What makes OpenAI’s text robot ‘malicious’?
Elon Musk-backed firm warns that artificial intelligence programme could be used to spread fake news
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.
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.
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.
Create an account with the same email registered to your subscription to unlock access.
Sign up for Today's Best Articles in your inbox
A free daily email with the biggest news stories of the day – and the best features from TheWeek.com
-
Magazine interactive crossword - May 3, 2024
Puzzles and Quizzes Issue - May 3, 2024
By The Week US Published
-
Magazine solutions - May 3, 2024
Puzzles and Quizzes Issue - May 3, 2024
By The Week US Published
-
Magazine printables - May 3, 2024
Puzzles and Quizzes Issue - May 3, 2024
By The Week US Published
-
AI is causing concern among the LGBTQ community
In the Spotlight One critic believes that AI will 'always fail LGBTQ people'
By Justin Klawans, The Week US Published
-
When even art is artificial
Opinion The AI threat to human creativity
By William Falk Published
-
The push for media literacy in education amid the rise of AI
In the Spotlight A pair of congresspeople have introduced an act to mandate media literacy in schools
By Justin Klawans, The Week US Published
-
The complex environmental toll of artificial intelligence
The explainer AI is very much mostly not green technology
By Devika Rao, The Week US Published
-
Artificial history
Opinion Google's AI tailored the past to fit modern mores, but only succeeded in erasing real historical crimes
By Theunis Bates Published
-
AI is recreating the voices of mass shooting victims
The Explainer The parents of these victims are using the AI to try and lobby Congress for gun control
By Justin Klawans, The Week US Published
-
The murky world of AI training
Under the Radar Despite public interest in artificial intelligence models themselves, few consider how those models are trained
By Austin Chen, The Week UK Published
-
Is Google's new AI bot 'woke'?
Talking Points Gemini produced images of female popes and Black Vikings. Now the company has stepped back.
By Joel Mathis, The Week US Published