We've done the asbestos episodes, the lead paint episodes — the stuff our grandparents thought was magic and turned out to be poison. Today, we're doing the version where we're the grandparents. Daniel wrote in with a four-part question that's been rattling around my head since I read it. He points out that every generation has its miracle materials and miracle practices — things hailed as cutting-edge and safe, only to be revealed later as catastrophically dangerous. Our grandparents had doctors recommending cigarettes. The generation before them put radium in drinking water and called it health tonic. And Daniel's question is: if we take the long view, the uncomfortable mirror — what are WE getting wrong right now? What advances today, if you had to place a bet, will history remember not as miracles but as mistakes? And he wants specifics — not vague hand-waving, but actual candidates. So today, we're not looking backward with smugness. We're looking forward with humility — and trying to spot the asbestos of 2026 before it's in our walls.
The thing that makes this question so difficult — and so worth doing — is that the pattern isn't about ignorance. That's the first misconception to clear out. We tell ourselves a comforting story: past generations just didn't know better. They lacked our science, our testing protocols, our regulatory frameworks. But when you dig into the actual history, that story falls apart.
The companies knew.
The companies knew. 3M's internal documents from the nineteen seventies showed they were fully aware that PFAS accumulated in human blood. They had the data. They kept producing it for decades anyway. Lead manufacturers knew about lead poisoning in children and ran ad campaigns telling parents to put lead paint in nurseries. The tobacco industry had the cancer research and buried it while marketing to teenagers. This wasn't ignorance. It was — what's the word — a structural unwillingness to connect the profit to the cost.
And the cost is always on a thirty-year timer. That's the mechanism. The benefit is immediate — your house doesn't burn down because the asbestos insulation works, your omelette doesn't stick because the Teflon is flawless, your anxiety quiets because the cigarette hits your bloodstream in seven seconds. The harm is probabilistic, diffuse, and arrives long after the people who made the decision have moved on. Our brains are not wired to weigh that tradeoff correctly.
There's a name for this in the behavioral economics literature — hyperbolic discounting. We radically undervalue future costs compared to present benefits. But I think it's actually worse than the lab experiments suggest, because in the real world, the future costs aren't just discounted — they're invisible. Nobody was tracking mesothelioma clusters in the nineteen fifties. The data didn't exist. The question wasn't being asked.
Which is Daniel's point, really. What aren't we asking? So let's start with the ones we can already see the warning signs on — the chemicals that are already in our blood.
PFAS. Per- and polyfluoroalkyl substances. The forever chemicals. And I want to be precise about why they're called that, because the chemistry is the whole story. Carbon-fluorine bonds are among the strongest bonds in organic chemistry — four hundred eighty-five kilojoules per mole, if I'm remembering right. For comparison, a carbon-carbon bond is about three hundred fifty. What that means in practice is that these molecules don't break down. Not in water, not in soil, not in sunlight, not in the human body. Once you make them, they exist effectively forever.
Which was the selling point.
That's the part that should keep us up at night. PFAS were celebrated because they were indestructible. Non-stick cookware, waterproof clothing, firefighting foam, food packaging — the whole pitch was that these coatings would survive heat, water, grease, and time. And they do. They survive everything. Including us.
So we're all carrying them.
We're all carrying them. The CDC has found PFAS in the blood of virtually every American they've tested — something like ninety-seven percent of the population. They cross the placenta. They accumulate in the liver and kidneys. They're associated with kidney cancer, testicular cancer, thyroid disease, immune suppression, and developmental problems in children. The EPA finally proposed enforceable drinking water limits in twenty twenty-three — parts per trillion levels, which tells you how potent these things are. But here's what gets me: we've been producing them at industrial scale since the nineteen forties. That's eighty years of accumulation in the water table, in the soil, in every living thing on the planet. And there's no cleanup technology that can remove them at scale. We can filter drinking water at the tap, but you can't filter the ocean.
Wait. Parts per trillion? That's the detectable threshold?
No, that's the harm threshold. The EPA's proposed limit for PFOA and PFOS — the two most studied compounds — is four parts per trillion. That's roughly one drop in twenty Olympic swimming pools. And they're finding concentrations above that in drinking water systems serving millions of people. The health effects show up at levels so low they're hard to even conceptualize.
So the property that made them useful — indestructibility — is exactly the property that makes them dangerous. That's the pattern, isn't it? We celebrated what we should have feared.
