AI may reverse political polarization. Don't get excited yet

Political polarization has become so extreme that when Democrats and Republicans find common ground on a single issue — like their shared hatred of AI data centers — it becomes major news.

But those same data centers could also be mending the nation’s political divisions. The large language models people increasingly rely on to understand the world are far from benign, but they are quietly — if unintentionally — promoting consensus in both how and what the public learns about the world. For better or worse.

It seems impossible to think that such a thing could happen. Research and commentary dedicated to analyzing today’s fragmented media ecosystem have, at times, made it seem like we’ve reached the point of no return. Looking back at the roots of our extreme political polarization underscores why we may be ripe for a reversal of that trend.

An ever-shrinking overlap between the two major parties began in the 1970s, but it intensified in the 1990s and beyond. Why that happened is the subject of longstanding debate, but one particularly compelling explanation traces the shift back to evolutions in the media landscape during that same period.

The rise of radio and television broadcasting resulted in a handful of stations and networks catering to a mass audience in the 1950s and 1960s. That meant the average person got much of the same information, even if there might be slight variations between outlets.

Those of us old enough to remember that era can recall legendary anchors like Walter Cronkite , who presided over CBS Evening News , building a brand based on overt neutrality. Such was the faith in this kind of media figure that Cronkite typically closed his broadcasts with a revealing catchphrase: “And that’s the way it is” — as if there was no other version of the truth. This was problematic in that it tended to oversimplify reality, banishing dissident voices to the margins. As an antidote to polarization, though, it proved quite effective.

That carefully constructed shared reality fell apart in the late 1970s, with the rise of cable television. Then Congress effectively removed barriers to entry in 1984 with the Cable Act, which allowed the number of cable networks to explode. Instead of three or four broadcast networks, Americans could seek out channels that suited their particular political and cultural sensibilities. By 1998, the U.S. had 174 different national cable networks — it was 82 just six years earlier — along with thousands of local cable stations.

The rise of the internet only accelerated these trends, giving everyone with access to a computer the ability to go down various rabbit holes. The resulting amplification of conspiracy theories and other fringe ideas was bad enough — then social media said, “hold my beer” and effectively obliterated any vestiges of a national common ground. Different platforms gave rise to a collection of adversarial digital tribes living in their separate realities.

Consensus and moderation prevailed when citizens relied on a highly centralized and homogenous group of media outlets. When that system was replaced by a multiplicity of outlets and sources, polarization intensified to dangerous levels.

Enter AI. Critics have predicted that the technology’s chatbots and models will sow disinformation, further fueling polarization. Yet there’s some early evidence to suggest a different outcome.

Consider how people rely on the internet for information. In the past, someone might “Google” a particular topic and then spend an inordinate amount of time clicking through links, searching for results that interested them or perhaps resonated with their own biases. Or they might turn to their own social media connections, effectively amplifying their own prejudices.

AI has upended that system. Increasingly, we’re looking at what has been described as a “zero-click” future, where someone posing a query instead accepts an AI summary rather than exploring the cacophony of the internet. Or AI is relied on more directly, as when people pose questions to chatbots, which many often trust as much as, if not more than, other humans.

These developments represent a return to the spirit, if not the reality, of broadcast networks and moderating voices. This is inevitable, given the way that AI works.

Models are trained on vast collections of information in order to generate words and phrases. As such, they’re probability machines, producing the likeliest answer based on training data. This means that chatbots and AI-generated search results tend to move to the middle, distilling an otherwise invisible consensus about any given issue. Answers come across as a reassuring, neutral and believable summary that comes across as the objective truth. Something like a digital Walter Cronkite . Little wonder that, even as Americans tell pollsters that they don’t trust the “media,” they increasingly turn to AI for information.

Some hint of where we’re going was captured in a recent report on election disinformation from the Brennan Center for Justice . It found that AI chatbots, far from amplifying fringe theories of election fraud, did the opposite, gently pushing back on these claims and banishing them. A study published in the Journal of Computer-Mediated Communication found that individuals who interact with chatbots about political issues tended to moderate their views.

That’s all well and good, but we should not forget that AI models have a built-in bias toward conventional wisdom, not minority or fringe points of view. The public has been so conditioned to associate “fringe” with bad that we might forget those views can lead to a more equitable society. Many civil rights that we now regard as necessary began as ideas far outside the mainstream.

As our reliance on AI grows, a decades-long drift toward polarization may reverse. But the resulting consensus could come at a cost: a new kind of groupthink that is no less dangerous for democracy than the polarization it replaces.

This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners. Stephen Mihm, associate dean and professor of history at the University of Georgia, is coauthor of “Crisis Economics: A Crash Course in the Future of Finance.”