Last Monday, 15 films were awarded the Pulitzer Prize for Journalism. In a first for the prestigious award, he revealed that two of his winners used his AI to create their stories.
“We don't have direct knowledge of the precedent,” said Pulitzer Prize administrator Marjorie Miller. “Previous data has shown that winners may have used low-level machine learning applications. This is the first time we have explicitly asked the question.”
In March, Alex Perry revealed to the Nieman Institute that five of this year's 45 finalists used AI to research, report, and draft submitted articles. reported that it did. As the cycle of hype and fear surrounding generative AI plays out in newsrooms across the country, the top finalist ultimately went to machine learning for use in investigative reporting.
Local reporting award winner City Bureau and Invisible Institute's “Missing in Chicago” trained custom machine learning tools to comb through thousands of police misconduct files. The New York Times' Visual Investigation Desk trained a model to identify the crater of a 2,000-pound bomb in an area of Gaza marked safe for civilians. The article was one of several he won as part of the paper's international reporting awards package.
Miller also acknowledged three other finalists who disclosed their use of AI. These include a local news series on the government's response to Hurricane Ian by the Villages Daily Sun, a newspaper that covers a large retirement community in Florida; It included a Bloomberg investigation and explanatory reporting on whether it is fueling the global pandemic. About industries that use water.
I spoke to the journalists behind two Pulitzer Prize-winning AI-powered stories about how they brought machine learning to their investigations and what other newsrooms can learn from their work. I heard about it.
Community-based data journalism
This year's Pulitzer winner for local reporting was the series “Missing in Chicago,'' which exposed systemic failures in the Chicago Police Department's response to investigations into missing and murdered black women. The series, published by the Chicago-based nonprofit City Bureau and Invisible Institute, has been years in the making, and one pillar of her reporting was her machine learning tool, named Judy. .
“We used machine learning to parse the text of police misconduct records, specifically document types that have a narrative within them,” said Invisible Institute data, which shared Pulitzer with City Bureau reporter Sarah. Director Trina Reynolds Tyler said. Conway.
Reynolds-Tyler began building Judy in 2021 as part of the Invisible Institute project, which processes thousands of CPD misconduct files released by court order. The files spanned from 2011 to 2015. She involved members of her Chicago community in Judy's development process, and ultimately her 200 volunteer workers read and manually labeled the fraud files. In other words, these volunteers created Judy's training data.
Although they may not have been experts in AI, Reynolds-Tyler believed that people in this affected community inherently understood CPD data. Even if they didn't come into the company with a language for writing machine learning algorithms, they had lived experience that outsourced data labelers don't have. Ultimately, Judy was able to bring forth her 54 allegations of police misconduct related to missing persons in her four years there.
In their Pulitzer Prize-winning investigation, these 54 cases became a kind of roadmap for Reynolds-Tyler and Conway's other reporting. The themes of these 54 of his cases confirm the pain and neglect of families whose loved ones have gone missing in recent years. This proved that these cases were not isolated, but part of a history of systemic failure by CPD.
Reynolds-Tyler also encouraged other reporters who rely on machine learning tools to see the value of embedding them in the communities they cover and basing their data on real places and real people. I hope you understand. “We have to be in the business of taking people into the future,” Reynolds Tyler said of his implementation of AI in his investigative reporting. “They help us look and seek understanding.”
find a pattern
In the international reporting category, a December 2023 report by the New York Times Visual Investigative Desk was one of several recognized stories about the Gaza war. A Pulitzer Prize-winning team has trained a tool that can identify craters in the remains of a 2,000-pound bomb, one of the largest in Israel's arsenal. The Times used the tool to examine satellite imagery and confirmed that hundreds of bombs had been dropped by Israeli forces in southern Gaza, particularly in areas that had been marked safe for civilians.
“There are a lot of AI tools that are basically just powerful pattern recognition tools,” he said. Ishaan Jhaveri, is a team reporter and specializes in computational reporting techniques. He explained that if you need to sift through mountains of material for a research project, you can train AI algorithms to recognize what patterns you're looking for. It could be someone's voice in hours of audio recordings, a specific scenario described in a pile of OSHA violation reports, or in this case, the outline of a crater in an aerial photograph.
Jhaveri said the team determined that an object detection algorithm was best suited for the investigation. They turned to a third-party platform called Picterra to train their algorithm. This allowed us to manually select craters in satellite images uploaded to the platform, and slowly trained Picterra to do the same automatically at scale.
One of the advantages of turning to Picterra was its superior computing power. Satellite imagery can easily exceed hundreds of megabytes, or even gigabytes, Jhaveri said. “Local development of satellite imagery would naturally be clunky and time-consuming,” he said, suggesting that many newsrooms simply don't have the infrastructure. Platforms like Picterra handle the processing power.
After eliminating false positives (shadows, bodies of water, etc.), the visual survey team finally found that by November 17, 2023, there were more than 200 craters in southern Gaza consistent with this type of bomb, It posed a widespread threat to civilians seeking safety throughout the South. In its investigation, the Times noted that “more bombs were likely used than those captured in our reporting.”
“We didn't use AI to replace manual tasks; we used AI because these are the types of tasks that would be extremely time-consuming to do manually. [it would distract from] Other investigative work is also required,” Jhaveri said. In his words, AI helps investigative reporters find the needle in the haystack.
