Why History Matters for the Future of Health
Public health is often described as the science of preventing disease, prolonging life, and promoting health through organized effort. At its heart it is something less tidy. It is the record of how societies decided whose lives mattered, what they were willing to spend, and who was permitted to act.
That record is worth reading now, because the questions in it have not changed.
Every instrument public health has ever adopted arrived with the same problem attached. Someone had to decide the instrument was warranted. Someone had to find the authority to deploy it. Someone had to build the thing that made it work at scale, and answer for it when it failed.
This chapter is not a survey of medical progress. It is an examination of what had to exist around each innovation before the innovation produced health.
Ancient Wisdom and the First Public Health Systems
Evidence of organized health effort dates back millennia. In the Indus Valley, cities such as Mohenjo-Daro featured covered sewers and wells supplying fresh water to much of the settlement. These early planners recognized that clean water and waste disposal were not amenities but conditions of urban survival.
The Romans extended the idea. Aqueducts brought water from distant sources. A network of sewers, including the Cloaca Maxima, remains one of the oldest surviving sanitation systems in the world. Public baths made hygiene a social norm centuries before anyone knew what a bacterium was.
Two things are worth noticing about these systems, and neither is the engineering.
The first is that they required a governing body capable of compelling construction, funding maintenance, and enforcing use over generations. A sewer is not a discovery. It is a sustained institutional commitment rendered in stone.
The second is that their public health value depended on reach. A water supply protects only the people it reaches. Coverage was part of the mechanism, and extending coverage was an institutional achievement rather than an engineering one.
Epidemics as Catalysts for Collective Action
Epidemics have a way of exposing what a society is actually organized to do.
The Black Death killed a substantial share of Europe's population in the fourteenth century. In its aftermath, Mediterranean port cities developed formal systems for isolating arriving ships, travelers, and goods. Over time, forty day isolation periods gave us the term quarantine.
Quarantine is worth pausing on. It is not a medical technology. It is an assertion of authority over movement, property, and commerce, exercised on the basis of incomplete information about an invisible cause.
The innovation was legal before it was epidemiological.
Cholera produced a similar lesson five centuries later. Repeated outbreaks in nineteenth century London and other cities demonstrated the limits of individual precaution against contaminated water supplies. Moving away from a sick neighbor does nothing about the pump you both drink from.
Chadwick and the Machinery of Public Health
What happened next in Britain is the clearest example in this chapter of evidence becoming authority, and it is usually skipped.
Edwin Chadwick's report on the sanitary condition of the labouring population, published in 1842, argued that disease and poverty were connected through filth, drainage, and water supply, and that the cost of the resulting illness fell on the public purse. It was an argument built on numbers, and it was aimed at legislators rather than physicians.
The Public Health Act of 1848 followed. It established a General Board of Health and permitted the creation of local boards with powers over drainage, water supply, and nuisance removal. It was contested, its powers were limited, and the central board was dissolved within a decade.
None of that diminishes what it demonstrates. A body of evidence was converted into a statute, a statute into an administrative body, an administrative body into inspectors, financing, and a permanent obligation to act. That sequence is what turns a finding into a health outcome.
Chadwick was also wrong about the mechanism. He believed disease spread through foul air rather than contaminated water. The infrastructure his argument produced reduced cholera anyway, because clean water and drainage address the actual transmission route regardless of why anyone thought they should be built.
That is worth sitting with. The institutional apparatus outlived the theory that justified it.
Germ Theory and the Birth of Modern Medicine
Until the mid-1800s, prevailing theories of disease centered on miasmas, or on moral failing. Illness was read as punishment or personal weakness.
That began to change when Louis Pasteur demonstrated that microorganisms were responsible for fermentation and spoilage, and later for disease in humans. Robert Koch isolated the bacteria responsible for anthrax, tuberculosis, and cholera, and established the postulates that linked specific pathogens to specific diseases.
The scientific achievement is well documented. The institutional consequence is less often noted. Germ theory changed what public health authorities could justify requiring. Boiling water, sterilizing instruments, isolating cases, inspecting food: all of these became defensible impositions because a causal mechanism existed to defend them.
Evidence did not confer authority. It changed what existing authority could reasonably justify.
Evidence Without Acceptance
The reverse case is equally instructive, and it happened first.
