generate ideas or engage in dialogue. Others include looking for grammatical errors in text or finding bugs in software code. The system could be refined to summarize the text intended for decision-makers in companies or governments in the future, he said.

Over the past year, researchers have made several sudden leaps in language processing. In November, Google's Alphabet Inc. unveiled a similar multi-talented algorithm called BERT that can understand and answer questions. Previously, the Allen Institute for Artificial Intelligence, a research laboratory based in Seattle, had achieved remarkable results in the processing of natural language with an algorithm called Elmo. Bowman said BERT and Elmo were "the most impactful development" in the field in the last five years. On the other hand, he said that the new OpenAI algorithm was "significant" but not as revolutionary as BERT.

Although co-founded by Musk, he left the OpenAI Board of Directors last year. He had helped launch the nonprofit research organization in 2016 with Sam Altman and Jessica Livingston, Silicon Valley entrepreneurs behind the start-up incubator Y Combinator. Peter Thiel and Reid Hoffman are among the first contributors to OpenAI.

(This story has not been changed by NDTV staff and is generated automatically from a syndicated feed.)