7 Regulation
How does a society decide what counts as harm, and what should it do about the harms it recognizes?
- Explain why what gets labeled “harmful” and therefore gets regulated does not track a neutral measurement of damage.
- Contrast the rare-dramatic-attributable model of harm with the diffuse-pollution model, and say which one describes social media and AI.
- Distinguish a moral panic from an evidence-backed harm, and say what each implies for regulation.
- Name the instruments by which harm is regulated and the failure mode of capture.
Which is more dangerous: alcohol or cannabis? Most people answer on the basis of what is legal, and the law does not match the evidence. Alcohol kills hundreds of thousands of people a year and sits in supermarkets; cannabis, with a lower measured harm profile, was a criminal offense for most of a century, and its enforcement fell almost entirely on Black and Latino communities. The point is not which substance is worse; it is that the law does not first measure harm and then regulate it. We first decide who and what matters, then find or manufacture the harm that justifies the decision.
7.1 Harm Is Contested, Not Measured
Four cases show how harm is contested rather than measured. Consider sex. What was “private” versus what was “criminal” has been redrawn case by case, in someone’s interests, with no consistent underlying principle. Sexual behavior between consenting adults was criminally regulated well into the twentieth century, and the expansion of privacy was won, not discovered. Sodomy laws are a Victorian legal export, written in 1861 and 1885, and used to convict Oscar Wilde before spreading through colonial law; they are not an ancient universal. Marital rape was legal in most US states until the 1970s and 1980s, because the private sphere was the household, not the person.
Consider drugs. Portugal decriminalized personal possession of all drugs in 2001, pairing the change with treatment and harm reduction; HIV transmission and overdose deaths fell, and use did not rise. Consider guns. After the Columbine shootings, legislators went after violent video games, but the evidence never supported that hypothesis, and cross-national comparisons point to gun availability, inequality, and the history of racially organized violence as the real drivers. Consider privacy. Its historical unit was the household, not the person, and what counts as a protected sphere is won and redrawn, not discovered.
The common thread is that harm is a claim made by the powerful, the panicked, and the organized, not a fact that precedes regulation.
7.2 Two Models of Harm
Engineers are trained on a particular model of failure: the faulty valve that causes a boiler explosion. Harm in this model is rare, dramatic, and attributable. But the worst damage caused by industry has not worked this way. When leaded gasoline lowered the IQs of an entire generation of children, no single tank of fuel caused a measurable injury. No particular cigarette caused any particular cancer death. The harm is real and massive, but it is distributed across millions of small exposures, so that no single decision can be shown to have caused a single injury.
Social media’s harms are this second kind. The term cognitive pollution captures the pattern: radicalization, depression, and polarization are byproducts of engagement maximization, produced at scale and priced at zero. The causal chain is long, probabilistic, and hard to attribute, which is exactly what makes it hard to regulate. The philosopher Derek Parfit identified the philosophical twin of this problem: the non-identity problem. If our choices change who is born, then no particular future person can say we made them worse off, yet the choice still has victims in the aggregate. Harms to “future people” cannot name an individual victim, which is why they are so easy to postpone.
Once a society accepts that harm can be diffuse and cumulative, responsibility shifts from proving individual causation to holding the emitter accountable for the aggregate. Engineers now study pollution not out of conscience but because liability eventually forced them to.
The shift has a legal consequence. If harm is rare, dramatic, and attributable, responsible engineering means avoiding the specific decision that produces the failure. Under the pollution model, responsible engineers are accountable for long-term cumulative effects. A few recent court judgments in the United States suggest this shift is finally happening, but they will be contested, and as the overturn of Roe v. Wade shows, precedent is not a guarantee. Legislation that holds tech companies responsible for the damage their products do to users’ mental health can come sooner, will be more robust, and will have more impact than gestures like banning young people from social media.
The law has mostly not solved the problem of future harm either, which explains why our legal obligations to future people are so weak. Courts require an identifiable party who has suffered a concrete injury. Future generations fail this test on both counts. The workarounds are real but imperfect. Environmental impact assessments require governments to consider long-term consequences. Some constitutions are designed to be difficult to amend. In 1993, the Philippine Supreme Court allowed a group of children to sue the government on behalf of future generations in Oposa v. Factoran. Wales and Hungary have created parliamentary commissioners tasked with representing future interests. These are genuine innovations, but today’s legislators still have no electoral incentive to sacrifice for people who cannot vote.