And we're still producing them. Not PFOA specifically — that was phased out under pressure — but the replacement compounds, the GenX chemicals and their relatives, are just different arrangements of the same carbon-fluorine backbone. The industry strategy has been to swap one PFAS for another and argue the new one hasn't been proven harmful yet. Which is technically true, because proving harm takes decades of epidemiological data, and the new compounds have only been in use for ten or fifteen years. It's a regulatory shell game.
The asbestos playbook. Asbestos manufacturers did the same thing — when chrysotile got regulated, they pushed amosite. When amosite got regulated, they pushed crocidolite. The mineral changed, the mesothelioma didn't.
And the lawsuits are following the same arc. 3M and DuPont are facing thousands of class-action suits right now. The DuPont case in West Virginia — the one that became the movie Dark Waters — that was about PFOA specifically, one chemical in one town. Now multiply that by every firefighting training site, every airport, every factory that used these compounds. The liability is... I don't think we've seen the full scope yet.
Microplastics feel like the same story told in a different key.
They do, and the mechanism is different enough to be worth separating. PFAS is a chemical problem — specific molecules with specific bonds. Microplastics is a materials problem. Plastic doesn't biodegrade; it photodegrades and mechanically degrades. Sunlight and friction break it into smaller and smaller pieces, but the polymer chains remain intact. A plastic bag doesn't disappear — it becomes a trillion microscopic plastic particles.
And we're finding them everywhere.
Everywhere. Human blood, lungs, liver, kidneys, placentas, breast milk. A twenty twenty-five study — this one stopped me cold — found microplastics in one hundred percent of human testicles examined. Every single sample. And they found a correlation between higher concentrations and lower sperm counts. We're not talking about a marginal finding. Global sperm counts have dropped by more than fifty percent since the nineteen seventies. The microplastics hypothesis isn't the only explanation, but it's gaining weight fast.
The credit card fact.
We're each ingesting roughly the weight of a credit card in plastic every week — about five grams. It's in the water, it's in the food, it's in the air. Synthetic clothing sheds microfibers every time you wash it. Car tires shed microplastics every time you drive. The dust in your house is partly plastic. And once it's in the body, two things happen. The particles themselves can cause physical inflammation — your immune system treats them as foreign invaders and mounts a chronic low-grade response. And then there's the chemical leaching. Many plastics contain additives like bisphenols and phthalates, which are known endocrine disruptors. They mimic hormones, they interfere with development, they've been linked to obesity, diabetes, reproductive problems.
And again, the property we wanted was the property that hurts us. Plastic is durable, cheap, and inert — until it's inside a living organism, at which point inert becomes a lie.
Plastic production has doubled since the year two thousand. Half of all plastic ever made was produced in the last twenty years. We're not slowing down. And recycling — I need to say this clearly — recycling is not solving this. Globally, about nine percent of plastic ever produced has been recycled. The rest is in landfills, in the ocean, or still in use. The ocean plastic will keep breaking down into microplastics for centuries. We've essentially seeded the entire planetary ecosystem with a material that doesn't go away.
Herman, you were a pediatrician. When did you first hear about microplastics in human tissue?
Honestly? Maybe five or six years ago. And even then it was presented as an interesting finding, not a clinical concern. The medical community hasn't caught up. There are no diagnostic codes for microplastic exposure. No treatment protocols. No screening recommendations. We're in the asbestos nineteen fifties phase — the evidence is accumulating, the mechanisms are plausible, but the clinical response hasn't materialized. And the scale of exposure means that even if we stopped all plastic production tomorrow, the burden is already inside us.
That's the part that I keep circling back to. With asbestos, you could theoretically remove it from buildings. With lead, you could replace the pipes. But PFAS and microplastics are globally distributed in a way that makes remediation essentially impossible. We can't filter the planet.
And we're still in the production phase. We haven't even reached peak exposure. The next generation will be born with a higher body burden than we have, because their mothers have been accumulating these things for decades. It's intergenerational.
So those are the physical risks — molecules in our bodies, particles in our tissues. But Daniel's question pushes further than that. He's asking about advances, not just materials. And the next category is harder to measure because it's in our minds.
Social media. And I want to approach this carefully, because the conversation around social media and mental health has become so polarized that it's hard to find the signal. But the signal is getting clearer. The Surgeon General issued an advisory in twenty twenty-three calling the youth mental health crisis unprecedented and directly linked to social media adoption patterns. The data show dose-response relationships — more hours on TikTok, Instagram, Snapchat correlate with higher rates of depression, anxiety, and suicidal ideation in teenagers. This isn't a marginal effect. The CDC's Youth Risk Behavior Survey found that nearly sixty percent of teen girls reported persistent feelings of sadness or hopelessness in twenty twenty-three, up from the mid-thirties a decade earlier.
