Two students behind a computer screen, with one in an orange shirt pointing at the screen

ASU reporters are paving the future for investigative journalism using AI

Jake Garcia

October 30, 2024

For a line of work often referred to as the Fourth Estate, the role of journalism, the press and news media has been a longstanding pillar of American democracy. Professional journalists and reporters play an important part in holding the powerful accountable – a watchdog relationship that is contingent upon trust with the public and cooperation from public officials.

In today’s media climate, trust and cooperation come at a premium. It’s within that context that students at the Howard Center for Investigative Journalism at Arizona State University (ASU) are incorporating generative artificial intelligence (AI) and ChatGPT into their work.

As part of the AI Innovation Challenge – a collaboration between ASU and OpenAI where ChatGPT Edu licenses are provided at no cost to approved projects – 15 students and additional faculty and graduate assistants are using the large language model (LLM) for a variety of investigative reporting techniques.

Mark Greenblatt, a professor of practice at the Walter Cronkite School of Journalism and Mass Communication and executive editor at the Howard Center, is overseeing the widespread use of this technology for the first time this semester in an immersive reporting course, which includes a mix of undergraduate and graduate students.

The course — which Greenblatt teaches alongside Lauren Mucciolo and Angela Hill — is the capstone experience for the master’s degree program in investigative journalism. Students are tasked with hands-on investigative reporting projects where they uncover, research and publish stories of significant importance. For Greenblatt, incorporating ChatGPT into the class increases both efficiency and credibility.

“We are finding that people who have never used a large language model before are trying it out for the first time and it’s really changing their entire world,” he said. “It’s turning them into much more powerful investigative reporters at a much earlier stage in their career.”

Prompt engineering helps provide access to public records

Greenblatt acknowledged that one of the biggest struggles for investigative journalists is to convince government officials or agencies to allow access to public records and information.

Despite requirements in place that make it mandatory for certain records to be made available, officials often cite the burden it is for them to locate the information and will sometimes attach an exorbitant cost to large databases.

To solve this, Greenblatt and his students ask ChatGPT to write a piece of code, which they provide to government public relations officials, who often don’t have data backgrounds and do not understand how easy it can be to fulfill even large data requests. With the help of simple code from ChatGPT, the Howard Center provides tangible examples for officials to understand the journalist requests are not nearly as burdensome or time consuming as they initially predicted.

“ChatGPT will write the simple code, and we'll insert that into a letter back to the records custodian,” Greenblatt said. “It gives that person the ability to go to their IT department with something in writing. They give it to the IT department, they crunch the numbers, and then they give us the records that we wouldn't get otherwise. ChatGPT makes the conversation very easy for us to negotiate.”

Greenblatt cited one instance where a response aided by ChatGPT brought down a records request tabbed at over $2,000 to just $40. In other cases, students have used ChatGPT to run complex data analysis on large datasets – a process that would normally take hours is performed in seconds.

Eshaan Sarup, a student in the program, has used this technique to great effect.

“What ChatGPT is very good at is giving you a base code to start with,” he said. “It's been really useful for helping me analyze, visualize and merge data.”

For agencies reluctant to disclose information (and those attempting to hide it), Greenblatt and his students take ChatGPT a step further: They prompt it to write a legal argument and cite legal precedent to detail the unlawfulness of withholding information. Howard Center students are required to check the accuracy of every legal direction ChatGPT suggests, allowing student investigative reporters the chance to almost instantly find cases they might never locate on their own.

Greenblatt imparted that seeing a student journalist empowered in such a manner is inspiring.

“When you see a young reporter who has never fought for a record in their life, and you see them very quickly and very powerfully going up against seasoned lawyers and winning and getting public records, democracy is working better in the Howard Center right here at Arizona State University,” he said.

Accuracy is everything, so ChatGPT remains under scrutiny

Students at the Cronkite School are quickly made privy to the mantra: “If your mother says she loves you, check it out.” That mindset undoubtedly applies to their use of generative AI.

“What we teach in our unit is that you don't just lean on ChatGPT, you have to interact with it,” Greenblatt said, noting the additional requirements students must follow if using the LLM in their reporting process. “We don't just ask them to disclose that ChatGPT was used to do a particular calculation or to do a task. Students provide the hyperlink so that we can go back in and actually see the input, the prompt, the output, and how it was manipulated.”

Greenblatt admitted there have been instances where ChatGPT has referenced legal citations that don’t actually exist. Thus, remaining skeptical and continuous fact-checking are necessities to ensure accuracy and fairness.

“In journalism, we're only as good as the trust that we have with the public,” Greenblatt said.

Greenblatt doesn’t envision a scenario where AI completely takes the role of the reporter. Given that the core of investigative reporting is uncovering information that is not available publicly, and a large language model is only as good as information that is already accessible; having it attempt to create an investigative report from scratch would be a failure.

“You have to make sure you're actually using your brain and thinking through what you want,” Sarup said.

Still, for the specific instances already in use, Greenblatt views ChatGPT as a powerful tool and important teaching mechanism.

“We're opening up government and making democracy better because of ChatGPT, and because the students have access to it,” he said.

The Howard Center for Investigative Journalism at Arizona State is supported by the Scripps Howard Fund. Its mission is to train the next generation of reporters through hands-on investigative journalism projects. The Scripps Howard Fund supports philanthropic causes important to the E.W. Scripps Company, with a specific focus on journalistic excellence.