7.3 The Uncertainty Playbook
When a harm is alleged, the standard industry playbook runs in five steps: fund scientists to generate alternative hypotheses, insist that correlation is not causation, describe any call for regulation as an attack on freedom, create front organizations with neutral-sounding names, and delay—because delay is profit. The response then moves through four phases: denial, uncertainty, the market argument, and finally acceptance, at which point compliance reliably turns out to be cheaper than predicted.
The tobacco industry wrote the script. In 1950, Richard Doll and Austin Bradford Hill published a landmark study linking smoking to lung cancer. The industry’s response was a masterclass in making clear things seem murky: it formed a research committee, funded scientists, and insisted for decades that the evidence was inconclusive. Internal documents later proved the executives knew the risks long before the public did. Within four years, American cigarette manufacturers had formed the Tobacco Industry Research Committee, later renamed the Council for Tobacco Research. They hired scientists, funded studies, placed ads in newspapers, and issued press releases arguing that the evidence was inconclusive. Their goal was not to disprove the science: it was to create enough uncertainty that politicians felt they could not act and smokers felt they could not be sure. The same script ran for leaded gasoline, asbestos, and oxycontin, and it is running now for social media and AI (Oreskes and Conway 2010).
The argument from the market deserves special attention, because it fails in every case where it has been tested. Consumers cannot evaluate a placental crossing rate for a sedative, or the probability that a suspension design will cause a rollover, or the structural integrity of a crumple zone. Markets aggregate preferences for things people can evaluate; they do not reliably handle latent hazards that require technical expertise and longitudinal data to detect.
7.4 When Regulation Works
The history of pollution shows what regulation looks like when it succeeds. In December 1952, a coal-smog blanketed London so thick that people could not see their own feet, killing at least 4,000 people in four days— a figure researchers later revised to closer to 12,000. The government’s first response was to blame influenza, but the Clean Air Act of 1956 followed, and the great smogs did not return. On June 22, 1969, the Cuyahoga River in Cleveland caught fire— the thirteenth time it had done so since 1868— and the resulting photograph helped produce the Clean Water Act of 1972. The Thames had been an open sewer since the Great Stink of 1858, but enforceable sewage-treatment requirements turned it around, and in 1983 a salmon was caught in the river for the first time since the 1820s. In every case, what changed was not public disgust, which had existed for a century, but the legal and economic cost of pollution.
The same lesson can be read backwards. At Minamata Bay in Japan, the Chisso chemical company discharged mercury for decades; cats, which ate fish scraps, sickened first and were ignored as an early warning, and official recognition of the disease came only in 1968, more than a decade after the symptoms appeared. A cross-border disaster can sometimes be converted into enforceable rules: after a 1986 fire at a Sandoz warehouse sent thirty tonnes of pesticides into the Rhine, the Rhine Action Programme set binding pollution targets, and salmon returned to the river within a decade.
The pharmaceutical parallel is the precautionary principle. Thalidomide was approved in West Germany in 1957 and sold in forty-six countries, but Frances Kelsey, a pharmacologist at the US Food and Drug Administration, refused to approve it because the safety evidence was insufficient. The birth defects that followed where it was approved vindicated her, and the Kefauver-Harris Amendment of 1962 followed. Before it, you could sell a drug in the United States without proving it did anything; after it, you had to show that it worked. Where harm is plausible and irreversible, the burden of proof may rest on those introducing the risk.
Product safety tells the same story. Ralph Nader’s 1965 book accused General Motors of selling cars it knew to be dangerous; GM’s response was to hire private detectives to dig up dirt on him, and when the surveillance was exposed, GM’s president had to apologize on national television, and Congress passed the National Traffic and Motor Vehicle Safety Act that same year (Nader 1965). Sweden mandated front seatbelts in new cars in 1959. The evidence for their effectiveness was already solid: Nils Bohlin, an engineer at Volvo, had developed the three-point belt the year before, and Volvo had made the patent freely available to every manufacturer. This was not altruism—Volvo wanted to market its cars as safe—but the effect was the same. The United States did not require seatbelts in new cars until 1968, and individual states did not begin requiring drivers to wear them until the 1980s. Australia’s Victoria became the first jurisdiction anywhere in the world to require seatbelt use in 1970—a decade before most American states. Airbags, first demonstrated as viable technology in the 1950s, were not required in all new American passenger cars until 1998. At every step, car makers argued that the requirement was premature, that consumers would choose safety features if they really wanted them, and that regulation would raise costs and kill innovation. At every step, the data showed substantial reductions in deaths after regulations changed. When buyers compare cars, they mostly look at price, fuel economy, and styling. They do not systematically seek out crash-test ratings. This is not a character flaw; it is a predictable feature of how people make decisions under uncertainty about low-probability events.
