The Uncertainty Principle

How sure are we that being sure is the right thing to be?

The old saw, “if your mother says she loves you, check it out with a second source,” underscores a key principle of journalism: Let’s be really careful to check, and double-check, everything we print. Stories are about what we know; not what we don’t.

Should we rethink that? And might we serve communities better if we did?

Before you break out the tar and feathers, a clarification: I’m not suggesting that we back away from the core values of verification. Everything we publish should be true; or at least, we should have some confidence it’s true. If we quote a politician saying something outrageously untrue, at least we know it’s true he said it. (Ideally, we’d also fact-check it.)

I’m talking about something else: About what we choose to publish and not publish. About the economic forces and technological constraints that we sometimes mistake for mission.

Bear with me; it gets a little complicated. (When has it not, you ask?)

What we think of as news judgement is built around both what we think is newsworthy — ie, what we think is of interest to our audience — and what resources we have. A newsroom may have 50 ideas to pursue, but if it only has 10 reporters, it’ll likely just pursue the top 10. And that puts a premium — entirely reasonably — on stories where you can assert something, rather than write about things where information is more ambiguous.

Stories are built around resolving uncertainty — “The mayor is corrupt” is a much better headline than “We don’t know, maybe there’s something here, maybe there isn’t.” Sometimes we don’t assert the certainty directly, partly because of legal risk, but generally we only do that if we’re convinced, in our heart of hearts, that what we’re implying is true. “The mayor is living beyond his means, and he won’t explain how, but we notice that he has a lot of rich friends who have gotten a number of no-bid contracts from the city” is a story that has a core conviction dressed up as just-asking-questions uncertainty.

It’s part of our process: we hold stories until we can nail things down; we write around information we don’t have. And all of that makes sense; in a world of constrained resources and a one-size-fits-all distribution model, we have to make choices about what stories we pursue that are of interest to the most people. And that general means picking stories that have new information or resolve unanswered questions. Right now, we wouldn’t assign a reporter, let alone publish a story, saying it’s not clear how budget cuts might affect art classes in a specific school. There’s no economic upside in publishing something uncertain about a topic only a handful of people will care about.

But a world where AI intermediates information turns some of those dynamics on its head.

The cost of producing that story for a single reader — assuming we have the underlying facts and context at hand — collapses to nearly zero. Even better, we can get a very clear signal from that individual reader that that story is important to them — because they ask for it, whether explicitly or implicity, through their behavior. Imagine a story like this:

“The school board has cut 10% from the budget, and specifically cited art classes as being a wasteful use of taxpayer money. But the principal of your son’s school has a long track record of supporting art lessons as an important component of education, and has previously been successful in pushing the school board to reverse budget decisions.”

Would it be better if we could resolve the uncertainty — if we could assure the reader that art classes will survive, or that art classes will definitely be canceled? Sure. That used to be our job, and here we’d be shifting some of that responsibility to the reader — Here’s what we know, here’s what we don’t — and trusting that they have the motivation to make sense of it.

We wouldn’t be telling them anything that’s untrue — so we remain true to our principles — but we’d be much more ambiguous about a conclusion. It’s more akin to a barroom conversation with a knowledgeable reporter — or perhaps a cleaned-up version of a newsroom Slack channel on the subject — than the pronouncements of truth you might find on the front page.

Is this more what a news story looks like in an AI age? And do we serve readers better if we embrace this mission?

Some part of this idea is not radical: We’ve always lived with some level of uncertainty in stories, not least in headlines like “Uncertainty over oil prices roils markets.” (Leaving aside the fact that I’ve never heard an actual human say “roil” in conversation.)

What’s radical is the idea that we might do this for stories that fall outside our current window of what’s newsworthy (I’ve written about this before) and, more importantly, might be prepared to present an untidy, incomplete picture of the world to readers, albeit rigorously fact-checked and with all the relevant context included.

That’s yet another cultural hurdle we’ll have to confront.

Are there legal risks to doing this? Sure. But if we build the systems right, there should be no more risk that we face with that hypothetical story about the mayor and his friends I invented above. We aren’t speculating, per se; we’re telling you what we know and what we don’t. It just happens to be much more personalized to a query you have.

There is also a darker side to this future: One where incomplete information is weaponized rather than simply helpful. A world where a parent can ask, are my kid’s art classes likely to be canceled is also a world where a partisan voter can ask, show me why the mayor may be corrupt. That’s a system where our work is used to confirm bias rather than offer perspective.

To be clear: That use case will happen, with or without us. It’s doubtless already happening in multiple chat windows today. The question is whether, if we build systems like this, we’re contributing to the problem or the solution. Systems that simply respond to queries are at risk of confirming users’ biases; what we need to do is design systems that can bring broader perspectives and context so readers get as clear-eyed a view of the world as possible — even if they disagree with it. That’s a design question, not a technology one. And if we don’t do it, someone else — who may not share our mission and principles — will.

Of that, I’m certain.