In the 1840s, working in the obstetric clinics of the Vienna General Hospital, Ignaz Semmelweis observed that maternal mortality from puerperal fever was far higher in the division staffed by physicians and medical students than in the division staffed by midwives. He identified a plausible route of transmission, physicians moving between autopsies and deliveries, and instituted a requirement that they cleanse their hands.
Deaths in his division fell sharply.
The finding was correct. It was also rejected. Semmelweis lacked a mechanism his colleagues would accept, since germ theory did not yet exist to supply one. He lacked institutional standing sufficient to compel practice beyond his own division. And the finding carried an accusation that senior physicians were themselves the vector, which is a difficult argument to win against the people whose behavior you are asking to change.
He died in 1865 without seeing the practice adopted.
Joseph Lister's antiseptic methods, introduced two decades later and eventually supported by germ theory, faced professional resistance of their own before becoming standard surgical practice.
The pattern in both cases is the same and it is the pattern this book is about. Correct evidence, in the hands of someone without the authority or the standing to act on it, does not become practice. Something else has to happen.
Vaccines and the Power of Prevention
Edward Jenner's work with cowpox in 1796 produced the first successful vaccination against smallpox, a disease that had killed and disfigured for centuries. Over the following two centuries, vaccination transformed the control of smallpox and many other infectious diseases, eventually contributing to the eradication of smallpox itself.
It also produced organized resistance almost immediately. Anti-vaccination leagues in nineteenth century England opposed mandatory smallpox vaccination laws on grounds of personal liberty and distrust of government.
That resistance is instructive precisely because the vaccine worked. Efficacy did not settle the question. What was being contested was not whether the intervention was effective but whether the state was entitled to impose it, and whether the people being asked to accept it had reason to trust the institution asking.
Those two questions have never gone away.
Smallpox: The Vaccine Was Not Enough
The eradication of smallpox is one of public health's clearest demonstrations that an effective instrument and an effective system are not the same thing.
The vaccine had existed for more than a century and a half when the intensified eradication effort began in 1967. Smallpox was still endemic across large parts of Africa, Asia, and South America. Having the instrument had not been sufficient, and mass vaccination campaigns alone had not closed it out.
What worked was a change in method. Surveillance systems to find cases quickly. Local reporting networks that made finding them possible. Containment around each case, vaccinating contacts and contacts of contacts rather than entire populations. Cold chains to keep vaccine viable in places without reliable power. Freeze-dried vaccine and the bifurcated needle, which reduced the dose required and could be used by workers trained in a day. Sustained international coordination and financing. And local health workers, community leaders, and volunteers doing the finding, persuading, and vaccinating in settings the coordinating bodies could not reach directly.
The last naturally occurring case was recorded in 1977, and eradication was certified in 1980.
The instrument was necessary and it was nowhere near sufficient. What produced eradication was surveillance, logistics, financing, coordination, and local participation, assembled around a vaccine that had been available since the eighteenth century.
Mapping Disease
In 1854, during a cholera outbreak in London, the physician John Snow mapped cases in Soho and identified a contaminated water pump on Broad Street as the likely source.
Snow presented his findings to the local Board of Guardians, which agreed to remove the pump handle. The outbreak was already subsiding, so the intervention cannot simply be credited with ending it. What matters here is the institutional sequence. Snow produced evidence, and acting on it required someone with authority to make a decision.
At nearly the same time, Florence Nightingale used data visualization to show that poor sanitation killed more soldiers in military hospitals than battlefield injuries. Her coxcomb charts converted statistics into an argument, and the argument moved political leaders to fund hospital reform.
Together, their work demonstrated both the power and the limit of analysis. Analysis identifies. It does not intervene.
Professionalizing and Globalizing Public Health
In the late nineteenth and early twentieth centuries, health boards and sanitary inspectors evolved into formal health departments. A profession emerged, trained in statistics, disease control, and policy.
Internationally, repeated cholera pandemics pushed nations toward shared quarantine standards and data sharing, culminating in the first International Sanitary Conferences, which were precursors to the World Health Organization.
This period is easy to skip because it produced no discoveries. It produced something arguably more consequential: standing bodies with defined authority, permanent staff, and the obligation to act on what they learned. Before this, response to disease was episodic and improvised. After it, there was an apparatus.
Most of what public health accomplished in the twentieth century depended on that apparatus existing.