The hardest lesson is that a disaster alone does not produce regulation. In 1911, a fire at the Triangle Shirtwaist Factory in New York killed 146 workers, mostly young immigrant women, because the managers had locked the stairwell doors. The owners were acquitted of manslaughter and collected the insurance. What turned the fire into fifty-six pieces of legislation was the fire plus the unions and reform organizations that could turn grief into legislative pressure (Orleck 1995).
7.5 The Ozone Hole That Closed
In 1974, two chemists at the University of California published a paper predicting that chlorofluorocarbons would destroy the ozone layer that shields Earth from ultraviolet radiation. CFCs were used as propellants in aerosol cans and refrigerants in air conditioners, and while the paper’s authors did not yet have a hole to point to, they had atmospheric chemistry on their side.
The chemical industry’s response was to fund counter-research, hire lobbyists, and describe the scientists as alarmists whose work was too speculative to justify regulatory action. This was the same playbook the tobacco industry had been running for two decades, and for a while it worked. The Alliance for Responsible CFC Policy argued that the science was uncertain. Industry representatives testified before Congress that banning CFCs would cost hundreds of thousands of jobs. Du Pont, which held a large share of the CFC market, said in 1975 that it would stop making CFCs only if a worldwide scientific consensus emerged and the appropriate regulatory bodies took action. This was not a promise to act; it was a description of conditions the company presumably believed would never be met.
Eleven years later, in 1985, a team from the British Antarctic Survey reported a massive thinning of the ozone layer over Antarctica every southern spring. Their data was so far outside expected ranges that they initially assumed their instruments were broken. NASA confirmed the finding using satellite data that had been sitting in archived files for years, after automated quality-control software had flagged the anomalous readings as errors.
Industry resistance collapsed in just two years, and the Montreal Protocol was signed in 1987. Its design explains why it worked when so many other environmental agreements have not. It set binding phase-out schedules for ozone-depleting substances, with different timelines for developed and developing countries. It established trade sanctions against non-signatories, which meant that countries outside the agreement faced economic costs for staying out. And it created the Multilateral Fund, which transferred technology and money from wealthy countries to developing ones. This last element gets the least attention and does the most work.
When the protocol was negotiated, China and India were skeptical. Both were industrializing rapidly, both had growing demand for refrigeration and air conditioning, and both pointed out that the damage to the ozone layer had been caused almost entirely by wealthy countries. The demand that they now forgo the same technologies their competitors had used looked like a way of keeping them poor. The Multilateral Fund changed the calculation. By 2023 it had disbursed over $4 billion to help developing countries transition away from CFCs. China became one of the largest recipients of technology transfer funding and one of the most consistent compliers with phase-out schedules. Neither country did this because its leaders suddenly became environmentalists: compliance became economically rational once the fund made alternatives affordable.
Du Pont, which had spent years arguing that alternatives were technically impossible, announced shortly after the protocol was signed that it had developed workable substitutes and would accelerate their commercialization. What had been technically impossible became technically straightforward once the regulatory framework made the old product unmarketable.
The substitutes were hydrofluorocarbons. They did not destroy the ozone layer, but they turned out to be extremely potent greenhouse gases, some thousands of times more warming per molecule than carbon dioxide. The Kigali Amendment, adopted in 2016, added HFCs to the list of controlled substances and set phase-down schedules for them. Climate scientists estimated that full implementation would avoid up to 0.4°C of warming by 2100.
This is not a story about individual consumers making better choices. Millions of people did not read scientific papers and switch to pump-action hairspray. The mechanism was a binding international agreement with differentiated obligations, a technology transfer fund, and trade sanctions against non-participants.
7.6 Settlement Versus Solution
The 1998 Master Settlement Agreement is the cautionary tale. Forty-six states sued the tobacco companies to recover $206 billion in Medicaid costs, and won. But it was a settlement, not a solution: it monetized the harm rather than removing it, and the industry survived and moved into new markets.
