Manufacturing irresponsibility
A modus operandi from tobacco to AI
I use AI incessantly. I use it to think through problems, to structure arguments, to research things I don't have time to research properly, to draft, to revise, to catch what I missed. I am, by any measure, more productive than I have ever been, and I am more productive in ways that would not be available to me if I had only myself to rely on. There is only so much I know. There is only so much I can take in. AI expands that range, and I would be lying if I said I wanted to give it up.
I also don't fully trust it. Not the tool—the tool is what it is—but what it's doing to me while I use it. I don't know yet which muscles are atrophying, because I'm using the tool to avoid noticing. Neither does anyone else. We're all inside the experiment, and the people running it have no incentive to publish the results.
That's the position from which I want to write about AI and responsibility: inside the thing, using it, grateful for it, suspicious of it. Not outside, not above, not exempt. If the argument I'm about to make is right, it applies to me first.
Here is the argument. The past sixty years of American business have produced a recognizable pattern, and AI is its latest instance. The pattern has three moves. First, engineer the environment so that the convenient choice becomes the default—the path of least resistance, made into a product. Second, profit from the resulting behavior. Third, when the costs of that behavior become visible, relocate responsibility onto the individual. You chose this. You should have known better. It's your body, your attention, your life.
Tobacco wrote the playbook. Pharma refined it. The food industry industrialized it. The attention economy moved it from the body to the mind. And now AI offers to do something none of its predecessors could: it doesn't just degrade your health, your appetite, or your focus. It offers to take over the very faculty—judgment, deliberation, the ability to respond—that responsibility depends on. That's not a difference of degree. It's a difference of kind, and it's why the pattern matters more now than it did when the product was a cigarette.
But before we get to AI, we have to see the pattern clearly. Which means starting where it was written.
The template
The tobacco industry didn't invent doubt. It industrialized it.
By the mid-1950s, the science linking cigarettes to lung cancer was no longer speculative. Richard Doll and Austin Bradford Hill had published their landmark studies in 1950 and 1952. By 1954, the American Cancer Society had found that smokers died of lung cancer at rates far higher than non-smokers. The industry's own internal research—buried for decades, later released through litigation—confirmed what the public studies suggested. The people selling the product knew what it did.
They also knew what to do about it.
In 1953, the heads of the major tobacco companies met at the Plaza Hotel in New York. What came out of that meeting was an organization called the Tobacco Industry Research Committee, later the Council for Tobacco Research—a body whose purpose was not to study tobacco but to manufacture the appearance of controversy about it. The logic was simple and, in retrospect, almost elegant: if the science is settled, the product is indefensible. If the science is contested, the product is a matter of opinion. And if it's a matter of opinion, then the person who smokes has made a choice.
That was the first move: engineer the environment so the convenient choice stays available. Not by hiding the evidence—by drowning it. The industry funded sympathetic researchers, published its own journals, placed op-eds, and cultivated the language of 'balance'. Every news segment that gave 'both sides' equal time was doing the industry's work for it. The goal was never to prove cigarettes were safe. It was to make sure no one could say, with the authority of consensus, that they weren't.
The second move was the obvious one. Cigarettes were sold, aggressively, to everyone—including, for a long stretch, to soldiers in wartime rations and to teenagers through cartoon mascots and glossy campaigns. The profit was extracted at scale, for decades, with full knowledge of the cost. That's not a conspiracy theory; it's in the documents. The industry's own files show executives discussing the health risks internally while denying them publicly. The gap between what they knew and what they said is the space in which the whole operation lived.
Then came the third move, the one that matters most for our purposes. When the costs became impossible to hide—when the surgeon general's report landed in 1964, when the lawsuits began, when the death toll could no longer be euphemized—the industry didn't defend the product. It defended the choice. Smoking was framed as a personal decision, a matter of individual liberty, something adults did with full information and the government had no business interfering with. The harm wasn't the product's fault. It was the smoker's. You should have known. You could have quit. It's your body.
This is the template. Not just 'sell a harmful product'. Plenty of businesses do that. The template is the three-part structure: shape the environment so the convenient choice is the default, extract profit from the resulting behavior, and when the bill comes due, hand it to the individual. The genius of the pattern is that it converts a systemic harm into a personal failing—and in doing so, it makes the system invisible.
