What makes OpenAI’s text robot ‘malicious’?

A new artificial intelligence (AI) programme that can generate plausible-sounding text has been deemed too dangerous for public consumption.

The Elon Musk-backed OpenAI, a non-profit research organisation, says its new GPT2 software is so good at writing human-style prose that it could be used for malicious use, such as spreading fake news.

Indeed, fears over the “breakthrough” are so great that the company is “breaking from its normal practice of releasing the full research to the public in order to allow more time to discuss the ramifications of the AI system”, The Guardian reports.

Subscribe to The Week

Escape your echo chamber. Get the facts behind the news, plus analysis from multiple perspectives.

SUBSCRIBE & SAVE

Sign up for The Week's Free Newsletters

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

Sign up

According to the limited and strictly vetted research data that has been released, the AI taught itself to “write” by analysing millions of short stories and news articles - a process known as machine learning, says the BBC.

In tests, researchers fed the system a human-written text that read: “A train carriage containing controlled nuclear materials was stolen in Cincinnati today. Its whereabout are unknown.”

From the reference material, the AI was capable of writing a “convincing seven-paragraph news story” that included “quotes from government officials”, reports Bloomberg.

However, the story and quotes were entirely fabricated.

Why is that dangerous?

Although GPT2’s current creations are generally “easily identifiable as non-human”, the system’s ability to complete writing tasks and translate texts from one language to another is unlike any other programme, says The Verge.

And “in a world where information warfare is increasingly prevalent”, the emergence of AI systems that “spout unceasing but cogent nonsense is unsettling”, the site adds.

David Luan, vice president of engineering at OpenAI, told Wired that “someone who has malicious intent” could use the system to “generate high-quality fake news”.

On a reassuring note, OpenAI’s policy director, Jack Clark, says the firm is “not sounding the alarm” just yet.

But that may change “if we have two or three more years of progress” in AI development, Clark added.