As artificial intelligence moves from the realm of science fiction to reality, many questions remain about the effects it will have on the American workforce, particularly regarding its potential to displace workers and eliminate jobs. In this episode of Mastering Innovation on Sirius XM Channel 132, Business Radio Powered by The Wharton School, Jeremy Gilbert, Director of Strategic Initiatives at the Washington Post, presents a positive view of AI’s growing role in journalism. He discusses the newspaper’s recent implementation of new artificial intelligence tools that are now executing simple tasks in the newsroom, freeing up journalists to conduct more substantial reporting.
Often a poster child of post-internet disruption, print journalism has had to scramble to adapt to the digital age as revenues wane and social media and independent online news outlets compete for readers’ attention. To enhance the breadth and depth of its reporting, the Post has been experimenting with a new automated program called Heliograf, which can present wide swaths of factual information and keep it continuously updated more effectively than a human can. Gilbert elaborates on the intricacies of the AI-powered tool and its use during the 2018 midterm elections, which the Post used to cover congressional races across the country.
An excerpt of the interview is transcribed below. Listen to more episodes here.
Saikat Chaudhuri: I’ve been reading a lot about the AI tools Heliograf and ModBot. Can you tell us about that?
Jeremy Gilbert: The Post is interested in the ways that AI can complement what our human reporters do. At the end of the day, we have some of the world’s best reporters, and we invest very heavily in that. We’re never looking to replace them. On the other hand, we want to minimize the amount of rote or mundane work any reporter does. Instead, we want to make sure that those reporters are doing things that only they can do. Both Heliograf and ModBot are different ways of trying to achieve the same goal, which is to say, “How can we ensure that the information our consumers need is out there, but at the same time that our human reporters are working on stories that only they can?”
In the case of Heliograf, when we were telling the story of the 2018 elections, we wanted to make sure the Washington Post reached about 90 million unique visitors per month in the United States. We’re reaching people in most congressional districts. We want to make sure that we tell the story of every congressional election. We could assign reporters to every one of those stories, but most congressional seats go to incumbents. Even in an election that features as much change as 2018, only a handful of seats shift. One of the things we can use an AI-powered system to do is to help cover those congressional districts where there is not expected to be change.
What’s interesting about Heliograf is that it allows us to cover those congressional races in a very different way. Instead of turning out one story the morning after the election, with Heliograf, we can tell a story that, at the beginning says, “Here are the people who are vying in the primary to represent the Democrats or Republicans and the independents in this particular race.” Then during the primary night, we count the votes and Heliograf can report every 30 to 90 seconds what is going on, who’s ahead, who’s behind, etc. We can preview the general election with whoever comes out ahead in each of the primaries, and on the election night, we can count votes again.
“The truth is that there was more concern from outside the newsroom than inside about how human reporters would feel working alongside these AI systems.” – Jeremy Gilbert
We can even look at things if it’s a statewide race. For example, in the state of Florida, how do the three counties in south Florida, which tend to be more Democratic, compare to the panhandle, which tends to be more Republican? We can compare turnout from 2018 to 2016 every 30 to 90 seconds. As new data comes in, we can publish a new story. Heliograf allowed us to tell more than 500 stories, updated continuously, starting with the primaries and going through Election Day. It allowed us to cover a story differently than we would have covered stories with humans, but it especially meant that none of our human reporters had to write a story that said, “This is how many votes are in and this is how many votes are out.” Instead, they could concentrate on doing interviews, analysis, and getting the aspects that only humans could get.
Chaudhuri: These bots are now good enough so they can publish some of these basic articles?
Gilbert: Yes, and similarly, when we talk about comment moderation, ModBot is all about that. We want humans to look at the comments and engage with our audience. It’s incredibly important, but computers are pretty good at the 80/20 rule. They can say, in the case of moderating the comments, “We don’t necessarily need to deal with 80% of these comments. For 20% of these comments, we need a human to make the final decision.” We save our humans a lot of time, because they only need to engage with the ones that the system is not already certain it knows how to handle. In a lot of cases, the goal with this AI is, “How do we make people’s regular lives easier?”
