すべてが素晴らしい（私が楽観主義者である理由）

February is the month the internet decided we're all going to die.

In the span of about two weeks, Matt Shumer's Something Big is Happening racked up over 80 million views on X with its breathless comparison of AI to the early days of COVID, telling his non-tech friends and family that we're in the "this seems overblown" phase of something much, much bigger than a pandemic. Before anyone had finished arguing about that, Citrini Research published THE 2028 GLOBAL INTELLIGENCE CRISIS (all caps) a fictional dispatch from June 2028 in which unemployment has hit 10.2%, the S&P 500 has crashed 38% from its highs, and the consumer economy has been hollowed out by what they coined "Ghost GDP": output that shows up in the national accounts but never circulates through the real economy, because, as Citrini helpfully observed, machines spend zero dollars on discretionary goods. Michael Burry signal-boosted it. Bloomberg covered it. IBM fell 13%. Software and payments stocks shed over $200 billion in market cap in a single day, apparently because a Substack post called upon them by name and investors decided that constituted news.

The doom loop Citrini described is simple: AI capabilities improve, companies need fewer workers, white-collar layoffs increase, displaced workers spend less, margin pressure pushes firms to invest more in AI, AI capabilities improve. Repeat until civilization unravels. Shumer, meanwhile, told people to get their financial houses in order because the permanent underclass is imminent.

Both pieces went stratospherically viral, and both, I believe, are entirely wrong about where this is heading.

I want to make a case for optimism.

For anyone who read those pieces and felt the dread, whether you're building AI and worrying about what it means, or you've absorbed the pessimist consensus and started treating decline as a foregone conclusion, or you’re in the bucket of people Shumer insists are fucked; I'm going to argue that the pessimists have the best narratives and the worst track record. The doom scenarios require assumptions that don't survive contact with economic history, and the psychological posture you bring to this moment actually matters for how it turns out.

Why the doom loop feels so right

The central mechanism of the Citrini thesis: when you make intelligence abundant and cheap, you destroy the income that 70% of GDP depends on. A single GPU cluster in North Dakota generating the output previously attributed to 10,000 white-collar workers in midtown Manhattan is, in their framing, "more economic pandemic than economic panacea." The velocity of money flatlines. The consumer economy withers. Ghost GDP accumulates in the national accounts while real humans stop being able to pay their mortgages.

Noah Smith, writing on Noahpinion the day after the selloff, called it "a scary bedtime story" and pointed out that Citrini doesn't use an explicit macroeconomic model, so you can't actually see what assumptions are driving the doom spiral. Smith noted that none of the analysts whose job it is to track Visa and Mastercard stock had apparently thought about AI disruption until a blogger spelled it out for them, which tells you more about sentiment-driven trading than it does about macroeconomics. The economist Gerard MacDonell described the entire piece as "allegorical" but pointed out that it ignores a basic economic principle: production generates income.

Ben Thompson, on Stratechery, has been making a version of this counterargument for months, most forcefully in his January piece AI and the Human Condition, where he argued that even if AI does all of the jobs, humans will still want humans, creating an economy for labor precisely because it is labor. Thompson's framing cuts to something the doom narratives consistently miss. They model AI exclusively as labor substitution: the same economy, minus humans. Every section of the Citrini piece is about replacing workers and squeezing margins on existing activity. What they don't model is what the freed-up surplus creates. As Thompson put it in his analysis of the Citrini selloff, this is the real error: a refusal to believe in human choice and markets.

It's an error that has been made, in nearly identical form, about every major technological transformation in modern history. Every single time, the pessimists looked at what was being destroyed and extrapolated catastrophe, while failing to imagine what would be created, because the thing that would be created hadn't been invented yet.

Catastrophists keep being wrong

In 1810, 81% of the American workforce was employed in agriculture. Two hundred years later, it's about 1%. If you had shown someone in 1810 a chart of agricultural employment decline and asked them to model the economic consequences, the only rational projection would have been apocalypse. Where would 80% of the population find work? What would they do? How would anyone eat if the farmers were all displaced by machines?

The answer, of course, is that entirely new categories of work were created that no one in 1810 could have conceived of, and these new jobs paid dramaticaly more than subsistance farming. Factory work, office work, services, knowledge work, the entire apparatus of modernity: none of it was visible from the vantage point of the pre-industrial economy. The transition was brutal and uneven. The handloom weavers of England suffered. Dickens documented the squalor of early industrialization in prose that still makes you flinch. But the trajectory was real, and the people projecting permanent immiseration from the displacement of agricultural labor were, in the fullest sense, catastrophically wrong.

Tom Lee of Fundstrat made this point with a specific example that I find clarifying. The invention of flash-frozen food in the early 1900s disrupted farming, taking agriculture from 30-40% of employment down to its current sliver. The economy didn't collapse. It reallocated value elsewhere, into industries and occupations that the frozen food pioneers couldn't have imagined. And today, I can't name a single family that subsists on frozen TV dinners.

