Like it or not (and my own feelings are mixed), artificial intelligence is being used by news organizations, and there’s no turning back. The big question is how.
The worst possible use of AI is to write stories, especially without sufficient human intervention to make sure that what’s being spit out is accurate. Somewhat more defensible is using it to write headlines, summaries and social-media posts — again, with actual editors checking it over. The most promising, though, is using it to streamline certain internal operations that no one has the time to do.
That’s what’s happening at Sahan Journal, a 6-year-old digital nonprofit that covers immigrants and communities of color in Minnesota. It’s one of the projects that Ellen Clegg profile in our book, “What Works in Community News.” And according to Lev Gringauz of MinnPost (one of the original nonprofit news pioneers), the Journal has embarked on a project to streamline some of its news and business functions with AI. (I learned about Gringauz’s story in Nieman Lab, where it was republished.)
Bolstered with $220,000 in grant money from the American Journalism Project and OpenAI, the creator of ChatGPT, the Journal has employed AI to help with such tasks as processing financial data of the state’s charter schools, generating story summaries for Instagram, and adding audio to some articles.
The real value, though, has come in bolstering the revenue side, as the Journal has experimented with using AI to retool its media kit and to understand its audience better, such as “pulling up how much of Sahan Journal’s audience cares about public transportation.”
“We’re less enthusiastic, more skeptical, about using AI to generate editorial content,” Cynthia Tu, the Journal’s data journalist and AI specialist, told Gringauz. Even on internal tasks, though, AI has proved to be a less than reliable partner, hallucinating data despite Tu explicitly giving it commands not to scour the broader internet.
And as Gringauz observes, OpenAI is bleeding money. How much of a commitment makes sense given that Sahan Journal may be building systems on top of a platform that may cease to exist at some point?
Two other AI-related notes:
➤ Quality matters. In his newsletter Second Rough Draft, Richard J. Tofel has some useful thoughts on the panic over Google’s AI search engine, which has been described as representing an existential threat to news organizations since it will deprive them of click-throughs to their websites.
Tofel writes that clickbait will be harmed more than high-quality journalism, noting that The New York Times and The Wall Street Journal have been hurt less than HuffPost, Business Insider and The Washington Post. “If there is one overriding lesson of publishing in the digital age,” Tofel writes, “it remains that distinctive content remains the most unassailable, the least vulnerable.”
Though Tofel doesn’t say so, I think there’s a lesson for local news publishers as well: hyperlocal journalism should be far less affected by AI search than national outlets, especially for those organizations that emphasize building a relationship with their communities.
➤ Here’s the pitch. Caleb Okereke, a Ph.D. student at Northeastern, is using AI to screen pitches for his digital publication Minority Africa. He writes that “we are receiving 10x more pitches than we did in our early days after launch,” adding: “With a lean editorial team, we faced a challenge familiar to many digital publications: how do you maintain depth, fairness, and attention when the volume scales but the staff doesn’t?”
He and his colleagues have built a customized tool called Iraka (which means “voice” in the Rutooro language) and put it to the test. As he writes, it’s far from perfect, though it’s getting better.
“As of now, editors are using Iraka individually to provide a first-pass on submissions, testing its utility alongside regular human review,” Okereke reports. “Every pitch is still manually read, and no editorial decisions are made solely based on the model’s output. This staged integration allows us to observe how the tool fits into existing workflows without disrupting the editorial process.”