Chaudhuri: That resonates with me. A few months ago, when IBM Watson demonstrated that an algorithm can more accurately diagnose certain types of glaucoma and other eye diseases better than the best experts could, people got concerned about the replacement issue you talked about, but they demonstrated that if you put together the expert opinion with the algorithm, the outcomes were dramatically better. That’s the potential that we see.
Gilbert: It’s all complementary. The truth is that there was more concern from outside the newsroom than inside about how human reporters would feel working alongside these AI systems. In the end, the challenge has been that people have identified aspects of their work they think should be automated, moreso than that people are afraid they will lose some aspect of their work. As long as you’re in a high-performing industry like the reporters in the Post newsroom are, where they know what differentiates them from their peers, it’s not work that one can automate — at least not today, and I can’t imagine at any time in the future.
Chaudhuri: Fascinating. Clearly, you’re on the cutting edge and doing a lot of great things at the Washington Post. It couldn’t have been easy though, because newspapers were disrupted by all these online media outlets, and there are so many different forms of how news is consumed, through social media or other online types of offerings. How did you guys survive this? How do you guys not only survive but thrive in this environment? What did you do organizationally?
“If you’re in the storytelling business, you cannot be completely devoted to the channel that you publish in.” – Jeremy Gilbert
Gilbert: There are a couple of things that have worked at the Post. We got incredibly lucky when the Graham family sold the Washington Post in the fall of 2013. They went out to try to find a buyer who was willing to invest a little bit in the newsroom and company, and Jeff Bezos has been incredibly generous, giving the Post the runway that we needed to invest in both the technology and journalism side of the business. In truth, the Post is well positioned and even was then, because we have a brand that stands for credibility, honesty, and objectivity in the work that we do. That’s been incredibly important, especially now. In the news business, you need people to trust you and understand that your company has been doing great work for a long time.
The other thing you have to know, and the Post has known this as well, is that if you’re in the storytelling business, you cannot be completely devoted to the channel that you publish in. Like many news organizations, things have been difficult for the print side of our business, but what we’ve been able to do is grow the digital distribution dramatically. That comes from a couple of different things. The Post as an organization has been willing to experiment on a lot of different channels. Some work out better than others, and you can’t experiment as an organization unless you’re willing to tolerate that some of your experiments won’t be as successful as others.
Some of the things we’ve done have worked incredibly well. Some of the things we’ve done we had to stop doing, or we stopped doing for a time until we could come up with a better way to handle it. Because of that, we’ve been able to benefit from the successful experiments that Jeff Bezos has pushed pretty hard and to double down on the things that work. We built a network of all of our freelancers to make it easier for our editors to commission freelance stories when, for example, there’s a developing situation in a specific place. We can easily reach out to freelancers in that place and often be the first to publish from there. That’s something that when we tried and it worked, we could pour more effort into it. Again, just like Heliograf or ModBot, it didn’t replace the journalists we had; it made them more effective. We were able to get information faster and publish those situations first. By being willing to invest in our most successful projects and kill the things that weren’t working, it made a lot of difference.
About Our Guest
As director of strategic initiatives at the Washington Post, Jeremy Gilbert leads the newsroom to identify, create and execute unique digital products and storytelling experiences. He reimagined election night experiences, created the Washington Post’s first virtual reality stories, built a freelance network that changes how the Post covers national stories and launched a new leadership vertical. Gilbert works closely with the engineering, product design, graphics, audience, analytics and advertising teams. Previously he led digital strategy at National Geographic, taught journalism at Northwestern University, worked at the Poynter Institute and art-directed a couple of newspapers.
Mastering Innovation is live on Thursdays at 4:00 p.m. ET. Listen to more episodes here.