For years, a primary metaphor for the Internet has been the “town square,” an infinite space for free expression where everyone can have their say. But as scaled digital platforms have grown to dominate most of modern life, metaphors that focus solely on speech have failed to explain our current civic dysfunction.
Perhaps the better way to understand the Internet is to compare it to a much older infrastructure problem: city-wide sanitation systems. Posted content is akin to water; websites and other interfaces are analogous to pumps; and unintended feedback loops equal the risk of infection. A public health framework for understanding the Internet would focus not on online information per se but on how it is generated, disseminated, and consumed via digital platforms.
The genesis of this model lies in the two-century-old history of early clean water advocates in Victorian England. At the time, the life-threatening diseases ravaging cities—cholera, typhus, tuberculosis, and scarlet fever—were not new. What was new were modern living conditions. Infections that might have taken weeks to spread through a village suddenly ravaged entire populations within days, and no one understood what was causing the massive outbreaks.
The Victorian working class knew who to blame when disease broke out: doctors. Mobs attacked members of the medical establishment, leaving government officials unsure how to weigh doctors’ safety against the public interest. Why anger? The traditional response to disease—quarantines—had become ineffective in industrialized cities, leading the public to mistrust those who profited from treatment.
The first serious approach to the problem was taken by a coalition of doctors, liberal advocates and social reformers beginning in the 1830s. Known as miasmists, they pushed the idea that noxious air was the culprit in epidemics. If a neighborhood didn’t pass the smell test, the argument went, you immediately knew it was already too late to be saved.
Miasmists, including prominent ones such as Florence Nightingale, have an ambivalent legacy. They were among the first to emphasize that disease had not only biological but also social and economic causes, a crucial insight. But at the same time, they were completely wrong about the role of air in the spread of popular diseases of the time, a reflection of an elitist worldview and overwritten morality.
This tension manifested itself during two important events. One was the Great Stink of 1858, where a combination of hot weather and poor waste management turned the Thames into a cesspool. The stench was so bad that even the curtains in the Houses of Parliament had to be caked with whitewash. No one was safe from the foul air, and with the assumptions of the miasmists, that meant no one was safe from disease. But in fact no major eruption followed the Great Stink.
Second was the pioneering work of a brilliant physician, John Snow, who for years had suspected that water (not air) was the actual cause of urban epidemics. In a painstaking natural experiment, Snow showed that the Broad Street pump was the source of the 1854 cholera epidemic in the Soho area of London. His data revealed that residents living across the city became ill if they accidentally got water from the pump, although a nearby brewery that got its water from another source had no recorded cases. There was no other plausible explanation: Some as yet undiscovered mechanism, located at the pump, was responsible for the infection. Although Snow was careful to frame his findings so as not to explicitly reject the miasma theory, the implications were obvious.
After much debate, over the next 20 years London implemented the world’s first modern sewage system. And from 1850 to 1900, urban disease changed from a problem of individual circumstances and negligence to a problem of economic dependence and social interconnectedness. When it became clear that not only medical personnel but also efficient water lines and safety valves were needed, public policy shifted from one-time treatments to longitudinal assessments of population health, fueled by new mechanisms of evaluation. Public health efforts had changed: If cholera epidemics continued, they did so only because cities refused to provide drinking water to vulnerable populations.
Today we live in an online version of the Great Stink and are in dire need of John Snow’s methods. Evidence is quickly mounting that social media is causing major harm on a large scale, particularly in terms of declining mental health and societal trust. But since these effects are not directly measurable (except what has been revealed from whistleblowers and difficult natural experiments), like Snow, we are left to speculate about causes while trying to find better data.
What would it take to build something like sanitation infrastructure for social media or generative AI? As we argue in detail in a recently released project, that would mean building evaluation tools to connect design features—like the feedback loops embedded in content recommendation systems—with population outcomes like mental health outcomes.
To extend the metaphor, current technology interventions tend to focus on moderation strategies that are centered on specific users and individual pieces of content. This is similar to the role of nurses in public health, crucial and under-resourced providers of well-being. But just as no one should believe that good stewardship is the best way to deal with dirty water, content moderation is insufficient to deal with dysfunctional platform architectures.
Modern platforms already function as experimental laboratories, running randomized controlled trials over and over again to improve results based on business goals. What we need are tools to assess the pump – the models and interfaces of platforms that determine how populations are exposed to content over time – to assess whether restrictions need to be implemented to protect at-risk groups. For potential concerns such as effects on mental health or systemic reductions in trust, platform effects would be evaluated alongside internal growth and revenue metrics. And just as epidemiologists learned to focus on infants and children as particularly vulnerable subpopulations, today’s researchers must pay special attention to crucial risk vectors, such as chronic social media use by youth.
Sanitation not only made epidemics easier to control and mitigate; it made the diseases themselves easier to understand, eventually leading to germ theory. Beginning with the early experiments of Louis Pasteur, the new science of bacteriology confirmed the existence of microorganisms, which John Snow had only suspected. When the specific bacterium responsible for cholera was identified under a microscope, a new cornerstone of public health was established.
We are in a similar moment now: We have strong ideas about the causal mechanisms that may mediate harm from products (such as interpersonal comparisons between teenagers leading to mental health problems). But just as 1850s London did not need germ theory to begin evaluating the effects of water and establishing sanitation systems, the first step to mitigating damage in large-scale models is to establish baseline effects independent of explanation. The lesson from public health is that such baselines will be necessary to build consensus about what platforms and large language models need to measure and optimize for.
We can continue to treat tech platforms like a town square where the loudest, ugliest voice wins the day. But instead of metaphors that blame individuals, or encourage us to just log out when things get harmful, we can embrace the standard of public health. The solution will come not from more content moderators or ever smarter chatbots but from new infrastructural commitments: pipes, valves and pumps that would actually keep users safe.