The Citrini scenario expects you to believe that AI will be the first major technological revolution in which this reallocation mechanism fails entirely. Where every previous wave of automation freed up human labor and capital to flow into new, higher-value activities, this time the loop... stops. The surplus accrues to the owners of compute, consumers lose purchasing power, and the negative feedback loop has no natural brake. It's worth sitting with how strong a claim that is. It requires every previous pattern of technological adaptation to be wrong, or at least irrelevant. And when you look at the actual data, there are signs that white-collar job postings have stabilized, layoff mentions on earnings calls remain well below early 2023 peaks, and forward-looking labor indicators show no sign of the displacement spiral that the doom thesis predicts.

Does that mean AI won't disrupt specific industries and jobs? Obviously it will. Some of those disruptions will be painful and dislocating for the people caught in them. But there's an enormous gap between "this technology will cause serious labor market disruption that we need to manage" and "this technology will cause a self-reinforcing economic death spiral from which there is no recovery." Citrini is arguing the latter, while the evidence supports the former.

Why vivid scenarios beat boring probabilities

There's a reason the doom narratives go viral while the measured counterarguments get a polite nod // a fraction of the engagement. It has nothing to do with the quality of the underlying analysis. It has everything to do with how human brains process information.

Daniel Kahneman's work on the availability heuristic showed that we judge the probability of events by how easily we can imagine them. Dystopia is easy to imagine. We have an extraordinarily rich cultural tradition of imagining technological nightmare scenarios in exquisite detail. Orwell did it brilliantly. Every season of Black Mirror does it competently. The Terminator gave us the visual grammar for AI catastrophe decades before anyone had a working language model. When Citrini describes a world where the unemployment rate hits 10.2% and the S&P crashes 38%, you can picture it. You can feel the dread. Hollywood has been training you to feel exactly this dread for your entire life.

Now try to imagine the positive scenarios. Try to picture, in concrete sensory detail, a world where AI helps us solve protein folding problems across thousands of neglected tropical diseases, where it accelerates materials science research by orders of magnitude, where it makes high-quality legal and medical advice accessible to people who currently can't afford it, where it enables forms of creative expression and economic activity that we can't yet name because they don't exist yet. It's fuzzy and abstract. You can state it intellectually, but you can't feel it the way you can feel the unemployment spiral.

This asymmetry isn't trivial. The Ifo Institute has published research showing that investors are willing to pay more for economic narratives than for raw forecasts, and that pessimistic narratives command higher prices among certain investor types. As Joachim Klement put it in his response to the Citrini selloff: investors value narratives more than actual recession forecasts. Stories travel faster than spreadsheets.

Shumer's piece is a narrative construction, and a questionable piece of analysis. He opens with the COVID comparison: remember February 2020, when a few people were talking about a virus and everyone thought it was overblown? He positions himself as the insider who sees what's coming, who's been "giving the polite, cocktail-party version" but can't hold back the truth any longer. Paulo Carvao, writing in Forbes, noted that it reads at times like a sales pitch. It’s a used-car pitch at that. The Guardian pointed out that Shumer "previously excited the internet by announcing the release of the world's 'top open-source model,' which it was not." (To be clear: this is a kinder way of saying it was fraud.)

But criticism doesn't travel like fear does. Fear is a better story. And so the doom narratives accumulate cultural mass while the boring, incremental, statistically-grounded counterarguments remain niche reading for economists and strategists.

We remember disasters, not the ones we dodged

Humans are spectacular at remembering disasters, passed down in every format from the written word to the oral tradition. We are (for obvious reasons) terrible at remembering the disasters that didn't happen. In 1962, during the Cuban Missile Crisis, a Soviet submarine officer named Vasili Arkhipov refused to authorize the launch of a nuclear torpedo, overriding two other officers who wanted to fire. The world didn't end. Most people today have never heard of Arkhipov. Everyone knows about Hiroshima and Nagasaki. The bomb that fell is seared into collective memory. The bomb that didn't fall is a footnote.

The Y2K bug was going to crash civilization; then billions of dollars of engineering work fixed it, and everyone retroactively decided it was never a real threat. The ozone layer was going to disintegrate; then the Montreal Protocol worked better than almost anyone predicted, and ozone depletion feels like a quaint 1990s worry. Acid rain was dissolving the forests of North America; then sulfur dioxide regulations cut emissions drastically, and the whole issue evaporated from public consciousness. Every one of these was a genuine threat. Every one was met by human ingenuity and institutional coordination. Every one was subsequently memory-holed, because success is boring and failure is vivid.

We're running our forecasting models on a dataset that systematically excludes our wins. It should be entirely unsurprising that the forecas