The road to that settlement ran through the Surgeon General’s report of 1964, fourteen years after Doll and Hill’s paper, which stated clearly that smoking caused lung cancer and recommended action. The industry spent the next thirty-four years fighting a rear-guard action. What eventually forced a reckoning was a combination of litigation, public health advocacy, investigative journalism, congressional hearings, and the slow accumulation of economic costs borne by state Medicaid programs. The key legal move was a shift in how states argued their cases. Rather than proving that smoking had harmed specific individuals, which the industry could defeat by arguing about individual risk tolerance, states sued to recover the cost of treating sick smokers. Mississippi was first in 1994, and by 1998 forty-six states had reached the Master Settlement Agreement.
It was a victory, but it took forty years to go from a clear scientific finding to a partial legal resolution, and even then the industry survived and moved into new markets. Other countries responded slowly, but they did respond. Australia required plain packaging for all tobacco products in 2012: no logos, no distinctive colors, just the brand name in a standardized font on a drab olive-brown background, surrounded by graphic health warnings. The industry challenged the law in the High Court and lost unanimously. Philip Morris then filed a challenge under an investor protection treaty between Hong Kong and Australia, arguing that plain packaging violated the treaty’s provisions on foreign investors. The tribunal dismissed it in 2015 on the grounds that Philip Morris had restructured its Australian operations specifically to gain access to the treaty—a move so transparently opportunistic that even the arbitrators were not impressed. What eventually made the difference was not better science but organized pressure operating through multiple channels simultaneously: litigation, public health campaigns, investigative journalism, and international organizations like the World Health Organization, whose Framework Convention on Tobacco Control gave national health ministries political cover and legal tools they had previously lacked. None of these were sufficient on their own. It was the combination that worked, albeit slowly. The same limit appears in antitrust. The 1911 breakup of Standard Oil is cited as the definitive remedy, but several successor companies immediately reconsolidated, and John D. Rockefeller’s personal fortune increased after the breakup as the stock prices of the subsidiaries rose (Stoller 2019). The remedy addressed the legal structure of the monopoly without altering the underlying concentration of power.
Some market structures build harm in by design. A tontine was a pooled investment whose subscribers’ shares grew as the others died, creating an obvious incentive to hasten their deaths, and most governments eventually banned tontines for that reason. The platform race to be the last surviving competitor reproduces the same incentive in a structure that is not yet banned.
Capture is the recurring enemy, and it has an international cousin. When Australia required plain packaging for cigarettes, Philip Morris restructured its holdings through a Hong Kong subsidiary to sue under an investor-state dispute settlement clause— a move so transparently opportunistic that even the arbitrators were not impressed. Regulatory capture also has an extreme form. In South Korea, chaebol executives were convicted and then pardoned as too important to the economy to jail. “Too big to jail” is not a technology-industry invention.
Capture also has a technical form. In the savings and loan crisis of the 1980s, control fraud—executives using their control of a company to route contracts to related parties and manipulate reported earnings— was made possible by deregulation that removed oversight and by accounting standards that gave executives wide latitude (Black 2005). The auditors who certified the books were paid by the companies they audited, which gave them a powerful reason not to find problems. The same structure produced Wirecard decades later.
The enforcement machinery can also be theater. The EU’s record fine against Google of €4.34 billion for Android antitrust violations in 2018 amounted to roughly two weeks of Google’s annual profit at the time. Google appealed, had the fine reduced slightly, and continued the practices it had been fined for. A fine that is smaller than the profit from the conduct is not a deterrent; it is a licensing fee.
7.7 Moral Panics and Wrong Diagnoses
A moral panic is a disproportionate, often exaggerated public alarm that can coexist with a real harm— and often targets the wrong harm while leaving the real one unaddressed. The lineage runs through dime novels in the 1880s, Fredric Wertham’s Seduction of the Innocent against comics in the 1950s (Wertham 1954), rock music, and Dungeons and Dragons, before arriving at violent video games in the 1990s. Wertham’s testimony brought comic books before a Senate subcommittee in 1954 and led directly to the Comics Code Authority, a self-regulatory body that banned dark content so comprehensively that it effectively gutted the medium for a generation. Video games arrived in force in 1993, when a Senate subcommittee held hearings focused on two games: Mortal Kombat, which allowed players to rip an opponent’s spine out as a finishing move, and Night Trap, which featured vampires draining blood from actresses. The hearings generated excellent television and led directly to the creation of the Entertainment Software Rating Board, the voluntary age-based rating system still in use today.
The video-game panic is the clearest modern example. Dave Grossman offered a more careful version of the alarm. His claim rested on military training research: fewer than a quarter of soldiers in World War II actually fired their weapons in combat, so the Army redesigned its training using operant conditioning with human-silhouette targets to raise that rate in Korea and Vietnam. Violent video games applied the same conditioning to children without consent or any ethical framework. Unlike the public panic, his argument was internally coherent. Unfortunately, the scientific literature did not validate it.
