AI has done some things to journalism. In fact, that’s really an understatement. Ask someone who’s been in the field for more than five years, and they’ll probably keep you there in the conversation for the better part of an hour, explaining what happened when these technologies were turned loose on an industry that deals, primarily, with the flow of information.

To be fair, AI is coming for everything. In some years, it will be making sweaters, delivering milk, and all of those other things that humans do, but in some essential ways, it came for journalism first.

Now, those who are concerned are looking at how to re-invigorate a moribund local news system, and evaluating how AI has helped or hindered this pursuit.

Mountain View, CA: The Next Endeavor

At the Next Endeavor event Sept. 14-15, in Mountain View, California, which Imagination in Action and partners put on (disclaimer: I help to run these events) we had a panel talking about local news.

If there’s somebody just right to lead a conversation like this, it’s Nina Gregory, and I also like how she framed it starting out:

“There is so much news about AI. But today we’re not going to talk about that. We’re going to talk about how AI is uh entering the newsroom, the process of gathering, reporting, writing, and distributing the news.”

With that, she turned to the panelists, Ken Doctor, Gabriel Sama, and Queena Kim, for more.

“We really view (AI) as a force multiplier, and we’re starting to use it in everyday activities,” said Doctor, who is the CEO of a pub called Lookout Santa Cruz. “And there are really three different ways that we’re using it now.”

One of those three ways, he noted, is in product creation.

“Neighborhood newsletters, in the old days, in the old print days that I came out of, would have taken large staffs in suburban neighborhoods collecting a lot of information,” he explained. “It costs a lot of money for all of that labor to do that. What we are able to do with AI is locate crime reports, school reports, transportation reports on a localized basis, by zip and subzip, and then create nine newsletters which are almost completely automated.”

A human, he said, reads everything before it goes out.

“A process that would have taken probably 40 hours of work in a given week, at least, if not 80, takes us 15 minutes each morning,” he said. Other uses he cited include spanish translation and geocoding.

Sama, of Columbia’s Graduate School of Journalism, chronicled his career changes since 2013.

“Back then, innovation moved way slower,” he said. So it was easier to first understand what was happening, and then to try to explain it to readers. With AI, I mean, I was in Meta in 2022, and I didn’t last very long, because I was part of the first layoffs in the company’s history. So I stayed there only about 10 months. But during that time, I came to understand that the company was in it for the long term, and also, that whoever had the resources and the money was going to expand this technology as quickly as possible, and that’s exactly what’s happened. And when it comes to coverage, I feel that we’re barely understanding how the technology is advancing, and it’s already changing.”

He described a loss of numbers, and a growing confusion about the way forward.

“Honestly,” he added, “I feel that most reporters are a little behind, and also you have to consider that back in 2013, there were way more technology journalists in the Bay Area … we not only are experiencing a technology that’s moving really fast, but there are very small numbers of people devoted to covering it, or, a smaller number of people devoted to covering it than ever. So that’s very problematic, and the stakes are only getting higher.”

He noted the potential impact of the doom-reporting of late:

“Most headlines the past two weeks have used the term ‘doomsday’ in multiple ways,” Sama said. “So I assume that for most readers, that’s terrifying, and so this is something that needs to be explained properly, and I don’t feel that there’s the bandwidth to do it.”

When Gregory asked Sama what standards we should be applying in newsrooms, he didn’t hesitate to espouse the idea that the same principles of originality that were upheld in traditional journalism still apply.

“When you copy and paste out of an LLM in journalism, we have a word for that,” Gregory said, supporting his assertion. “What is that called? That’s plagiarism. Plagiarism.”

“I think there’s been a lot written,” she began, “about how there’s a pretty big divide between how young people are looking at artificial intelligence. And so they’re, interestingly, very skeptical of it. The students at Berkeley are skeptical about AI.”

“We had a sort of working group with staff, professors, and also students, and the staff was pretty jazzed on it, you know, because they wanted this idea of getting your productivity up. Professors were sort of, huh, this is sort of interesting. … But the students were highly skeptical, and they were very skeptical about using a lot of these tools, because they didn’t want to be training the LLMs on how to do their job, basically. They were concerned about the environmental aspects of it, and so I think that there’s a big divide which has been very well documented.”

She talked enthusiastically, though, about the use of a tool called Descript.

