Stray Thoughts

Is journalism incentivized for polarization? And does AI offer a possible solution?

TL;DR: Maybe yes, and maybe yes.

For insights on both questions (and answers) I turn to the ever-insightful Jonathan Stray, a senior scientist at the UC Berkeley Center for Human-Compatible AI (and recovering journalist).

We’ll get to the first question in a bit, but to skip ahead to the second, Jonathan recently released results of a fascinating experiment to see if different algorithms in social media feeds might lower polarization. His latest research shows that, perhaps some algorithms could, while also increasing time on site — meaning, there’s a business incentive to do that.

That’s a really interesting finding that points to one possible path out of the culture wars and shouting matches we seem to find ourselves embroiled in. And I want to suggest other, possible ways that we could harness AI to lower the temperature — or at least build bridges between warring sides, even if I don’t have the empirical backing that Jonathan’s work does.

But let me back up, and get to that first question: Does the way we practice journalism contribute to polarization?

That’s a provocative question, to say the least, but in all the time I’ve known Jonathan, he’s never shied away from asking the tough ones. He brings a wealth of different perspectives to every question, not least because of his wide-ranging and diverse career — from software to journalism (I first met him about 15 years ago when he was about to graduate from Hong Kong University’s Journalism and Media Studies Center) to teaching to looking at how AI systems select what information to show and writing about conflict resolution/reducing polarization.

He raises the question — and gives a good answer — on a really smart and wide-ranging podcast interview he did late last year (and which I only discovered now.) A lot of it has to do with broader questions of polarization and how to reduce it — and it’s all interesting — but one part caught my eye, which comes after a discussion about how lowering the heat on culture wars is less about fact-checking or proving one side is right or wrong, but more about presenting “pluralistic fairness,” which, as Jonathan puts is, is ‘an answer that people who hold multiple views, who hold differing views would all agree is fair.”

Which is, sort of, the idea that rather than trying to bludgeon the other side into agreement with your — obviously superior — facts, the right thing to do is to expose both sides to reasonable arguments from both sides; at the very least, it helps us not caricature the other side as crazies and lets us see their point of view, even if we disagree with it.

But that’s often not what we do, per Jonathan:

…journalists are not trained to think about peacebuilding and conflict, with some exceptions. But generally, that’s just not what they think about and the frame they use. So I’ve been thinking a lot about this, in combination with some other folks in the journalism and journalism training spaces and the bridge-building space: what journalism would have to look like to be non-polarizing in the face of divisive political actors?

And it’s not very complicated, really. It has to be pluralist. You have to charitably represent the views on both sides, or all sides, which used to be a mainstream tenet of journalism, but actually disappeared for a number of reasons, including economic reasons.

And more troublingly, he notes, the structural economic and business incentives point us in the wrong direction. (And, no, it’s not a screed about clickbait; it’s an argument against serving your loyal readers too well.)

There’s a fundamental economic shift that happened in journalism over the last two decades or so, which is that news organizations went from being majority ad-supported to majority audience-supported subscriptions. This sounds great and in many ways it is a more sustainable model. However, advertisers don’t really care about politics, right? They just want their products sold. They don’t care which side the newspaper is on or whether it reports this viewpoint or that viewpoint. Audiences do.

They certainly did when the Washington Post, on orders from owner Jeff Bezos, decided to stop endorsing political candidates; a quarter-million readers walked out in that first week. One way to look at that is as a stinging rebuke of a craven decision; another is that you buck your audiences’ world view at your economic peril. And the same could be said for another number of publications — Fox News, for example. That’s a lesson for any editor or publisher to take note of: that there’s little business upside to trying to present as clear a picture of other points of view.

And in any case, most news articles are built on the idea of synthesis: of ingesting and analyzing a flood of disparate information and ideas, and coming to conclusions to be presented to readers. They’re not, by and large, organized to show multiple different views and not choose between them.

To be sure, some newsrooms are trying to fill this space: Tangle, which presents the best three arguments on any given topic from both (or more) sides, is probably the poster child for this. Isaac Saul, who founded it, will present his — or his colleague’s — view as well, but only as one perspective among many. I don’t always agree with Isaac, but I always learn something from his newsletter.

And as a plug: Semafor, which I also work, has its version of this as well: Our story format, or “Semaform,” separates out and clearly labels the facts from our reporter’s analysis — and also includes the best counterargument to it, under a section labeled “Room For Disagreement.” Stories often also feature “The View From” other players or parties on the topic. (Is this paragraph self-serving? Sure. But I’m proud of what we did.)

The problem is scalability. Tangle is tiny, and Semafor is a startup with big ambitions but a still-modest newsroom. How can we make this idea work at scale — and equally important, not make it feel like required reading that readers will gloss over?

Or as Jonathan notes:

And it’s very akin to the classic problem in journalism, which is sometimes called the ‘eat your broccoli problem”, which is like, “Yeah, yeah, you want to read about the latest celebrity gossip, but maybe you should know something about the war in Ukraine,” which is going to be cognitively and emotionally more difficult.

Can — and you knew this was coming — AI help?

Maybe. The bot I built that takes reporting on the same topic from multiple outlets and summarizes what they all agree on, pulls out the parts that are unique to each, and shows how each news organization frames the issue is one step in that direction. It addresses two issues: One, it lets you see multiple perspectives, and two, it does it in a way that feels less like homework and more like a way to efficiently parse a lot of information. I can’t guarantee someone will simply skip over the New York Times or Fox News part of the output, but I can say that the bot positions them as ways to see what your favorite news outlet may have missed.

Or my deconstruction bot, which doesn’t do exactly the same thing, but carefully — and, to be fair, somewhat over-pedantically — takes apart a story and shows readers what unstated assumptions they embed, is another way to offer a meta-view of an issue and push users to question what they’re reading.

And I’m certainly no polarization expert or researcher, unlike Jonathan. I tend to think of this from the supply side; what can newsrooms do? He tends to look at the landscape; what information do people get, and how do they engage with it, which is an equally important — possibly more important — frame.

Will any of this actually work, and more importantly, work at scale? We won’t know until we try.

But the broader idea that Jonathan is flagging is worth sitting with for another beat: That journalism, as it is practiced today, is perhaps as much a part of the problem as it is the solution.