Why the Intelligence Crisis Scenario Is Wrong

The Canary Is Fine: Why the Intelligence Crisis Scenario Fails on Its Own Terms

A recently circulated piece from CitriniResearch, styled as a “macro memo from June 2028,” paints a vivid picture of economic collapse driven by artificial intelligence. The S&P 500 is down 38%. Unemployment has hit 10.2%. White-collar workers are driving Ubers. It is well-written, internally dramatic, and almost entirely wrong, not because the authors are silly, but because the economic reasoning contains a series of compounding errors that, once identified, cause the entire edifice to collapse.

The essay deserves engagement precisely because the fears it articulates are widely shared. What follows is an attempt to show why those fears, while emotionally compelling, do not survive contact with basic economic logic.

Ghost GDP is not a thing

The essay’s most evocative concept is “Ghost GDP”: output that appears in the national accounts but never circulates through the real economy. A GPU cluster in North Dakota produces the equivalent output of 10,000 white-collar workers, but unlike those workers, it doesn’t eat at restaurants or pay a mortgage.

This fundamentally misunderstands what GDP measures. GDP is not “stuff produced” in some abstract sense; it is the market value of goods and services purchased. If a GPU cluster generates output that nobody buys, it does not show up in GDP. If it does show up, someone is buying it, which means money is circulating, which means the “ghost” framing is incoherent.

The essay reports nominal GDP growth in the “mid-to-high single digits” while simultaneously describing a consumer economy that is withering. Pick one. If consumers are not spending, where is the GDP coming from? If it is coming from business-to-business transactions and capital investment, then the composition of demand has shifted, but demand itself has not disappeared.

The accounting identity the essay cannot escape

GDP splits between labour and capital. Call labour’s share L and capital’s share K, so GDP = L + K. The essay claims that GDP is rising, labour’s share is falling (from 56% to 46%), and corporate profits are strong. That means K is growing both as a share and in absolute terms. The equity market capitalises future K. Interest rates are falling (the essay has the 10-year dropping from 4.3% to 3.2%). Rising corporate profits, falling discount rates, growing economy: the net present value of future earnings goes up, not down.

Yet the essay describes a 38% drawdown in the S&P 500. This is not a minor inconsistency. To make the maths concrete: if GDP grows at 6% nominally and capital’s share rises from 56% to 60%, aggregate profits grow by roughly 14%. Discount that at 3.2% rather than 4.3%, and equity valuations should be significantly higher. The essay needs an enormous risk premium expansion to make its scenario work, and the mechanism it proposes for that expansion does not hold together.

Comparative advantage does not take a holiday

The essay’s central metaphor is the “Intelligence Displacement Spiral”: AI replaces workers, companies reinvest savings into more AI, AI improves, more workers are replaced. A feedback loop with “no natural brake.” It is also a description of every productivity-enhancing technology in history, and the natural brake has always been the same: comparative advantage.

Ricardo demonstrated in 1817 that even if one party is more productive at everything, both parties still benefit from specialisation. This is not an optimistic assumption; it is a mathematical consequence of opportunity cost. AI capacity is finite and must be allocated, and rational allocation means deploying AI where its advantage over humans is greatest, which necessarily leaves tasks where humans retain comparative advantage.

The essay’s implicit response is that AI improves so quickly that comparative advantage shifts faster than humans can reallocate. This is a claim about speed, not about the existence of the mechanism. The distinction matters enormously. If comparative advantage holds but adjustment is slow, you have a transition problem. If comparative advantage somehow does not hold, you have a civilisational crisis. The essay writes as though it is describing the latter while only providing evidence for the former.

It is worth noting what happens at the theoretical extreme. If AI becomes so capable and so cheap that it has no opportunity cost at all, then scarcity itself is largely solved. In that world, “unemployment” means something closer to “leisure” than “destitution,” because the goods and services people need are essentially free. You cannot have mass suffering in a world of radical abundance; the concepts are mutually exclusive.

Scarcity shifts, it does not disappear

The essay equivocates between intelligence becoming “abundant” and intelligence becoming “less scarce.” These are not the same claim. AI has not made intelligence abundant in any meaningful sense; it has reduced the cost of a specific type of cognitive labour, namely analytical, procedural, text-based reasoning. That is a long way from saying that everything humans do is no longer needed.

When one input becomes cheaper, scarcity shifts to its complements. This is one of the most reliable patterns in economic history. Mechanised agriculture made raw physical labour less scarce, and scarcity shifted to the ability to operate and maintain machines. Computing made arithmetic less scarce, and scarcity shifted to knowing what questions to ask of the data. In each case, the pessimists fixated on what had been devalued and ignored what had become more valuable as a direct consequence.

The same rebalancing is already visible with AI. As the cost of generating a first draft, a code prototype, or a market analysis falls toward zero, the value of knowing which draft to pursue, which prototype to ship, and which analysis to trust rises. Taste, trust, physical presence, the ability to take responsibility for a decision: none of these are close to being automated, and all of them become more valuable precisely because the commodity cognitive work around them is now cheap.

