Can AI Liberate Journalism?

Can a computer cover the courthouse as well as a reporter with a notebook and a pulse?

That question popped into my mind as I read two recent news stories. One was about how artificial intelligence programs like ChatGTP could write news stories as well or better than humans. The other covered how the nation’s largest news chain slashed its staff so drastically that those who traditionally cover court houses — local news reporters — could be a dying breed.

I must confess a bias. I started out as a courthouse reporter for the Des Moines Register and I can’t imagine a computer developing sources such as Betty Klunder or Dot Knauth, two Polk County courthouse clerks who tipped me off to many stories that landed my byline on page one of the Register. Both helped me navigate the intrigues and gossip of the courthouse corridors that gave me great start in journalism. How could a computer top Betty and Dot?

photo by Gertrūda Valasevičiūtė

Then I met Richard Boyd here in North Carolina, and he made me question whether a computer could beat me at my own game. Boyd is a thought-provoking high-tech entrepreneur, investor, author and speaker who has decades of experience in high-tech applications for everything from education to health care and national security. He’s worked for defense contractor Lockheed Martin, who bought his last company and where he created and led Virtual World Labs, a virtual reality simulation that allowed the Pentagon to drop troops into a computer-generated world where they would encounter languages, cultures and military threats that real soldiers would soon reckon with overseas.

He’s founded high tech companies in Chapel Hill and helped create several pioneering computer gaming companies, including Red Storm Entertainment with author Tom Clancy; iRock Entertainment with Ozzy Osbourne; and Timeline Computer Entertainment with author Michael Crichton. In other words, he knows the high-tech world better than Betty and Dot knew the courthouse and certainly better than me.

Boyd says technology like the recently unveiled ChatGPT-3 (and recently 4) enables computers to do more than writing; they can also do basic reporting tasks, such as taking highly structured data like a baseball game or a police blotter and write something that is virtually indistinguishable from what a human reporter might write. The systems do have some limits.

“Because it (the data) is highly structured, there’s no reporting behind the statistics,” he told me in an interview. “You know, who got the hits, how did the pitchers do, how many errors, how did the innings play out, that kind of thing. And they’ll use language from previous reporting to mix it up and make it sound like it’s something original. But it’s not. I think that anything that has a pattern to it or is formulaic like court reporting, and again I’m thinking more of something like the police blotter, how many break-ins were there, or events happening, that’s all formulaic.”

But Boyd says the journalistic limits of formulaic reporting doesn’t mean a machine with new generative AI capacities couldn’t be configured to do sophisticated court reporting much better than I did in my days at the Register.

“I think today we have a system that can read court stenographer information, if it has access to it, and distill that down into some kind of reporting based upon a huge base of examples of similar cases,” Boyd says. “And I think we’re now at the point where that same system, if it needed to get more information from individuals, could email, or text or even call with a computer-generated voice and get responses. I think you could do that today.”

“So, if the system can see you and start asking questions, can it effectively be a polygraph, can it tell if you are telling the truth or not? And the answer is yes,” he says, “with some level of confidence. So, what do you have then? You have a lot of the components that makes a good court reporter.

“That’s the world we live in today,” Boyd says. “I think the fundamental question is not whether a machine (or computer) can do journalism the way a human would, but could it report the way a machine would do it and is that better or worse?

Boyd readily acknowledges the limits of a computer-generated journalism model. How would a source, for example, react to a call, email or text from a computer asking it a question?

“Would people respond to questions the same way as they would to a skilled reporter.,” he asks, “someone who is trying to build trust and ask questions of to get (a source) to open up the same way we see in movies or probably the way you experienced? I think that’s more difficult.”

But he cautions, machine learning and artificial intelligence systems enable computers to do many things that people didn’t think they could a few years ago, like play chess better than Garry Kasparov, the Russian world chess champion defeated in a six game match by IBM’s Deep Blue supercomputer in 1997.

And they have advantages that human reporters could never match, such as perseverance. “Machine learning systems have no choice but to stick to whatever their utility function is. They never sleep and they don’t need health insurance. They have access to all the information on the net and they can read it at machine speed,” Boyd says, “Humans have not had an upgrade since the Pleistocene (the advent of modern humans), but machines are upgrading every second. They’ve got storage and processing speeds and methods of manipulating information. It’s possible now for machines to talk to each other and ask questions and make phone calls. Could an artificial intelligence system that’s managing the information in a courtroom talk to other artificial intelligence systems,” he asks, “and report at machine speed? At that point, I’d say a reporter is at a disadvantage unless someone tells the machine not to talk to other AI reporters. There are a lot of interesting issues here.”

One of those interesting issues is what will the profit seeking hedge funds and newspaper chains that control most of the nation’s local media do if they adopt the tools of artificial intelligence and machine learning?

If you have any doubts about that question, read Joshua Benton’s devastating profile of Gannett, the nation’s largest newspaper chain, in Nieman Lab. He details how the company slashed half its U.S. staff in four years, including many reporters who cover mayors, city councils and, yes, courthouses, at its 217 daily and 175 weekly newspapers to preserve profits and pay its hefty debts. What would Gannett do if it could replace reporters with machine learning and AI systems? I think I know that answer.

Journalists need a way to work on their own, not for companies like Gannett. AI systems and machine learning could blaze that trail. Boyd says the nation’s best hope is to figure out a way to arm individual reporters with machine learning and artificial intelligence systems that have common goals with journalists.

“Humans that pair well with AI machine learning systems will be able to outperform organizations and individuals that don’t,” he says. “I think we are in a Simulation Century right now, and a stand-alone human being is insufficient, a handicap. The central question of the century is how to achieve the right balance of humans to machines to optimize outcomes. In the short term, a good reporter augmented with a machine learning system that is serving their goals is going to be a real powerful thing.”

—James O’Shea

James O’Shea is a longtime Chicago author and journalist who now lives in North Carolina. He is the author of several books and is the former editor of the Los Angeles Times and managing editor of the Chicago Tribune. Follow Jim’s Five W’s Substack here.