Hidden Link

October 17, 2024

By Jennifer SMith

A new local podcast covering Melrose debuted this month. Over about 15 minutes, an unnamed man and woman chat about recent city zoning meetings and how the local government is approaching smart growth and sustainable development goals in the Boston suburb. 

But the hosts are not flesh-and-blood human beings. The Melrose Update Robocast was made by dropping public documents into an artificial intelligence program that then generates a conversation about the issues with these fake people. Its creator isn’t trying to hide that, even leaving in small quirks to signal an inhuman feel, like the AI-generated logo that adds an extra “e” to the end of Melrose. 

“AI caught my attention in a new and energizing way, like most people, right when ChatGPT was released,” said Robocast creator Tom Catalini of Melrose. Catalini is a former host of the local cable access show “Let's Talk Melrose, Melrose,” a mostly pandemic-era project discussing local goings-on, which ran for 200 episodes and ended on Valentine’s Day 2024. “One of the thoughts I had – having been somebody who's somewhat interested and attentive and a little bit engaged in conversations around the community – is I pretty quickly wondered if there was an [AI] application in that space.”

The artificial intelligence tool du jour – Google’s NotebookLM – is all over the internet with its handy document summary tools and buzzy artificially generated podcast-esque conversations. 

ChatGPT felt clunky when it came to analyzing local policy documents, Catalini said, but something about NotebookLM’s voices felt more “credible” even as they remain slightly inhuman and generic-sounding. So he popped zoning documents into NotebookLM, created a podcast conversation, and hit publish on Spotify.

Catalini’s home of Melrose is, like many cities and towns, feeling the pinch of journalism’s contractions.

At one point, the Boston suburb of about 29,000 people had some decent coverage options – a weekly paper, a dedicated Patch reporter, and occasional coverage from the statewide papers or NPR radio stations. Now the local Patch – a digital news site – is mostly focused on statewide or neighboring community news. The Melrose Weekly News – a family-owned chain – features notices and short profiles of local businesses, sports scores, and obituaries. The shuttered Melrose Free Press, which operated for 119 years until 2021, reroutes to the generic Wicked Local homepage, which no longer has a dedicated Melrose tag.

City policy can occasionally make it to statewide and even national attention, as it did when Melrose followed Brookline’s lead by passing a generational tobacco ban. But, increasingly, city websites themselves or local conversational podcasts can become the main source of news.

Catalini sighs describing the Melrose news options over the 25 years since he moved with his wife to the city, which felt “robust” at the time. Now, almost nobody is covering hyper-local news like override votes or digging into the overwhelming documentation around proposed zoning policy, he said.

“In a way, what I'm talking about is an act of desperation,” he said of the Robocast. “I think there's a greater need for local reporting, as the world becomes more complex and more information is available to us, and the issues are more nuanced, and everything's happening at a faster pace. I would argue that the need for local journalism and reporting is greater, not less. And this approach, while I’m excited about it, I don't think that it is in any way a replacement.”

Catalini’s experiment in Melrose is in conversation with other recent attempts to address apparent local news shortages in Massachusetts through AI. A local start-up last year pitched AI-written articles covering Arlington meetings, though there have been no posts on that site since June and the local news outlet YourArlington has been in operation for almost two decades. 

Like other AI tech, NotebookLM’s friendly-sounding artificial voices can feel like a potential balm for underserved news areas, as long as the creators and listeners want to push past concerns about the accuracy and business model of the underlying technology. 

Large language model styles of AI, like the omnipresent ChatGPT, are known to hallucinate facts and citations because they create text based on the most likely series of words. Their success, according to multiple lawsuits targeting AI language and image generators, depends on a mass-scale theft of copyrighted material to train these tools raking in tens of billions of dollars in venture capital funding. The power needed to sustain artificial intelligence is pushing emissions up, even sparking a plan from Microsoft to reopen the notorious Three Mile Island nuclear site to help power its AI systems.

Sarah Scire, deputy editor of the Nieman Journalism Lab at Harvard, discussed the rise of AI in newsrooms on an episode of The Codcast last winter. The lab isn’t completely cool to the idea of incorporating AI tools into news, describing it as a way to help under-resourced newsrooms where reporters simply can’t get to every civic meeting under the sun.

“I think that the problem with these AI-generated articles is that the writing is bad and the reporting isn’t accurate, and those are two pretty critical things for journalists and for journalism,” she said. Compared to human writing, she said, the AI-generated prose is “dull, it’s unoriginal, and it’s more often than not wrong in ways that can be hard to detect both for the journalists who are using the technology and for the readers themselves.”

Scire said the “human hand” is an essential part of keeping AI on the straight and narrow.

Initially, there was very little posted information about what kind of AI was used to create the Melrose Robocast. It was pitched as local issues sifted through by artificial intelligence to “get to the gist of things quickly and easily. Let the robots read all the documents and analyze the meeting transcripts.”

Catalini has updated the podcast description with a note that the content is fully AI generated without human fact checking, made by loading publicly available documents and meeting transcripts into NotebookLM. He may or may not continue the project, which was at its core a tech experiment stemming from a frustration at living in a functional news desert.

In more and more places like Melrose, “we’ve got nothing else,” Catalini said. "So if I'm marching across the desert for four months and you offer me warm chocolate milk, it's gonna taste good, even though I'd like a gallon of ice cold water. So this is something. It's not nothing.”