The essay treats AI and humans as pure substitutes in a zero-sum contest. The far more likely equilibrium, and the one every historical analogy supports, is that they are complements.

The price mechanism still works

The essay tells the story of a Salesforce product manager who loses her $180,000 job and ends up driving for Uber at $45,000. This is presented as evidence of structural economic breakdown. It is actually evidence of the price mechanism working exactly as theory predicts.

Labour is repricing to reflect a new supply-demand equilibrium. This is painful for the individual, genuinely and seriously painful, but it is not a market failure. The question is whether the new equilibrium wage provides an acceptable standard of living, and here the essay ignores half of the equation. If AI is expanding supply across the economy, prices fall. The same $45,000 buys substantially more when software, legal services, financial advice, travel booking, and dozens of other categories have seen their costs collapse.

The mechanism is straightforward: the technology that displaces the job also cheapens the output. Mechanised agriculture destroyed millions of farm jobs but made food radically cheaper. Automation displaced factory workers but made manufactured goods so affordable that service-sector workers on modest wages could own things that would have been luxuries a generation earlier. Falling prices do at least as much for living standards as rising wages, and there is no reason to assume this pattern reverses with AI. The essay provides none.

The mortgage problem is real but solved

The essay’s strongest section concerns residential mortgages. $13 trillion in debt underwritten against income levels that may no longer hold. This is a legitimate concern about nominal rigidities: mortgages are fixed obligations that do not adjust when the borrower’s income falls. A displaced worker’s cost of food and software may drop, but her $4,200 monthly mortgage payment stays the same.

We faced almost exactly this situation in 2008, and the playbook is well-established. The Federal Reserve buys distressed mortgage-backed securities (they purchased over $1.7 trillion of agency MBS between 2008 and 2014). Banks receive capital injections. Forbearance programmes keep borrowers in their homes. The currency is debased sufficiently that nominal asset values stabilise, keeping the collateral base intact and the banking system solvent. The real adjustment happens underneath the nominal floor that policy provides.

What makes the AI scenario considerably more tractable than 2008 is that the underlying economy is genuinely productive. In 2008, extend-and-pretend was papering over loans that should never have been written against assets that were never worth their sticker price. In an AI-driven displacement scenario, the economy is producing more real output than ever. You are buying time for the distribution of that output to sort itself out. The collateral is good; the income streams just need to reallocate. That is a far easier problem than one where the collateral itself was fraudulent.

The feedback loop has a brake

The essay’s most repeated claim is that the AI displacement spiral has “no natural brake.” The brake is straightforward: revenue.

If displaced workers spend less, aggregate demand falls, firms sell less, and their revenue declines, which means they have less money to invest in AI. The essay tries to sidestep this by arguing that AI investment is “opex substitution” rather than new capex, so it continues even as revenue declines. Yet opex substitution has a floor. A company that was spending $100 million on employees and $5 million on AI, now spending $70 million on employees and $20 million on AI, has already captured the easy savings. If revenue is falling because customers are disappearing, the total budget continues to shrink. The company cannot invest more in AI than it earns.

There is also the demand-side brake: if consumers have less money, the products AI helps produce do not sell, which means the return on AI investment falls. Capital allocation follows returns. If the return on AI declines because nobody can buy what it produces, capital flows elsewhere. This is just the price mechanism operating across time.

What the essay actually describes

Strip away the drama and the fictional Bloomberg headlines, and the CitriniResearch memo describes something fairly mundane: a rapid technological transition that causes a temporary financial dislocation. The railway boom and bust of the 1840s, electrification in the early 20th century, the dot-com cycle, the 2008 financial crisis: each time, new technology created enormous wealth, the transition was disorderly, financial structures cracked under the strain, policy intervened, and the economy emerged wealthier on the other side.

The essay wants to argue that “this time is different” because AI substitutes for human intelligence itself. Every general-purpose technology in history has substituted for whatever the binding constraint on productivity was at the time. The plough substituted for human physical effort. The printing press substituted for human copying labour. The spreadsheet substituted for human calculation. Each time, the scarce input became less scarce, prices adjusted, and humans reallocated. “This time is different” is the most expensive sentence in the history of economic forecasting, and the essay provides no rigorous reason to believe it applies here.

The real risk

None of this means the transition will be pleasant. People will lose jobs they valued. Communities built around specific industries will struggle. The financial system will experience stress that requires active policy management.

The risk is not that the economics are broken. The risk is that the politics are. If governments are too slow to deploy the tools they have, or if populist anger leads to counterproductive policy, the adjustment will be slower and more costly than it needs to be. This is a serious concern, but it is a profoundly different thesis from “the economy no longer resembles the one any of us grew up in.” The economy will be fine. Whether our political institutions manage the transition competently is genuinely uncertain.

The canary is not dead. It is not even unwell. It is just moulting, which, if you have never seen it before, can look alarming. Give it a minute.