“There’s still a lot of human labor that needs to go into it,” she said, of the processes that use this technology to expedite the understanding of audio. “It’s not a perfect tool, but it gets us way farther than we’ve ever been. And it’s so (important to) a lot of the editing that we do now.”

On the minus side, she gave the group an anecdote about her friend in the Bay area who is a reporter.

“He trained an AI agent to go through all the county minutes of meetings, what’s happening in the city council,” she said. “This is all stuff that he had actually started doing when he was a baby reporter. He was working for a big-shot reporter who didn’t have time to do that. So he’d say, ‘Hey, will you go and look through all this stuff?’ It was, you know, I’m not going to say that this was glorious work, but it was training, right? He was able to look out for things. The senior reporter would be like, ‘Hey, what about this? This is interesting.’ You’d go look for it.”

AI, she suggested, makes all of that obsolete.

“There’s another element that’s lost in this transcription too,” Gregory added, “which is not just the factual and skill-building training, but as an audio professional, when you have to listen to the tape and you have to listen to people talk again and again and again, you learn a lot of nuance that informs your skill as an audio producer and editor.”

Another example that got brought up is using NotebookLM to go through SEC filings: sure, it’s faster, and a lot more efficient, but what do you not see, as a human, when you’re not reading along, but just getting what AI pulls?

I really liked Kim’s explanation here:

“The pack reporters always go to the most obvious parts of that document,” she noted. “It’s sort of the brilliant mind, or the beautiful mind, that finds a weird little stray comment, you know, and says: ‘that’s really interesting.’ … That sort of human touch is lost, and you wonder, you know, what that’s going to do going forward, what the ripple effect of that is.”

Doctor disagreed, contending that the human touch is not lost.

“The ability to do journalism now, and into the 2030s,” he said, “if you are not using AI to do accountability and investigative reporting, that little that you lose from going through books like you might have in a Dickensian era is (that it’s) sped through. You can speed through so much and understand so much. We’ve used it. We are using it routinely now.”

“You can take city budgets, for instance, for the last three years. We did this in Eugene, Oregon, we took the city budgets in Eugene, and we ran them through there and said, ‘What are all the changes over these three years?’ Now, that would take hours and hours and hours. So you essentially had it just compare, and tell you any anomalies between the three years, and it surfaced all of these analyses.”

Here’s more of his testimony around this:

“In Santa Cruz, there are huge issues around pesticide use in South County on farm workers,” he explained. “There are all the ICE issues of roundups. Our ability to get to ICE documents that affect our people in both of our cities, and a lot of really deep pesticide research, is greatly enhanced by using AI. Those are just two recent examples.”

And then, he concluded, there’s outreach.

“The business impact of (AI) allows us to do more to reach people,” he said, “and I think it’s going to help us.”

Toward the end, Gregory asked the panel:

“Are there models, maybe, for independent journalists or others … in your experience of going both on the tech side and in the newsroom side?”

Sama was last to comment, with this lengthy explanation that I’ll provide in detail but split up into a sequence:

“I feel more pessimistic than I’ve ever felt in the past, or at least in the past 15, 20 years,” he said. “There was always a model to follow and replicate, at least when it came to making money, to paying the bills, to hiring people and paying their salaries. I felt that there was always something to emulate. Right now it’s very confusing.”

Another part of his analysis had to do with Google and its dominance in the 1990s. “Google is not perfect, and it was very difficult to figure out their algorithm,” he said, “(but) people were able to hack it, and there were some rules behind it that we were able to emulate and replicate and follow, and there was a benefit. I’m talking SEO, and sort of the death of SEO, which is happening right now for newsrooms. SEO was a thing. It was a very confusing thing, but it was a thing.”

He elaborated on this some more:

“Google never told us how their algorithm worked, at least when it came to publishers,” Sama said. “This was our way of figuring out the hidden rules. There’s nothing like that right now with LLMs. And LLMs present one very serious problem. While Google was a gatekeeper that we didn’t choose, but it was a gatekeeper, LLMs not only are behaving like that, but they are also an interface that presents our information in a scrambled way. So one thing that we do is that we verify, we attribute, we fact-check, and we present that information to you guys in a way where we can say: ‘we stand by this.’ And right now it’s very difficult to do that when it comes to LLMs. And going back to your question, which means that there’s really no current modern new business model to pursue, because it’s very confusing.”

I’ll leave it there. This discussion, I thought, caught a lot of the dire aspect that you’d get if you, say, had a career in journalism. What will happen, ultimately? We’ll see.