And it worked. For decades, the tobacco industry wasn't just profitable; it was respectable. Its executives testified before Congress. Its ads ran in magazines. Its position was treated as one legitimate side of a legitimate debate. The people who paid the price were told—by the industry, by the culture, by their own doctors in some cases—that they had done it to themselves.
The template has been run again and again since. What changes is the product. What stays the same is the move.
The maturation
Tobacco taught American business that you could sell a harmful product for decades if you were willing to manage the appearance of controversy. Pharma learned the lesson and improved on it. The difference is that tobacco fought the institutions that might have stopped it. Pharma learned to become them.
The clearest case is opioids. In the late 1990s, Purdue Pharma introduced OxyContin with a claim that was, on its face, remarkable: that the drug's time-release mechanism made it less addictive than other opioids. This wasn't a fringe position—it was the core of the marketing. Purdue trained its sales force to tell doctors that the risk of addiction was 'less than one percent', a figure that had no reliable scientific basis and that the company's own internal documents contradicted. The sales force was paid on volume. The more prescriptions, the more money, and the more money, the more pressure to expand the market.
But the interesting part isn't the lie. It's the infrastructure Purdue built around it. The company funded patient advocacy groups that pushed for broader access to pain medication. It sponsored continuing medical education courses—the courses doctors are required to take to maintain their licenses—that taught the company's own position on addiction risk. It placed friendly researchers on guideline panels. It cultivated a network of key opinion leaders, respected physicians whose endorsements carried weight with their peers, and paid them to speak. By the time the prescribing guidelines were written, the guidelines had been written by the company's allies.
This is the maturation. Tobacco's strategy was to contest the science. Pharma's strategy was to produce the science, then distribute it through the institutions that doctors trust. The doubt was no longer manufactured in a public relations office. It was built into the medical literature, the continuing education system, the professional guidelines, and the patient advocacy groups. A doctor prescribing OxyContin in 2001 wasn't making a reckless choice. They were following the standard of care—a standard that had been shaped, quietly, by the company selling the drug.
And when the bill came due—when addiction rates rose, when overdose deaths climbed into the hundreds of thousands, when the pills ran out and people turned to heroin and then fentanyl—the third move kicked in again. The narrative wasn't that Purdue had engineered the crisis. It was that patients had abused the medication. That doctors had overprescribed. That individuals had made bad choices. The Sackler family, whose name had been on museum wings and university buildings, was recast as a philanthropic dynasty undone by a few bad actors. The system was exonerated by being dissolved into a million private failings.
The pattern held. Engineer the environment—not just the ads, but the professional standards themselves. Profit from the behavior. Relocate responsibility to the individual. What changed was the sophistication. Pharma didn't need to convince the public that its product was safe. It only needed to convince the people whose job was to evaluate the product. And it did, by hiring them, funding them, and shaping the institutions they worked in.
There's a version of this story that's about one company, or one family, or one drug. But the pattern is bigger than Purdue. It's the structure of an industry that has learned to internalize its own oversight. The FDA, the medical journals, the continuing education system, the clinical guidelines—these are supposed to be the immune system. The pathology is that the pathogen has learned to live inside them.
Which brings us to the third case, and the one that takes the pattern furthest before AI: an industry that didn't just shape the environment of choice, but built it.
The industrialization
Tobacco wrote the template. Pharma refined it. Big Food did something different: it didn't just manage the perception of the product, or capture the institutions that evaluated it. It restructured the environment in which the choice was made.
The tobacco lineage
The connection between Big Food and Big Tobacco isn't a metaphor. It's a bloodline.
Philip Morris and R.J. Reynolds built multi-billion-dollar food businesses between the 1980s and the mid-2000s, spanning seventy-six countries. They transplanted the cigarette playbook directly onto food. 'King-size' became supersized sodas and jumbo cookies. 'Low tar' became 'low fat' and 'light' processed foods—same move, different aisle: give the consumer the feeling of a healthier choice while keeping the addictive core intact. The food scientists called the product development 'engineering', designing for bliss points and vanishing caloric density—foods your brain registers as barely eaten.
