Moneycontrol PRO

News website used AI to write articles. Now, it has to make ‘substantial’ corrections

American website CNET experimented with an internally-designed AI tool. It did not go as planned.

January 27, 2023 / 03:57 PM IST
Lessons a news outlet learnt after using AI. (Representational image)

Lessons a news outlet learnt after using AI. (Representational image)

In November 2022, American tech and finance news website CNET, began testing an artificial intelligence tool to churn out articles.

It wasn't ChatGPT but an internally designed AI engine to help editors create basic explainers on the topic of financial services.

They published 77 articles with Artificial Intelligence  assistance . Out of those, they had to correct 41.

One of the stories wrongly stated that a person would "earn" $10,300 after a year, if they deposited $10,000 in a savings account that has a three percent interest.

Another one of CNET's personal finance articles misstated that home equity loans do not affect private mortgage insurance payments requirements.

In another story, plagiarism was a problem. CNET said later replaced phrases in its article "how to close a bank account" that weren't completely original.

CNET's editor-in-chief Connie Guglielmo said her team did a full audit of articles after errors were pointed out.

Elaborating on how they used AI, Guglielmo said editors created story outlines first and then expanded and refined AI drafts before publishing them.

They have paused AI usage for now.

"We've corrected these stories, with an editors' note explaining what was changed," Guglielmo wrote in an editorial. "We will restart using the AI tool when we feel confident the tool and our editorial processes will prevent both human and AI errors."

"We identified additional stories that required correction, with a small number requiring substantial correction and several stories with minor issues," she added.

With regards to plagiarism, the editor said CNET was working on additional methods by which similar published content could be flagged. They will also take more steps to curb misinformation.

To inform readers, CNET has a disclosure on its pages that says articles were written with AI's help but fact-checked and reviewed by its editorial staff.

AI-assisted articles have the two bylines -- one for their CNET Money vertical and another for the editor who reviewed the article.

first published: Jan 27, 2023 12:33 pm