The architecture of the internet creates risks for democracy

The first line of evidence comes from naturalistic quasi-experiments from which we can infer the causal impact of the rollout of internet hardware on relevant outcome measures. For example, the rollout of broadband in the US 20 years ago was affected by state “right-of-way” laws, which govern how easy it is for telecommunications companies to lay cables along public roads and land corridors. Some states imposed far more onerous conditions than others before digging could commence. Using this variation in regulation as an independent variable, one study showed that broadband availability increased affective political polarization.

SIGN UP FOR THE AWARD-WINNING SCIENCEADVISER NEWSLETTER The latest news, commentary, and research, free to your inbox daily

There is now a solid body of evidence showing that internet availability is causing a variety of outcomes that adversely affect democracy. However, these studies leave unanswered the question of why and how these effects occur. Why would access to fast broadband make people more polarized and more extreme?The answer may have something to do with platform algorithms, such as curated newsfeeds (e.g., on Facebook) or ranking of posts (e.g., the “for you” feed on X). Algorithms have long been in the sights of researchers and regulators as potential culprits of polarization because of their opacity and their known focus on maximizing user engagement and platform dwell time with little regard for the quality of curated content.Recent auditing studies have examined the political implications of algorithm design in the US and Germany. In the US, after Elon Musk endorsed Donald Trump for president in 2024, Republican-leaning X accounts received a measurable boost in visibility relative to Democrats, and Musk’s own posts accumulated 17.1 billion views between July and November 2024—surpassing all political campaign advertising on the platform. During the German federal election in 2025, an algorithmic audit of X, TikTok, Instagram, and YouTube found that around half of all party-related content that was algorithmically recommended to young users across platforms involved an extreme-right party, doubling its audience share relative to the content’s original upload rate on TikTok. Center-left parties, by contrast, were suppressed.

A similar result was observed in an analysis of all posts by 436 German politicians on X during the election campaign, which similarly found that the X algorithm disproportionately amplified content from parties at the political extremes, in particular on the extreme right, and systematically suppressed parties in the center of the political spectrum.Those algorithmic biases have demonstrable behavioral consequences. A recent field experiment re-ranked content expressing antidemocratic attitudes and partisan animosity on X. When anti-democratic content was down-ranked, participants’ outgroup animosity and negative emotions declined compared to participants in the control condition who were exposed to the standard X algorithm—implying that the algorithm is favoring, or at least failing to guard against, anti-democratic content. The changes observed in that experiment were comparable in magnitude to 3 years of increasing affective polarization in the US.Remarkably, many platforms are demonstrably aware of the risks they pose to democracy. Under EU legislation (the Digital Services Act; DSA), platforms must file annual systemic risk assessments of their operations and how they might affect democracy, and Bing, X, Snapchat, and TikTok all highlight the risks of echo chambers in their reports.Fortunately, the problems emerging from algorithmic curation are, in principle, solvable. The experiment that identified the problematic role played by the X algorithm in prioritizing anti-democratic content also identified a potential solution: the experiment was possible only because the researchers developed an algorithm that could downrank anti-democratic content—suggesting that the same technology could be deployed by platforms at scale in the interest of democracy.Finally, a fundamental attribute of social media is that they give rise to “homophilic” networks. Homophily refers to the natural tendency of individuals to form bonds with similar others. Birdwatchers join birdwatching groups online, and bowling aficionados hang out with fellow bowlers, and so on. A unique attribute of online networks is that they permit homophily to emerge even for fringe views: people who think the Earth is flat can link up online as easily as birdwatchers—which would be impossible in real life because “flat-earthers” are few and far between.Second, if a person holds a political belief that they think is widely shared, when in fact it is a minority viewpoint, then it is unlikely that that person would be satisfied by a government acting on behalf of a majority-based mandate.

As a result, that person might feel disenfranchised from “remote elites” that do not serve “ordinary people”—in other words, the familiar vocabulary of populist politics emerges naturally from people holding false-consensus beliefs.A recent large study in Germany confirmed the link between false-consensus beliefs and populist attitudes. Respondents were asked to estimate public support for seven controversial policies (e.g., abolishing the right to asylum, higher taxes on the rich) and indicated their own opinions on those issues. Individuals who systematically overestimated support for their own positions scored higher on all three dimensions of populist attitudes—popular sovereignty, anti-elitism, and Manichean worldview—with the association holding across the left-right spectrum. The fundamental ability of online social networks to bring together people—the very essence of social media’s appeal—thus carries within it the seeds of the corrosion of democracy by facilitating the creation of a multitude of diverse and sometimes fringe communities that all feel empowered by like-minded peers but ignored by a government acting on behalf of the true majority.So will democracy survive the internet? Perhaps, but the available evidence calls for robust protective action, such as debiasing algorithms and devising tools to help people be better calibrated to the prevalence of their own views, lest we end up in a polarized and fractured authoritarian society.