The Washington Post is using AI to create personalized podcasts : NPR

This is not your mother's podcast, or your father's, or anyone else's. Washington PostThe new “Your Personal Podcast” offering uses artificial intelligence to customize podcasts for its users, combining the algorithm you might find in your news feed with the convenience of portable audio.

The podcast is “automatically personalized based on your reading history” Mail articles, the newspaper says on its help page. Listeners also have some control: with the click of a button, they can change their podcast's theme mix or even swap its computer “hosts.”

The AI ​​podcast immediately made headlines and drew criticism from people. question its accuracyand the motives behind it.

Nicholas Cua, critic and staff writer for Vulture and New York magazine WHO writes podcast newsletter, says AI podcast is an example Maillarge-scale digital experiments, but one of them was not successful.

“This is one of many technology-driven, digital-first experiments they are doing” aimed at “attracting a larger audience and penetrating new demographics,” he says. These broader efforts range from generative artificial intelligence tool for readers To digital publishing platform. But in this case, Cua adds, “it seems to compromise the core idea of ​​the news product.”

On that background page, the newspaper emphasizes that the podcast is in early beta testing and “is not a traditional editorial podcast.”

Bailey Cattleman, head of product and design for the company Mailcalls it an “AI-powered audio briefing”—and listeners will soon be able to respond.

“In the next episode, they'll be able to really interact and ask more questions to further understand what they just heard,” Cattleman tells NPR.

No matter how technically difficult it may sound, there are a lot of questions on the accuracy of the new podcast – even its ability to pronounce names correctly Mail journalists he quotes. Semaphore reported these are the errors that the employees are referring to Mail, included “misattributing or inventing quotes and inserting comments such as interpretations of source quotes” as the newspaper's own position.

The newspaper's app asks listeners to “check the information” by checking the podcast with its source material.

In a statement, the Washington Post Guild, which represents editorial and other staff, tells NPR: “We are concerned about this new product and its proliferation,” arguing it undermines MailThe mission of Russia and the work of its journalists.

Referring to the newspaper's established practice of making corrections if an article contains an error, the guild added: “Why would we support any technology that meets other, lower standards?”

So why Mail Are you producing a podcast on artificial intelligence? And will other news and audio channels follow suit?

Here are some questions and answers:

Isn't AI podcasting already a thing?

” Mail has definitely come out on top among traditional publishers in the US,” Andrew Deck tells NPR. But he adds that the newspaper is not the first to experiment with AI-powered podcasts in the broader news industry.

Deck, who writes about journalism and artificial intelligence for Harvard University's Nieman Lab, points to examples such as BBC's My Club Dailyan AI-powered football podcast that allows users to listen to content related to their favorite club. In 2023, he adds, “the Swiss public broadcaster used voice clones of real radio hosts on the air.”

News outlets have long offered an automated feature that converts text articles into computer-generated voices.

Even outside the news industry, artificial intelligence tools for creating podcasts and other audio are more accessible than ever. Some promise to simplify the editing process, while others can synthesize documents or websites into something similar to a podcast conversation.

Why do publishers want to experiment with AI podcasts?

“It's cost-effective,” says Gabriel Soto, senior director of research at Edison Research, which tracks the podcast industry. “You're cutting out a lot of the resources and people needed to make a podcast (studios, writers, editors, and the hosts themselves).”

And if a brand can create a successful AI-powered virtual podcast in today's highly competitive podcast market, Soto adds, it could become a valuable intellectual property in the future.

Dec says what if MailThe experiment worked, the newspaper “may be able to significantly increase and expand its audio journalism offerings without investing in the workforce that expansion would typically require.”

In the interview, Cattleman emphasizes that the new product is not intended to replace traditional podcasts: “We think they have a unique and enduring role, and that's not going away at the Post.”

What's unique about Mail Do you have a podcast?

For Deck, the level of customization it promises is innovative. The ability to tailor a podcast to an individual, he says, is “probably beyond what any podcast team in journalism can create by hand right now.”

IN example Mail publishedlisteners can choose from voice options with names like “Charlie and Lucy” and “Bert and Ernie.”

Cattleman says her team started with the idea that there is no “one size fits all” for audiences when it comes to AI and journalism.

“Some people want a really direct briefing style; some people prefer something more conversational and louder,” she says.

Cua says the addition of an AI-powered podcast is an attempt to make stories accessible to a wider audience.

He says that with the podcast Mail seems to be trying to reach young people who “don't want to read anymore, they just want to listen to the news.”

The main goal, according to Cattleman, is to make podcasts more flexible and attract younger listeners who are constantly on the go.

Description of the underlying process MailIn the AI ​​podcast, Kuttleman says, “Everything is based on Washington Post journalism”.

The LLM, or large language model, converts the story into a short audio script, she said. The second LLM then checks the accuracy of the script. Once the final script is stitched together, Cattleman adds, the episode is narrated by a voice.

Will listeners embrace an AI news podcast?

Edison Research's Soto says one in five podcast consumers say they've listened to a podcast narrated by artificial intelligence.

But he adds that for podcast listeners“many prefer human interaction, adopting artificial intelligence tools to assist with content creation but not podcast execution or hosting.”

The new AI podcast is a bit like Deku's hyper-personalized choices for users offered by TikTok and other social networks.

“There is a level of familiarity
and perhaps comfort with algorithmic curation among younger audiences,” he says.

But while younger audiences tend to be tech-savvy, many also care about authenticity and connection.

“Community is at the core of why people listen to podcasts,” Soto says.

There's also the idea of ​​a host or creator identity that drives engagement with TikTok and other platforms.

“These authors built relationships with their audiences—and perhaps even trust—even if they didn’t communicate with the sources themselves,” says Deck. “This type of news content is a far cry from the disembodied banter of AI podcast hosts.”

What are the potential downsides of AI podcasts?

One major potential impact is the loss of jobs and, for companies, the loss of talent.

“Automation is kind of erasing the entire voice productivity industry,” Cua says. “There are people who do this for a living,” he adds, who could “put out better versions of these records.”

There are also concerns that if AI selects a story and controls its presentation, it could create an echo chamber, ignoring context or skepticism that a journalist would likely express.

“AI-powered news personalization tends to be firmly entrenched in the camp of giving audiences what they want to know,” says Deck.

Dec says he's willing to give. MailAI Podcast, some time to see how this ends. But Dec has a major problem: “I can say this straight out: generative AI models are hallucinating.”

And when AI models are wrong, he says, they often get it wrong.

Blurring the lines between human and artificial intelligence voices can also raise issues of trust, a critical factor for a news organization.

As Soto says, “What happens when your audience expects real content from you, but finds AI instead?”