This isn't inference. Historians have documented the parallel science strategies: fund research that manufactures uncertainty, attack unfavorable findings, redirect attention away from the product, and treat the absence of consensus as if consensus were impossible. The Sugar Research Foundation launched a project in 1969 whose early results pointed toward sucrose as a possible factor in heart disease and bladder cancer—then terminated the funding, unpublished, undisclosed. In 1967, the sugar industry secretly funded a New England Journal of Medicine review that pointed the finger at fat and cholesterol while minimizing sucrose. (There's legitimate scholarly debate about how far to read the 'sugar conspiracy', but the pattern—industry funding science to serve commercial ends—is well established.)
The environment itself becomes the product
But here's where Big Food goes further than either predecessor. Tobacco managed perception of the product. Pharma managed the institutions that evaluated the product. Big Food managed the conditions of choice itself.
Federal agricultural policy over the past half-century has systematically pushed American farmland toward corn and soy—two crops that now cover nearly two-thirds of U.S. cropland and supply nearly half the average American's daily calories. In the 1970s, Agriculture Secretary Earl Butz told farmers to plant 'fencerow to fencerow', and the subsidy structure rewarded them for doing it. The result is a food system in which the cheapest calories are the engineered ones, and the raw ingredients for processed food are effectively underwritten by the government.
Then the environment is furnished. School lunch programs. Hospital cafeterias. Vending machines in public buildings. SNAP-authorized retailers. The default settings of daily life—the places where people eat without deliberating, because not deliberating is the point—are stocked with the products the system is designed to produce. You don't choose the processed option because you were persuaded to. You choose it because it's there, it's cheap, and the alternative takes effort you don't have at 6 p.m. on a Tuesday.
This is the escalation. Tobacco had to fight the environment; pharma had to infiltrate the environment; Big Food built the environment. The convenient choice isn't marketed into prominence—it's engineered into the infrastructure. The path of least resistance isn't a tendency anymore. It's a subsidy, a supply chain, and a shelf placement.
The blame machine
And when the costs show up—obesity, diabetes, metabolic disease—the third move arrives on schedule. The story isn't that the environment was built to produce this outcome. It's that people make bad choices. They should eat better. They should exercise more. They should have more discipline. Personal responsibility, again, doing the work that structural critique would otherwise do.
The irony is that this is the case where personal responsibility is most obviously a dodge. You can avoid cigarettes by not buying them. You can, with effort, avoid opioids. But you cannot personally opt out of a food system that has restructured the cost, availability, and default settings of what you eat. You can make better choices at the margins. You cannot exit the environment. And the industry knows this—which is precisely why it frames the problem as individual, because the individual frame is the one that lets the system stay standing.
The pattern holds: engineer the environment, profit from the behavior, relocate responsibility. What changed is the depth. Tobacco blurred the truth. Pharma captured it. Big Food made it irrelevant.
Which brings us to the last step before AI. If food restructured the physical environment of choice, the next industry restructured the attentional environment—and in doing so, moved the pattern from the body to the mind.
The bridge: The attention economy's antechamber
If Big Food restructured the physical environment of choice, the attention economy restructured the first door through which anything enters your awareness.
The logic of that door is simple: attention is finite, information is infinite. Whoever controls how your attention is allocated controls the conditions under which you think about anything—and, more importantly, whether you still have the conditions to think at all.
From 'what you choose' to 'whether you can choose'
The attention economy isn't in the business of making you prefer a particular product. It's in the business of designing an environment in which you are continuously unable to process anything deeply.
Autoplay. Infinite scroll. The instant feedback of the like. These aren't content features; they're infrastructure. Their common trait is that each interaction is shorter than the time it would take you to form a complete judgment. Before you've judged, the next stimulus has arrived. Nicholas Carr described the shift precisely in The Shallows: reading gives way to browsing and scanning, and the medium stops being a means to an end and becomes the end itself.
This isn't about what you consumed. It's about how you consumed—your cognitive mode was reshaped.
The 'thinking muscles' that get reshaped
The mechanism is neuroplasticity run in reverse. The brain strengthens what it uses and prunes what it doesn't. The attention economy continuously calls on your capacity to handle high-frequency, fragmented, instant-feedback stimuli, while leaving your capacity for sustained focus, delayed judgment, and depth unused—and therefore, over time, less available.
