Five AI Projects for Local Newsrooms
Aimee Rinehart, AI Program Manager
Ernest Kung, AI Product Manager
October 2023
We started the AP’s Local News AI initiative in August 2021 in relative obscurity. We sent out a 32-question survey and had to hustle to explain to newsrooms why it was important to take this survey, and why AI readiness was poised to be important to newsrooms. No one could have guessed that the release of ChatGPT would change everything in terms of awareness and over-hyped interest on the topic if AI, specifically generative AI.
Now, two years on, we heard back from at least six newsrooms who demurred to take our first survey ask how they can get involved with our work. Don’t worry: a follow-up survey is forthcoming, and we hope that with AP, there will continue to be ways to be involved in the topic and share best practices.
In all, we received almost 200 survey responses from news leaders in all formats and from every state, Guam, Puerto Rico and the District of Columbia. We wrote a report of the findings with a list of what everyone else is doing regarding tools and profiles of newsroom leaders who convey curiosity and also concern about AI. We share those conflicted feelings and have come to rely upon our network of now more than 4,000 to share insights and explore limits.
Interest in Generative AI, that is the wholesale making of text, images, videos and audio, has been high, and we caught the fever, too. In fact, three of the five projects include facets of Generative AI — OpenAI’s GPT 3.5 and Whisper to be specific. We found that the results were consistently strong and cost effective.
None of the projects write full articles. Two of the projects use templated AI, or natural language generation (just like AP’s earnings reports from 2014), because hurricane alerts and police incidents need to be reliable, and so we could not leave these jobs to generative AI.
Newsroom and University Partnerships
The projects’ superpower were our university development teams. At Northwestern University’s Knight Lab led by Professor Jeremy Gilbert, Chief Nerd Joe Germuska led the student team of Mame Coumba Ka and Maria Aragon for El Vocero de Puerto Rico. Knight Lab Senior Engineer Scott Bradley and University of Michigan Professor Mark Newman worked with Michigan Radio. Professor Serdar Tumgoren at Stanford University worked with students Kalyn Epps, Ryan Leahy, and Ozge Terzioglu for the KSAT-TV project. Professor Prasad Calyam at the University of Missouri led students Hemanth Sai, Xiyao Cheng, Kiran Neupane, Anirudh Kambhampati, Roshan Neupane, and Harsh Joshi to work on the Brainerd Dispatch and WFMZ-TV projects. Using student developers kept the projects costs low, but it also gave students a project that was meant to be useful and used — not a bridge to nowhere. Students also come with a healthy dose of optimism and fresh ideas and we believe our industry can benefit from both. Already, we know of at least one of our student developers being hired by a news organization. It’s a good start and we need more people like them.
Overcoming Challenges
The local news leaders who brought us these projects have somehow managed, in the face of severe economic challenges, to continue to dream about how their newsrooms can run more efficiently. Many of the news leaders said they see projects like these as a way to attract and retain talent by removing tedious tasks from staff. We worked with managing editor Renee Richardson at the Brainerd Dispatch who said the automated police blotter we created with the University of Missouri was estimated to spare her staff hours each week to focus on more enterprising work.
Several of the people we worked with at the start of the project changed jobs or left their newsroom, which mirrors industry trends. Maria Arce brought us the El Vocero de Puerto Rico project and left to study newsroom crisis response plans at the University of Michigan. Bernice Kearney, news director at KSAT-TV, left after 30 years at that station to be the news director at the KPRC-TV. Brad Gowland, the newsroom technologist for Michigan Radio who we profiled in our report, left the station to work for a different department at the University of Michigan. In these cases, the projects continued despite their departures.
We signed agreements with all newsrooms to define intellectual property rights and liability, which proved to be too high of a hurdle for one broadcast news group. Looking back, we might have taken another strategy with collaboration agreements. Another project almost stalled because an executive several states away was unsure about the return on investment of time in that newsroom to work on the project versus keeping workflow the same. In these moments, we felt for the news leader who brought us these projects because they saw their staff needed help. Having a project killed by someone on the other side of the country felt deeply unfair to the people who are left to contend with the realities of too much information and not enough people to help sort and make sense of it all.
We managed to save one project for the original newsroom but had to look for another home for a separate project and found a locally owned station in Pennsylvania that had already participated in earlier phases of the Local News AI initiative. WFMZ-TV welcomed the project and the assignment of identifying and marking thousands of emails to help train an AI to make a decision on inserting new planning calendar items.
What’s Next
After working with local newsrooms and riding the AI wave this year, one thing is certain: It will take all of us working together to figure this out. We are glad AP started this journey a decade ago and that we have delivered projects into five newsrooms that have the potential to scale to other newsrooms that can take the code base and operationalize it. Our goal has always been to help the journalists who remain in shrinking newsrooms contend with the onslaught of information. We hope other news leaders will consider these tools to help their staff with repetitive tasks. Our team has prepared case studies and open sourced much of the code, which you can find below in the next section. While a lot has been said about the threat of AI taking jobs away, we believe these tools might help keep journalists in the newsroom by reducing burnout.
Thank you to now-retired SVP of Strategy Jim Kennedy and to VP of News Partnerships Lisa Gibbs for bringing this project to AP. Thanks to the John S. and James L. Knight Foundation for the financial support to make these projects a reality.
Access the Resources
Project 1: El Vocero de Puerto Rico (Case Study | Open-Source Code)
Project 2: Michigan Radio (Case Study)
Project 3: KSAT-TV (Case Study | Open-Source Code)
Project 4: Brainerd Dispatch (Case Study | Open-Source Code)
Project 5: WFMZ-TV (Case Study and Open-Source Code available soon)