After Columbine, legislators went after games, but the scientific literature never supported the hypothesis: laboratory measures of aggression bear little relationship to real-world violence, and large longitudinal studies consistently find no meaningful effect (Ferguson 2015). The simplest check is cross-national: Japan and South Korea are among the highest per-capita consumers of video games in the world and have dramatically lower rates of violent crime than the United States. Gun availability, inequality, and the history of racially organized violence explain American violence far better than games do, but they are much harder to face.
In 2011 the Supreme Court settled the legal question, if not the empirical one. California had enacted a law restricting the sale of violent video games to minors. The Court struck it down in Brown v. Entertainment Merchants Association, applying the same First Amendment analysis it would apply to books or films. The majority noted that the research California presented did not establish a causal link between violent games and harm to minors, and that the burden of proof for restricting expression falls on those who seek restriction.
By then the panic was already fading, displaced by fresh anxieties about social media and smartphones. What had not faded were the harms the panic had never addressed. While legislators debated spine-ripping finishing moves, the real harm went largely unregulated: loot boxes, randomized rewards purchased with real money, marketed to children, and now recognized as gambling by several European regulators.
7.8 The Bridge
Social-media and AI harms are pollution-model harms. The trap is to respond with moral-panic framing—“ban the kids’ phones”— instead of diagnosing the actual harm: attention extraction, addictive design, discrimination, and surveillance. Regulating well means diagnosing the real harm rather than the one that generates the best testimony, using the pollution and product-safety models where they fit, and defending the regulator from capture. The European Union’s Digital Markets Act, which forces designated gatekeepers to allow interoperability and forbids self-preferencing, and India’s Unified Payments Interface, which built public payment rails before any private platform could achieve lock-in, are two early examples of regulation aimed at the actual source of the harm. Regulation has also moved faster than the industry claims is possible. The EU AI Act requires transparency for high-risk systems, mandates human oversight for consequential automated decisions, and bans specific applications outright. Before the GDPR took effect in 2018, industry associations described it as unworkable and predicted that it would destroy European tech competitiveness. By 2024 it had generated approximately $4 billion in fines and had changed how companies worldwide handle personal data, including companies with no European operations that simply chose to comply rather than maintain two systems. The argument that AI regulation will destroy innovation has been made about every major technology regulation in living memory, and has been wrong every time. The question is not whether the harms can be named; it is whether the people who benefit from them can be made to pay.
Harm is not a fact that precedes regulation; it is a claim made by the powerful, the panicked, and the organized. Regulating well means diagnosing the actual harm rather than the one that generates the best testimony, using the pollution and product-safety models where they fit, and defending the regulator from capture.
7.9 Exercises
- Rank the harms, then the laws.
- Consider the list “alcohol, cannabis, heroin, gambling, social media, AI-generated misinformation”. Rank the items by measured harm, then by how the law actually treats them. Discuss the gaps and what explains them.
- Two models of harm.
- Pick a harm (e.g., a bridge collapse, leaded gasoline, a data breach, or teen anxiety from Instagram). Classify it as rare-dramatic-attributable or diffuse-cumulative, and say which regulatory instrument each model naturally suggests. Which model were you trained to think in?
- Moral panic vs. real harm.
- Choose a concern (violent video games, kids and phones, “AI will end humanity”). List the features that make it look like a moral panic and the features that make it an evidence-backed harm. What is the actual harm the panic may be distracting from?
- Precautionary vs. wait-and-see.
- Replay the thalidomide decision: you are the regulator with evidence that is suggestive but not conclusive. Do you approve or wait? What does your answer imply for “move fast and break things” and for releasing open-weight AI models?
- Redesign the law.
- Pick a harm currently regulated on a moral or panic basis rather than an evidence basis (e.g., cannabis, a social-media feature). Propose a regulation that tracks the actual harm, and name the political obstacles to passing it.
- Beat the playbook.
- You are the regulator and the industry is running the five-step uncertainty playbook. For each step, name one counter-move (independent funding, burden-of-proof rules, naming the front groups, deadlines).
- Settlement or solution.
- Study the 1998 Master Settlement Agreement. Why did a $206 billion settlement fail to end the tobacco problem? What would a solution have required, and why was a settlement the outcome the parties preferred?