The Turn Toward Risk
Something changed in the middle of the twentieth century that matters more for this book than any single discovery.
The Framingham Heart Study, begun in 1948 in a Massachusetts town, followed a large group of residents over decades, recording their characteristics and their subsequent cardiovascular outcomes. It did not identify a pathogen. It identified associations. Blood pressure, cholesterol, smoking, and other characteristics were shown to be related to the probability of future heart disease.
The concept of the risk factor came substantially from this work, and it changed what public health could claim to know.
Before this, the field's causal language was largely infectious. This organism causes this disease. Afterward, it could say something different and stranger. This characteristic is associated with an elevated probability of that outcome, in populations, over time.
That is a probabilistic statement about a person who is not yet ill, derived from a population they resemble. It is very close to what a predictive model produces.
And it introduced the problem this book examines. A probabilistic finding about a person who feels well does not tell anyone what to do. Whether it produces health depends on whether the person is told, whether they can act on it, whether a clinical or public health system is organized to help them act, and whether anyone is responsible for what happens if nobody does.
Prediction entered public health long before artificial intelligence did. The field has been living with the problem of what to do with probabilistic knowledge for decades. Artificial intelligence has not created that problem. It has enlarged it.
When Evidence Meets Organized Opposition
Tobacco is the clearest demonstration that evidence alone changes nothing when the determinant of health is commercial and political.
By the early 1950s, published studies had linked cigarette smoking to lung cancer. The United States Surgeon General's report in 1964 concluded that smoking was causally related to lung cancer in men. The evidence was public, substantial, and contested by an industry with resources to contest it.
What followed was not a straightforward decline. It took decades, and it took instruments rather than findings. Warning labels. Advertising restrictions. Excise taxation. Restrictions on smoking in workplaces and public spaces. Litigation and the disclosure of internal industry documents. Cessation programs. Public education sustained across generations.
Consumption fell substantially over the following decades in many countries, and it fell because evidence was converted into law, price, and physical restriction.
Two lessons carry forward.
The first is the persistence of the gap. Strong evidence accumulated long before the full range of policy instruments that eventually changed exposure and behavior. Knowing was not acting, and the delay itself carried a public health cost.
The second is that the opposition was organized. Public health is accustomed to obstacles that are passive: insufficient funding, weak infrastructure, competing priorities. Tobacco control faced an opponent with a strategy, a budget, and an interest in the finding not being acted upon.
That is worth remembering when the subject is a technology sold by companies with a commercial interest in how it is evaluated.
Failures That Shaped Modern Practice
Some of the field's greatest failures are more instructive than its successes.
The United States Public Health Service study at Tuskegee, conducted from 1932 to 1972, enrolled Black men in Alabama in a study of untreated syphilis without adequate informed consent and deceived participants about the nature of the research. Even after effective treatment became available, researchers did not provide it as part of the study. The study became an enduring symbol of medical exploitation and institutional betrayal.
The relevance to what follows in this book is direct and it is not metaphorical.
Tuskegee was not principally a failure to generate information. It was a failure in what an institution did with its power, its knowledge, and its obligations to the people from whom that knowledge was obtained.
That configuration is worth naming carefully, because it recurs. An institution collects information about a population. The institution's interests and the population's interests diverge. The population has no visibility into what is being done with what was collected, and no mechanism to contest it.
Systems developed from data that poorly represent the populations in which they are deployed may perform unevenly across those populations. Models trained on historical decisions may also reproduce patterns embedded in those decisions.
But the deeper lesson is about standing. Communities that have been studied without benefit have reason to be skeptical of the next instrument, and that skepticism is not irrational resistance to be overcome with better communication. Rebuilding requires transparent process, participation in setting priorities, and mechanisms ensuring that benefits reach the people who supplied the data.
That is a legitimacy problem, and legitimacy is not something an accurate system can supply.
Community as Author
There is a counterexample worth placing alongside Tuskegee, because it shows the same institutional relationship running the other direction.
The early response to HIV and AIDS in the 1980s was slow, and the communities most affected were among those with the least institutional standing to demand otherwise. What followed was unusual in the history of public health. Affected communities organized, and they did not confine themselves to demanding attention.
They pressed for changes in how clinical trials were designed and who was eligible for them. They pressed for faster review of promising treatments and for expanded access outside trial protocols. They challenged surveillance and reporting practices on privacy grounds. They built community based prevention and support services when institutions did not. They developed sufficient technical literacy to argue with researchers about methodology on the researchers' own terms.
