In 1964, Philip Converse published what would become one of the most cited papers in all of political science, “The Nature of Belief Systems in Mass Publics.” ^{It was a result of a movement at the University of Michigan, where Converse taught, to do large-scale data analyses of the ideology of the American public — or, perhaps more accurately, the lack thereof.}

It’s common to think about political ideology as something that can be measured along a single left-right axis, with each individual person inhabiting a spot somewhere between its extremes. Draw a single straight line and you can fill it up with an array of ideologues. Starting from the “left,” you might run into Bernie Sanders, then Elizabeth Warren, then Kamala Harris, then Barack Obama, then Bill Clinton, then, I dunno, Jon Tester? Keep moving “right” and you’ll find Lisa Murkowski, then Mitt Romney, then Mike Lee, then Tom Cotton, then Ted Cruz, then Josh Hawley. Reasonable people can argue about the exact positions on that line, but it’s a straightforward, one-dimensional spectrum that defines the way pundits talk about American politics.

What Converse argued was that few Americans actually had an ideology coherent enough to be meaningfully placed on that line. “Conservatives” and “liberals” are each associated with a set of beliefs that, to the politically knowledgable, map together in logical ways. Do you oppose abortion rights? You’re supposed to also be anti-immigration, in favor of lower taxes on the wealthy, against regulations on businesses, and pro-gun rights. Ah, but do you think gun control is a good idea? You’re supposed to also support strong action against climate change, oppose prayer in public schools, back civil rights protections for disadvantaged groups, and favor a strong social safety net.

Those are groupings that make sense to people who appear on cable news panels. But Converse and his colleagues found that most Americans were nowhere as coherent in their individual policy views. One might be anti-abortion rights but also anti-climate change. Or support mandatory school prayer but also want to tax the rich. These Americans didn’t see these positions as mutually contradictory because they didn’t see these distinct issues as inherently entwined. Or to use the phrase that Converse memorably coined for it, they didn’t know “what goes with what” — which views political elites considered the “correct” ones for a particular point on that left-right spectrum.

When Converse was writing, the two major American political parties mirrored that mishmash of views. The Democratic Party contained both Black civil-rights advocates and Alabama white supremacists. The G.O.P. included both the far-right John Birch Society fringe and socially liberal Rockefeller Republicans. Some people considered that a problem in need of solving. And over the next few decades, the parties became more ideologically sorted — first among political elites, and then later among the general public. And a big part of that has been about regular people learning “what goes with what” — what mix of policy preferences define a “conservative” or “liberal” or “democratic socialist” or “right-winger.”

How do people learn that? They learn it from watching political elites, of course. Let’s say you like Alexandria Ocasio-Cortez — you think her spot on that left-right line roughly coincides with yours. Some new policy question comes up that you’ve never given a moment of thought. If you find out that AOC has a certain position on that topic, you’re very likely to adopt her position as your own.

But they also learn “what goes with what” through the media. If you love Fox & Friends and watch it religiously, the positions the hosts back every morning are quite likely to become your own, even if you’ve never given them a moment’s independent thought. (Ditto Rachel Maddow.) All of this has made the country awfully hard to govern. Ideas that might sound perfectly reasonable in the abstract sound ruinous as soon as they’re associated in your mind with the “other team,” whoever that team might be.

If it sometimes feels like Americans are fighting the same few fights over and over again — no matter the underlying issue, no matter the reliability of the evidence base — the ease with which Americans today learn “what goes with what” is a big reason why.

This brings us, of course, to the Los Angeles Times’ opinion section and one of the worst editorial “innovations” of the past few decades of journalism.

Earlier this month, the L.A. Times started putting little AI-generated taglines at the bottom of many (though not all) of its opinion pieces — including editorials, regular columns, and outside op-eds. These little snippets attempt to do three things:

• summarize the views expressed in the piece,
• highlight other views that are in opposition to the piece’s perspective, and
• assign the piece a spot on that left-right spectrum. As in, “This article generally aligns with a Center Left point of view,” or “a Right point of view,” and so on.

There are, of course, 28,928 things to complain about here. The AI, being AI, sometimes makes things up out of whole cloth, or decides to argue the Klan wasn’t so bad after all. Laura went over many of those things a couple of weeks ago.

But it’s that last one — the assigning of a one-dimensional ideological label to an opinion — that I hate with the fire of a thousand suns.

Let’s imagine you’ve just read an L.A. Times opinion piece. You’ve found it interesting enough to stick with it all the way to the final sentence. Maybe you find it persuasive, maybe you don’t, but either way you’ve just invested a few minutes into hearing its arguments. Then, underneath it, an AI tells whether you’re supposed to agree with it, based on whether or not it aligns with your self-conception. If you’re a self-identified conservative, and you find a piece convincing, what do you think is likely to happen when at the end you’re informed that you’ve just read some soy-boy, feminazi, DEI propaganda? Or if you’re a proud liberal who’s told that what you just read was Trumpist B.S.? You’re telling people “what goes with what” at the worst possible time — when they’ve just invested time in hearing out an argument that might change their mind. It’s as if someone thought American political discourse was too healthy and needed some roughing up.

I called up Lilliana Mason, the Johns Hopkins political scientist, to get her take on this. She wrote Uncivil Agreement: How Politics Became Our Identity (2018), which explains how American political identities have grown far more regimented over the years, far more closely entangled with how we see ourselves as a fundamental level. She was roughly as horrified as I was at what the L.A. Times was doing.

“The social psychology research really demonstrates that, when people read information that they think is coming from ‘their team,’ they are much more likely to immediately accept it,” she told me. “They’re even willing to change their own opinion in order to match their team’s when that occurs. So if you want your opinion pages to be persuading people, then this tool is doing the opposite.

“It’s making these arguments harder to to get through to the other side. As soon as you read that thing at the bottom, you realize, ‘Oh, I shouldn’t have believed any of this,’ or ‘Yeah, l’ve always thought this, I’ve always believed this to be true,’ even if it just told you about something totally brand new.”

Mason also made the very reasonable observation that not every opinion can be slotted onto a simplistic left-right spectrum. (“The ideological spectrum is one of the most complex concepts in political science. Like, we’ve written hundreds of books about it, and we still don’t fully agree on what it is and how to define it. So to say that Al could do it is, I mean, it’s laughable, right?”) And even for the ones where that might be plausible, it’s entirely unclear that an opaque AI can do it well. (This piece by pro-Trump columnist Scott Jennings, “President Trump came through for Los Angeles,” could, with a few more exclamation points, have come from the White House communications office. It aligns, the AI says, “with a Center point of view.”)

“My guess is that it’s looking for words that tend to be associated with liberal or conservative articles,” she said. “But you can imagine that that would make a lot of mistakes, because sometimes liberals or conservatives write critiques of the other side, and they use the other side’s language. I would be shocked if it was really good at this.”

But even if AI was perfect at the task, it’d still be destructive to democratic discourse to take every opinion and wrap it in an ideological box. Mason’s academic instincts proposed an experiment where the ideological labels were randomized — so different people were told the same article was either liberal or conservative — to see how their opinions were affected. Would conservatives adopt liberal views if they were told they were conservative, or vice versa? (There’s plenty of evidence that Americans have more liberal policy views than the candidates they vote for.) It’s an interesting question that would make for a good academic paper. But there’s no reason for the L.A. Times to be running that experiment on its readers.