Dustin Dwyer is a busy guy. For twenty years, this Harvard Nieman Fellow has been cranking out content in the Mitten State for Michigan Public, a huge public media organization with roots in radio that covers nine Nielsen markets and the entire southern half of Michigan’s lower peninsula.
When Dwyer’s not in the field as the West Michigan bureau chief, he’s working on an ingenious tool called Minutes, which tracks and transcribes meetings from over one hundred government institutions and school boards to ensure that important stories aren’t missed.
I spoke with him last week, while he was covering May local elections and a massive tornado that ripped through suburban Kalamazoo. Our edited conversation follows. -Chris
CB: What was the inspiration for Minutes?
DD: It started simply because I had a problem. I’m a bureau reporter that covers West Michigan. Because I’m the only person out here, I have a bunch of communities in this area that I need to keep tabs on.
We used to rely on community newspapers to find out what was happening in small community school boards or city council meetings. But as you know, newspapers have been shrinking – some have been going away completely. I was finding that no news organization was sending people to these meetings and reporting on them.
We don’t have the resources to send people to every single meeting. The thought came up that if we could find the meetings online, we can use this new speech-to-text technology to generate a transcript, which I can read through faster than I can watch the whole meeting. It started from there, and we’ve been building on it ever since.
CB: Have you found this tool has enabled you to find stories you wouldn’t have discovered otherwise?
DD: Multiple times. I’ve used it in different ways: one is to find out about something that happened in a meeting that was newsworthy; that’s happened a few times. For example, I had a story about the town of Utica [Michigan] passing an ordinance removing a law that made it illegal to curse in front of women and children. It was an outdated profanity ordinance that had been on their books for ages.
Another was I use it is to discover human interest stories. These are stories not where a major controversy happened, but finding those little nuggets in little towns, you know, that have otherwise been overlooked.
CB: News that doesn’t create a Twitter storm.
DD: Not a Twitter storm, but they certainly create some engagement for us when we use them as fun kicker stories. As much as I want Minutes to always be this great investigative tool, it’s also great for human interest stories from city councils. Because a lot of weird stuff happens at city councils! But the tool is also good for when you have a story, you can just go into the system and hit search and then you’ll find the community where that’s been an issue.
I used it on a story about police budgets a few years ago when there was a big push to defund the police. We went back and tried to look at all the budgets of cities to find out if they actually did decrease or increase their police budgets. Minutes allowed us to search through public comments at meetings to find out not only what the city did with the budget, but we were also able to get voices from communities that we wouldn’t have heard otherwise.
Tuesday, there was a tornado in Portage, Michigan. I’m getting that city’s meetings added to the system so I can review where residents brought up issues of infrastructure. There’s certainly going to be lots of news coming up in the upcoming weeks as they do clean up.
CB: You had this idea to use AI technology to help mine meetings for your reporting. How did you get from concept to an actual usable tool?
DD: That’s been a journey. The idea really got off the ground because we applied for the Google news industry innovation challenge grant. At that point we didn’t have a single software developer on staff. This money really allowed us to start walking down that road. We got feedback from reporters and tweaked it. Then we partnered with the Associated Press, who helped us get to the next step. We also relied on partnerships with the University of Michigan, who holds our broadcast license, where computer programming students helped us build an email alert system.
It’s been building one thing at a time, testing, and getting some feedback; then it’s figuring out a way to build the next thing and finding the right partnership to make it to get it done. We’ve really relied on those partnerships.
CB: What’s next on the roadmap?
DD: I’ve experimented with taking our transcripts and sending it to a GPT and asking it for a summary, which I think is going to be a really helpful thing.
CB: Adding summaries would make your work even quicker.
DD: I’ve seen the results of the summaries. I’ve also seen where it can go wrong. One problem is, how accurate is the initial transcript? We’ve spent a lot of time trying to get the transcripts to be more accurate. Another question is, how accurate is the summarizer? We’re working on those questions now. Currently Minutes results are not something that we publish on our website at Michigan Public as “The Truth.” It’s meant to give reporters that first pass and allow them to discover if the story is this interesting or worth spending more time on.
CB: If a City Council person says one word off mic, it could mess up the entire interpretation of the transcript.
DD: With different city councils, you’re dealing with very different sound quality. Our system is built with the assumption that someone is going to double-check whatever they find and then actually confirm what was said.
CB: Is Michigan Public the only organization currently using the tool?
DD: No. In fact, we’re working with NPR right now on recruiting more member stations to try it out. I’m reaching out to folks at member stations around the country to sign them up for the tool, getting their communities into our system, and then collecting feedback from them. I think we have seven states represented in our database.
CB: How can news organizations use this tool?
DD: Go to searchminutes.com or email me. We’re getting public media stations signed up as part of this pilot, but you don’t necessarily need to be a public media station to join.
- The AI Patrol is Ramping Up - October 25, 2024
- All-AI Radio Is Here - October 18, 2024
- Marketers Are All In on AI - October 4, 2024
Leave a Reply