This is where the pattern from the earlier cases reappears, but with a twist. Tobacco degraded the body's signals. Big Food degraded the environment of choice. The attention economy degrades the substrate of deliberation itself—not by taking your attention, but by making the kind of attention that thinking requires harder to sustain.
The third move, again
And when the costs become visible—fragmented attention, declining reading, the sense that nobody can concentrate anymore—the blame arrives on schedule. It's your phone. It's your discipline. It's your screen time. Manage it better. The environment stays standing; the individual absorbs the fault.
Which brings us to AI. Because AI does something none of the previous cases did. Tobacco, food, and the attention economy all degraded the conditions surrounding thought. AI offers to substitute for thought itself—and in doing so, it doesn't just erode the ability to think, learn, and therefore respond. It shapes the very boundary of what 'responsibility' can mean for the person using it.
That's the culmination. And it's where the argument has to go next.
The culmination
AI runs the same three moves. It engineers the environment—not through ads or subsidies or autoplay, but by inserting itself into the moment of work, the moment of decision, the moment of thought. It profits from the resulting behavior. And when the costs appear, the costs will be assigned to the individual. You relied on it too much. You stopped thinking for yourself. You should have known better.
But AI is not just the latest entry in the series. It is the first case where the thing being degraded is the very faculty that responsibility depends on.
What the earlier cases left intact
Tobacco degraded the body's signals. You could still think; you just couldn't trust what your body told you. Big Food restructured the environment of choice. You could still deliberate; the deliberative option was just made costly and scarce. The attention economy fragmented the substrate of deliberation. You could still, with effort, sustain a thought; the effort required was just raised, continuously, against your will.
In each case, something around the capacity to respond was damaged. The capacity itself remained—bruised, undermined, harder to exercise, but present. The person could still, in principle, know what was happening to them and answer for it. That's what made the third move—blaming the individual—even minimally coherent. You could have chosen otherwise. It would have cost you, but it was possible.
AI is the first case that offers to do the responding for you.
The offer
This is not a hypothetical. It's the pitch. The industry does not sell AI as a tool for thinking. It sells AI as a substitute for thinking—an agent that will handle your email, draft your memo, plan your trip, summarize the report, structure the argument, generate the options, and increasingly make the call. The promise is delegation. The promise is that you no longer have to.
And here is where the earlier analogy breaks, in a way that matters. When you use a calculator, you offload arithmetic but keep the mathematical judgment—whether the answer is plausible, what operation the problem requires, whether the result makes sense. When you use AI to think, you offload the judgment itself. Not the arithmetic of reasoning, but the reasoning—the weighing, the evaluating, the deciding what matters. And judgment is not a separate faculty you can keep while outsourcing the work. It is constituted by the work. You do not have judgment and then apply it; you build it by exercising it, and you lose it by not.
The knowledge condition
This is where the argument I've been building comes due. Responsibility, in the strict sense, is not a capacity you simply possess. It is a relation between you and what you understand. You can only respond to what you can see; you can only answer for what you know. Responsibility is the ability to respond, and that ability is a function of knowledge. Without knowledge, a being is not in a position to answer for anything—not because they're lazy or immoral, but because the condition of answerability hasn't been met.
So consider what happens when AI not only does the work but shapes the knowledge that would let you evaluate the work. The model summarizes the document—but you didn't read the document, so you can't tell whether the summary is faithful. The model drafts the argument—but you didn't build the argument, so you can't tell whether it holds. The model generates the options—but you didn't frame the problem, so you can't tell whether the options are the right ones. At each step, the knowledge that would make you a responder is the knowledge the tool is quietly replacing. The more fluent you become with the tool, the less you know about the thing the tool is doing—and the less you are able to respond to it.
This is not an argument against using AI. It's an argument about what using AI costs, and about who is positioned to bear that cost. For someone who already has deep knowledge in a domain, AI is a force multiplier on judgment they already possess. They can evaluate the output because they know the territory. For someone who doesn't, AI is a substitute for judgment they never built—and the substitution prevents the building. The gap between these two users is not a gap in tool fluency. It's a gap in the knowledge that makes responsibility possible at all.