Much of what they demanded was adopted, and some of it changed research practice permanently.
The lesson is not simply that advocacy works. It is that the people a health intervention is aimed at can be participants in designing it, and that interventions designed this way tend to be both more acceptable and better.
That possibility remains available now. Yet health technologies can still be designed for populations without giving those populations a meaningful role in determining what is built, how it is used, or what happens when it affects them.
The Infrastructure Imperative
Running underneath all of these histories is a requirement that is easy to state and hard to fund.
Vaccines work only with cold chains, trained personnel, and reliable supply. Antibiotics require laboratories for diagnosis, pharmacies for distribution, and education to prevent misuse. Sanitation systems need engineering expertise, ongoing maintenance, and sustained political commitment to fund both.
Artificial intelligence carries requirements of the same kind, and they are routinely underestimated. Effective systems need connectivity, electricity, hardware maintenance, software updates, data security, and trained staff at every level. They also need oversight mechanisms and quality assurance.
Pilot programs frequently fail to scale for exactly this reason. A diagnostic system that performs well in a well equipped urban hospital may fail in a rural clinic with unreliable power and limited technical support. A predictive model that depends on complete electronic records will produce unreliable results where record keeping is inconsistent.
Treating infrastructure as secondary to the algorithm reverses the historical lesson. It has never been the discovery alone that determined the outcome.
What the History Has in Common
Across these histories, several recurring conditions become visible.
Interventions that endured acquired local ownership rather than remaining external mandates. Smallpox eradication depended on local health workers and community leaders doing the finding and persuading. The HIV response was reshaped by the people it was aimed at.
They were adapted to the populations and settings in which they operated. The freeze-dried vaccine and the bifurcated needle existed because the original method could not survive the conditions of the places where the disease remained.
They were integrated into institutions capable of sustaining them. Chadwick's evidence became durable only when it became a board with inspectors and a budget. Semmelweis had a finding and no institution, and the practice waited two decades.
And they changed as evidence, conditions, and populations changed. The efforts that stalled were frequently the ones that could not revise.
None of those conditions belongs to the technology itself. All of them describe the relationship between an instrument, an institution, and the people whose health is affected.
How AI Differs from What Came Before
Having established the continuity, it is worth being precise about the discontinuity, because there is one.
Past innovations targeted specific pathogens or processes. A vaccine addresses a disease. A sewer addresses a transmission route. Artificial intelligence is notaddressed to anything in particular. It is a general capability for finding patterns, applied across surveillance, diagnosis, resource allocation, communication, and administration at once.
Three consequences follow.
It can change without notice. A vaccine formulation is fixed until deliberately reformulated. A model's behavior can shift as the data flowing into it shifts, without anyone deciding that it should.
Some systems are substantially less transparent than earlier public health instruments. Their outputs may depend on relationships across large numbers of variables that cannot always be translated into a simple account of why a particular result occurred. The mechanism by which a sewer prevents cholera can be explained to anyone.
And error replicates differently. A mistaken judgment made by one person may remain relatively localized. A mistaken assumption embedded in a widely deployed system can be reproduced across every case the system assesses, at speed and with remarkable consistency, until someone detects it.
These are real differences and they raise the stakes. What they do not change is the structure of the decision. The instrument still requires an institution capable of acting on what it produces, resourced to respond at the volume it generates, authorized to intervene, and trusted enough to be permitted to.
Lessons for an AI-Powered Era
Progress in public health has never been automatic.
Sanitation, germ theory, and vaccination transformed population health where communities funded infrastructure, institutions built and kept trust, and someone held the authority to act on what was known. Where any of those were missing, the science was just as correct and the outcomes were considerably worse. Semmelweis was right and it did not matter for two decades. The smallpox vaccine existed for a century and a half before smallpox was eradicated. The link between smoking and lung cancer was established long before the instruments that changed exposure were in place.
The quality of the instrument matters. But it does not determine the outcomealone. The capacity of the institution receiving it matters just as much, and sometimes determines whether the instrument produces any public health benefit at all.
The chapters that follow examine artificial intelligence with that lens. What it genuinely offers, where it fails, who bears the cost when it does, and what would have to be true for any of it to reach the people it is meant to serve.
The instruments changed. The question did not.