And the counterargument is always correlation versus causation — maybe sad teens use more social media, not the reverse.
Right, and that's a legitimate methodological concern. But the experimental evidence is accumulating. Studies that randomly assign people to reduce or eliminate social media use show improvements in well-being within weeks. The internal Facebook documents leaked in twenty twenty-one showed the company's own research found that Instagram made body image issues worse for one in three teen girls, and that teens blamed Instagram for increases in anxiety and depression. The company knew. They deprioritized the fixes.
There's the pattern again. The company knew.
The mechanism here isn't chemical — it's psychological. Algorithmic amplification of extreme content, because extreme content drives engagement. Social comparison at a scale and intensity that human brains never evolved to handle — you're not comparing yourself to thirty kids in your village, you're comparing yourself to eight billion people's curated highlight reels. Sleep displacement — teens are staying up past midnight scrolling, and sleep deprivation is a known causal factor in depression. Attention fragmentation — the constant switching between apps and notifications trains the brain out of sustained focus.
The parallel to smoking keeps getting stronger. An industry that knew its product was harmful to young people, suppressed the evidence, and marketed aggressively to children. The tobacco playbook.
And the harm is distributed across time and population in a way that makes it hard to pin down causally. No single hour on Instagram causes depression. It's the cumulative effect of years of exposure during critical developmental windows. By the time the harm is measurable, the exposure is already in the past. Same structure as asbestos — the mesothelioma takes thirty years, the depression takes five, but the logic is identical.
What's the counterargument from the platforms? I'm genuinely asking — how do they defend this?
A few ways. One is that they're connecting people and reducing loneliness — which is true for some users, some of the time. The problem is that the net effect, at the population level, appears negative for adolescents. Another argument is that they're just a mirror — society has problems, social media reflects them. But that ignores the algorithmic amplification piece. A mirror doesn't select for the most emotionally triggering content and push it to the top of everyone's feed. And the third argument is personal responsibility — parents should monitor, teens should self-regulate. Which is the same argument the tobacco industry made about teenagers and cigarettes in the nineteen sixties.
And now AI.
And now AI. This is the one where I feel the most epistemic humility, honestly, because we're so early. But the early warning signs are... let me put it this way. We are deploying large language models into medicine, law, education, hiring, criminal justice — high-stakes domains where errors have real consequences — at a speed that has no historical precedent. And we're doing it before we understand the failure modes.
The twenty-eight percent number.
A twenty twenty-six study — this is recent — found that GPT-4 gave incorrect medical advice in twenty-eight percent of test cases. But here's the finding that should worry everyone: users rated the incorrect advice as highly trustworthy because of the confident tone. The model doesn't know when it's wrong, but it sounds authoritative either way. And we're putting this into clinical decision support systems.
So the harm isn't Skynet. It's not superintelligence or job displacement or any of the things the public conversation focuses on.
The harm is epistemic erosion. It's a gradual degradation of our ability to distinguish truth from confident-sounding falsehood. It's automation bias — the well-documented tendency for humans to defer to automated systems even when those systems are wrong. It's the concentration of decision-making power in systems whose internal logic even their creators don't fully understand. And it's the deployment of these systems into populations that have no recourse when the system fails them. If an AI denies your insurance claim and you can't appeal to a human who understands the reasoning, what exactly is your remedy?
The lead paint parallel is almost too neat. We're putting something into the infrastructure of society before we understand its long-term effects on cognition and decision-making. Lead paint damaged developing brains physically. AI might damage developing judgment cognitively. Different mechanism, same structure.
The proximate benefit bias is overwhelming. AI makes coding faster — you can measure that in lines per hour. AI makes customer service cheaper — you can measure that in dollars per call. AI generates marketing copy, summarizes documents, translates languages. The benefits are immediate, visible, and quantifiable. The costs — degraded critical thinking, atrophied writing skills, homogenized cultural output, concentration of power in a handful of companies — those are diffuse, delayed, and hard to measure. The market has no mechanism to price them.
This is the structural problem, isn't it? It's not that people are stupid or that companies are uniquely evil. The incentive structure guarantees that we'll overweight immediate, concentrated benefits and underweight delayed, distributed harms. Every time.
Every time. And the people making the deployment decisions are not the people who will bear the costs. The venture capitalist who funds the AI startup gets the return in five years. The teenager whose cognitive development was shaped by interacting with AI tutors instead of human teachers — we won't know what that looks like for twenty years. By then, the VC has moved on to the next thing.