The boundary of responsibility
And this is the sense in which AI doesn't just erode responsibility—it redraws its boundary. If responsibility requires knowledge, and the knowledge is increasingly produced, filtered, and mediated by a system whose workings you don't understand, then the boundary of what you can be responsible for is being set by the system, not by you. The industry's third move—blame the individual—becomes not just a dodge but a category error. You can't be responsible for what you had no knowledge of. And if the environment is engineered to prevent that knowledge from forming, the individual isn't failing responsibility. They're being placed outside it.
Which means the real question isn't 'will people take responsibility for how they use AI?' It's 'what would have to be true for them to be in a position to?' And that's an environmental and institutional question, not a personal one.
Where I stand
I use AI incessantly. I'm more productive than I've ever been, and I attempt projects I wouldn't attempt alone, because there's only so much I know and so much I can take in. I'm not writing this from outside the pattern. I'm writing it from inside, using the tool, grateful for it, and trying to stay on the right side of the line I'm describing.
The line, as best I can draw it, is this: I use AI to extend knowledge I have, not to replace knowledge I don't. I use it to draft what I already understand, not to understand what I haven't. I keep the judgment in my hands—which means I have to keep doing the work that builds judgment, even when the tool would happily do it for me. It costs me time and effort that the tool would save. That cost is the point. It's what keeps me a responder rather than a passenger.
But I don't fully trust that I'll stay on the right side of the line. Nobody does. We're all inside the experiment, and the people running it have no incentive to publish the results.
The line
The arc from tobacco to AI is not a story about industries getting worse. It's a story about the target moving inward.
Tobacco worked on the body—it degraded the signals that tell you something is wrong. Big Food worked on the environment—it restructured the cost and availability of the choice, so that the convenient option became the default. The attention economy worked on the mind's front door—it fragmented the substrate of deliberation, making sustained thought harder to sustain. And AI works on the condition of response itself—it offers to take over the faculty that responsibility depends on, and in doing so, it redraws the boundary of what any of us can be answerable for.
The three moves never changed. Engineer the environment. Profit from the behavior. Relocate the blame. What changed is what's being engineered, and what's being lost. The earlier industries damaged the conditions around thought. AI is the first that substitutes for thought—and a substituted faculty is not a bruised one. It's a faculty you never built.
So where does that leave someone who uses the tool? Not outside the pattern. I've said where I stand—the line I try to hold—and I won't restate it. What I want to add is that the line isn't only personal. It's also a distinction worth naming, because it separates two things the industry wants us to blur.
There is a difference between responding to the path of least resistance and manufacturing it. Every market does the first. I price my books to a market of readers who don't want to deliberate about every purchase, and I don't think that makes me complicit in anything—it makes me attentive to how people actually choose, and respectful of their time. What the industries in this essay did is the second thing. They didn't respond to a tendency; they built an environment designed to produce it, and then billed the resulting behavior to the people inside it. The first is commerce. The second is manufacture. And the difference between them is the difference between a market and a machine.
That's the distinction the title is about. 'Manufacturing irresponsibility' is not a complaint that people are irresponsible. It's an observation that irresponsibility can be produced—engineered, like anything else—and that when it is, the producers have a strong interest in making sure the blame lands on the individuals they produced it in. The tobacco executive who says 'smoking is a choice' and the AI executive who says 'use it responsibly' are running the same play. Both know the environment does the work. Both need you not to know that.
So the question isn't whether you'll use AI. I use it. You probably will too. The question is whether you'll use it while knowing what it's doing—to the work, to the knowledge that makes the work valuable, and to the faculty that makes you answerable for it. That's all responsibility has ever meant in the strict sense: not abstinence, not purity, but staying in a position to respond. The precondition of response is knowledge. Whoever controls the knowledge controls the boundary of what you can be blamed for. And that boundary is where the fight over AI is actually being fought—not over whether the tool works, but over who gets to define what you can see.
I wrote this essay with AI. Every stage: the thinking, the structuring, the research, the drafting, the revision. I'm telling you because the essay makes no sense if I hide it, and because the fact that I used it doesn't refute the argument—it is the argument, in miniature. The difference between this use and the manufactured-irresponsibility case is not that I refused the tool. It's that I kept knowing what the tool was doing, and kept paying the cost of staying on the right side of a line I can't promise I'll always hold.
