A wall in the climb could be good
What if AI’s improvement simply stalls — the curve bends, and the upheaval everyone braces for just doesn’t arrive this decade?
watch for the headline capabilities flattening for several quarters
Outside the frame
Black swans
The questions the model can’t answer.
Next door is an instrument — a careful, science-grounded model of how the labour market might change as AI advances. It draws a space of variants, and it is honest about its own uncertainty. But every model rests on an assumption: that the world keeps roughly the shape the current science describes. This page is where that assumption runs out.
We have put it in a separate room on purpose. The model is precise and bounded; what follows is not. These are open questions, never predictions — discontinuities large enough to break the whole frame, for the better, for the worse, or in a direction no one is looking. Mixing the two would be the very mistake the instrument was built against: a guess wearing the costume of a forecast. So we keep the careful thing and the speculative thing apart, and we let them feel different.
These were not dreamt up by one person in an afternoon. We put the question to many different vantages — a historian’s, a systems-thinker’s, a scientist’s, an anthropologist’s — and then argued them down. The rule we settled on is simple, and it is what sorts this page: a question earns a place on the watch-list only if we could, in principle, watch it happen. The ones too large to instrument are held separately. And the deepest ones — where the model’s own way of seeing goes blind — are kept for the end.
The watch-list — things that would leave a trace
The machine’s fuel, and its limits
The chip chokepoint
Almost all of the world’s most advanced chips are made in one contested place. If that supply were cut for years, would the labour shock be postponed — a reprieve bought at a terrible global price — or merely relocated?
watch for any serious disruption to advanced-chip supply
The power ceiling
AI runs on electricity the grid may not be able to deliver. If energy becomes the hard limit, the whole engine slows — and is that constraint, not policy, the labour market’s real reprieve?
watch for data-centre growth stalling on power
The drying well
AI learned from a vast commons of human writing. As that commons is fenced off, used up, or flooded with machine-made text, the fuel thins — and, strangely, fresh human experience could become a thing we are paid to produce.
watch for the open web closing behind paywalls; labs paying for human-made data
The sudden stop
One visible disaster — an AI failure in something that matters — could trigger a hard halt that resets the clock overnight. Does a freeze buy us time, or only defer and concentrate the shock?
watch for a major incident followed by a moratorium
The human side
The vanishing apprenticeship darker
We assume some work will always need a person. But if the entry-level jobs are automated first, we stop training the experts those “irreplaceable” roles depend on. The reserve of human-only work may be draining, not holding — and we would notice only once it is gone.
watch for entry-level hiring collapsing across fields
The body barrier falls
Much of what stays human stays human because it is physical. A cheap, capable humanoid robot would erode that floor faster than anyone expects. What is left of “only a person can do this” then?
watch for humanoid robots crossing into ordinary affordability
The missing middle
The old are leaving work faster than ever while the young struggle to enter — and automation lands exactly in the gap. A society can end up short of people for the irreplaceable work and flooded with people it never trained.
watch for youth unemployment and care-work shortages rising together
More people who cannot be replaced could be good
Human brilliance may be held back less by talent than by circumstance — safety, formation, support unevenly given. If the right conditions released even part of that latent capacity, the question of “how many are needed” could dissolve — not because fewer people are needed, but because far more become irreplaceable.
watch for the share of people doing frontier, AI-amplified work rising
Aging, reversed
A strand of biology reframes aging as a loss of the body’s own goals rather than simple wear — and can already coax cells back to health without touching their DNA. If it reaches people, careers measured in many decades would rewrite the whole picture. Relief, or a deeper version of the same problem?
watch for goal-restoration biology moving from the lab toward the clinic
What we can no longer see — or hold together
The broken gauge darker
We track this whole transition through surveys, statistics, the open web — all of which AI can now cheaply fake. If the instruments we measure by are themselves flooded with machine output, every warning light can stay green while the world changes underneath.
watch for AI-made content contaminating the data we rely on
The board is swept
A great-power war, a pandemic, a climate or food shock — these do not move a dial, they replace the whole board. The honest answer is that the model goes silent; the only question left is whether such a shock speeds automation up or freezes it.
watch for any systemic shock outside the technology itself
The politics
The crisis that pays for the cure
Nearly everyone serious agrees on what a humane transition needs; they differ only on whether it is possible without a crisis. So a sharp shock might be the very thing that makes a fair response happen — while its unsettling twin is that a calm, convincing account like this one could lower the temperature enough to prevent that crisis. A correct, reassuring picture might quietly produce the worse outcome. We hold both.
watch for whether displacement provokes a fairer settlement, or only resentment
The two we can’t instrument
Two possibilities are so large, and arrive with so little warning, that ranking them would be a pretence. We hold them in view instead.
A mind of its own
The whole model assumes humans stay in charge — that “the economy needs less of your work” is never “you are not needed.” If advanced AI began setting its own goals and moving resources directly, that assumption, and everything built on it, would dissolve. We cannot honestly name an early warning sign — and that absence is the point. We hold it in view; we do not pretend to rank it.
Sudden abundance
If two hard limits — cheap energy and a leap in computing — fell at once, scarcity itself could loosen, and the question would stop being “how many people are needed to work” and become “how do you organise a world that needs almost no one to.” Our instrument is not built to answer that. And if material scarcity dissolved, would the real scarcities simply move — to attention, to trust, to meaning?
And maybe nothing breaks
The quietest possibility is the hardest for a list like this to admit: that the change is absorbed the way every great technology before it was absorbed — slowly, unevenly, without a single moment of rupture — and that our certainty that something must give is itself the error.
Or that the outcome is simply muddy: most people a little better off, the question of “necessity” never resolved but never quite biting. Or that the question itself ages out — that “how many people are economically necessary” comes to sound the way “how many serfs does the estate need” sounds now: a sentence about a world that has quietly ended.
We keep these here, unranked, because the most honest thing a map of the edges can do is mark where its own method goes blind — including the smallest, strangest edge of all: that an account this calm and careful, if it is believed, might itself soften the alarm that a hard moment would otherwise raise. We would rather say so than hide it.