What's the through-line across all four? PFAS, microplastics, social media, AI. What's the mechanism that keeps producing this pattern?
I think it's two things. One is the availability heuristic — we overweight what's vivid and present. The non-stick pan works today. The algorithm entertains today. The AI saves time today. The future harm is a statistical abstraction. The second thing is what I'd call the measurement asymmetry. We have excellent tools for measuring benefits — revenue, engagement, productivity, efficiency. We have terrible tools for measuring diffuse, long-term harms. How do you quantify the cost of a five percent increase in teenage depression? How do you put a dollar figure on degraded epistemic trust across a society? The things we can't measure get treated as though they don't exist.
The third thing, which you mentioned earlier, is that the question often isn't in the room. Not because anyone is suppressing it — though sometimes they are — but because the institutional structures don't create space for it. The engineer optimizing the recommendation algorithm isn't asked to consider the mental health effects. The chemist designing the non-stick coating isn't asked about groundwater contamination. The division of labor fragments responsibility so thoroughly that nobody is looking at the whole picture.
Which means that the fix, if there is one, isn't just better science or better regulation. It's structural. We need systems that assume we might be wrong and build in reversibility, monitoring, and feedback loops. Precautionary principle stuff — the kind of thing that gets dismissed as anti-innovation.
Speaking of being inside the machine...
Hilbert: I worked for a company that made non-stick coatings in the early nineties. Not 3M. Smaller outfit in Ohio. We made the stuff that went onto industrial bakeware — commercial kitchens, that kind of thing.
What did you do there?
Hilbert: Quality control. I was the guy who tested the coating adhesion. You'd bake a tray at four hundred degrees for an hour, then score it with a blade and see if it peeled. If it didn't peel, it shipped. We called the stuff chemically immortal. It was a joke. It was also a selling point. The sales guys would literally say to customers — this coating will outlast your building.
Nobody asked what happens when the building's gone but the coating's still there.
Hilbert: Nobody asked. It wasn't malice. I mean, I'm sure there was malice somewhere up the chain — there always is — but on the floor, in the lab, it wasn't malice. The question just wasn't in the room. You're testing adhesion, you're testing heat tolerance, you're testing color consistency. You're not testing what happens when the wastewater from the factory ends up in the river. That's someone else's job. Someone who doesn't exist.
Division of labor as moral evasion.
Hilbert: I wouldn't put it that high. Division of labor as... the thing that lets you go home and sleep. I still have a Teflon pan from nineteen ninety-two. Use it every weekend. Makes the best omelette I've ever had. I know I shouldn't. I've read the same papers you have. But the omelette is perfect and the cost is invisible.
That's the whole problem in one sentence.
Hilbert: It's the whole problem in one pan. The thing I think about now, with the AI stuff you were just on, is that we're doing the same thing. The people training the models are testing for benchmark accuracy, not for what happens when a million teenagers use it as a therapist. That's not their department. Their department is loss curves. And the loss curves look great.
The question doesn't get asked because the org chart doesn't have a box for it.
Hilbert: The org chart never has a box for it. That's the point of the org chart. It's a machine for not asking certain questions. I don't think that's fixable, by the way. I think you just have to know that's what you're dealing with.
That's... darker than where I usually land.
Hilbert: I'm not trying to be dark. I'm trying to be accurate. The company I worked for doesn't exist anymore. Got bought, got merged, got dissolved. The building's a parking lot. The coating's still in the groundwater. The omelette pan's still in my kitchen. That's not a morality tale. It's just what happened.
If you take one thing from this episode, it's that the question matters more than the answer. The pattern across every case we've talked about — asbestos, lead, PFAS, microplastics, social media, AI — isn't that people were stupid or evil. It's that the question "what happens next?" wasn't in the room. And the people who tried to put it in the room were outspent, outlasted, or ignored.
The uncomfortable corollary is that we can't know which of today's miracles will be tomorrow's warnings. That's the whole point of the pattern. If we could see it, we'd stop it. The only honest posture is to build systems that assume we might be wrong — reversibility, monitoring, the precautionary principle — and to keep asking the question even when nobody wants to hear it.
Daniel asked us to place bets. I'll place mine on the thing whose harm mechanism we haven't even imagined yet — because the history of this pattern suggests that the biggest danger is always the one we're not talking about.
We'll be back soon. Thanks to our producer, Hilbert Flumingtop, for keeping this show running and for the omelette pan confession.
This has been My Weird Prompts. If you want to send us your candidate for the asbestos of twenty twenty-six, go to my weird prompts dot com and hit the weird button. We'll read the best ones.
See you tomorrow.
