The intelligence curse
The resource curse applied to AI: if the income of states and firms comes from intelligence on tap rather than from people, the incentive to invest in people —education, jobs, social protection— disappears.
- Horizon
- 3–10 years
- Evidence
- Projected
- Consensus
- Low
Luke Drago and Rudolf Laine carry the natural-resource curse over to AI. The thesis, in their words, is that powerful actors —states and companies— stop having incentives to care about ordinary people, and that when they deploy general intelligence they lose the incentive to invest in people [336]The Intelligence CurseView source ↗.
The mechanism has four steps. One: automation advances inside existing organisations starting from the bottom, what they call pyramid replacement. Two: it also reaches atypical human talent, not only routine tasks. Three: non-human factors of production —capital, resources, control over the AI— weigh more than people do. Four, and this is the step that sets it apart from gradual disempowermentGradual disempowermentThe idea that humanity could lose control over its own future without any catastrophe, simply because more and more economic, political and cultural decisions pass to AI systems.For exampleLike a town that hands each of its services to an outside company until one day it notices it no longer decides anything of its own.: if income comes from intelligence on tap rather than from people, there is no return on investing in education, in jobs or in a welfare state for those left out. The analogy they use is explicit: AI resembles coal or oil more than it resembles the plough or the computer.
Its virtue is that it makes a falsifiable, dated prediction. If the mechanism is real, the damage shows up first on the entry rung, not spread across the whole ladder. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen report exactly that from administrative payrolls: a 19% gap in the relative employment of workers aged 22 to 25 in exposed occupations, operating through less hiring rather than through more separations [164]Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial IntelligenceView source ↗.
The criticism also has a name. Zanna Iscenko and Fabien Curto Millet argue that the diagnosis is faulty and that what has been observed is the predictable consequence of a classic macroeconomic shock [566]Looking for the Ladder: Is AI Impacting Entry-Level Jobs?View source ↗. The authors of the figure themselves ask that it be read as an early, descriptive indicator, not as a causal estimate [165]No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%View source ↗. Daron Acemoglu, by another route, estimates aggregate macroeconomic effects considerably more modest than the ones this scenario assumes [15]The Simple Macroeconomics of AIView source ↗. And one conceptual objection remains: the resource curse is not deterministic —Norway, Botswana— and its moderating variable is prior institutional quality, which the analogy leaves out.
Profile
- SpeedSlow
- ReversibilityHard to reverse
- ConcentrationVery high
Assumptions that must hold
That control of AI is concentrable: capital-intensive, with few players able to sustain the frontier.
That the tax base effectively migrates from labour to capital and compute without fiscal systems adapting in time.
That substitution advances from the bottom up, breaking the entry rung of each career first.
That prior institutional quality is not enough to defuse the curse, as it was with natural resources in Norway or Botswana.
What would refute it
That labour damage appears spread across the whole career ladder instead of concentrating at the entry rung. The pyramid-replacement prediction is directly falsifiable and this is its negative face.
That the fall in entry-level employment turns out to be attributable to the sharpest interest-rate tightening in four decades rather than to AI exposure.
That fiscal systems shift revenue toward capital and compute without public spending on education and social protection falling.
That the AI frontier becomes cheap and dispersed enough that control stops being concentrable.
Early signals
- Youth employment decline in AI-exposed occupationsObserved
It is the cleanest observational confirmation any of the six scenarios has: the damage shows up at the entry rung and operates through less hiring, not more separations. It remains descriptive, not causal.
- AI adoption by US firms (Census BTOS)Weak signal
Adoption is very uneven by sector and firm size, and pyramid replacement requires it to spread beyond Information and Finance.
«Not observed» is not a clean bill of health: it means nobody has seen it yet, which is different from it not happening.
View as table
| Indicators | State | Note |
|---|---|---|
| Youth employment decline in AI-exposed occupations | Observed | It is the cleanest observational confirmation any of the six scenarios has: the damage shows up at the entry rung and operates through less hiring, not more separations. It remains descriptive, not causal. |
| AI adoption by US firms (Census BTOS) | Weak signal | Adoption is very uneven by sector and firm size, and pyramid replacement requires it to spread beyond Information and Finance. |
See among the scenarios →Report a mistake in this entry →
Sources
- [336] The Intelligence Curse · Drago, Luke 2025
- [164] Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence · Stanford Digital Economy Lab 2026
- [165] No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19% · Stanford Digital Economy Lab 2026
- [566] Looking for the Ladder: Is AI Impacting Entry-Level Jobs? · Economic Innovation Group 2026
- [15] The Simple Macroeconomics of AI · MIT / NBER 2025