A Diagnosis, Not a Warning: Why the Discontinuity Holds

Most arguments about AI and work end up somewhere comforting. New jobs will emerge. People will adapt. The market will sort it out, as it always has. The Discontinuity Thesis is not comforting, and the reason is not pessimism. It is that the usual escape routes do not survive contact with the mechanics. This post sets out the structural case plainly, and then deals with the strongest objections to it.

The mechanism, stated without politics

Start with the Iron Law of Markets. A firm that can buy the same output more cheaply will do so, because the firm that does not is undercut by the one that does. This is not greed, it is competition. Now add the second input. Digital labour costs roughly thirty cents an hour against thirty dollars an hour for the human equivalent, and unlike the human it improves on an exponential curve rather than over a career. Put those two together and you get an engine of obsolescence that no individual employer can switch off.

This is the heart of it, and it is why the thesis rests on Unit Cost Dominance rather than on any claim that AI is conscious or creative or about to wake up. It does not need to be. It only needs to be cheaper at producing economically valuable output, and it already is across a widening band of cognitive work.

Why no single firm can opt out

The obvious response is that employers will choose restraint. They will keep people on, because consumers with wages are also customers. This is where the multiplayer prisoner’s dilemma bites. Any one chief executive who keeps expensive humans while rivals automate simply loses to those rivals and is replaced. The cooperative outcome, everyone keeps paying wages so demand survives, is not reachable, because defection always pays the individual defector even as it degrades the whole. The Nash equilibrium is mutual automation, and it is mutually destructive. Nobody chooses the cliff. Everybody walks off it because stopping is a competitive death sentence.

The circuit that breaks

Here is the part people miss. A capitalist economy runs on a wage-demand circuit. Firms pay wages, wages become consumer spending, that spending becomes firm revenue, which funds the next round of wages. Cut wages out of the loop and the loop does not just shrink, it breaks. You can have an enormous productive surplus and a collapsing market for it at the same time, because the people who would have bought the output no longer have income to buy it with. Production soars while demand falls toward the floor. That is the severance, and it is the difference between a recession and a discontinuity.

The escape routes, one by one

An argument is only as strong as the objections it can take, so here are the serious ones.

Physical work is the refuge. The arithmetic does not support it. If cognitive displacement puts tens of millions out of work, the physical roles available to absorb them are a fraction of that number, and they require different skills, locations and bodies. You cannot move seventy-five million displaced cognitive workers into thirty-five million physical slots, and you cannot retrain a forty-five-year-old analyst into a roofer at population scale. There is no absorption channel large enough, and the wages in the channels that do exist get compressed by the flood of entrants. Refuge for some is not refuge for the labour market.

The verifiers survive. The hope is that a class of humans stays employed checking and steering the machines. For a while, yes, and that is exactly the Verification Divide that the thesis predicts, an elite few who compound their advantage while the majority slide into cognitive obsolescence. But the boundary keeps moving. The tasks reserved for human verifiers are themselves automated as the systems improve, including the verification of AI by other AI. There is no stable task boundary to stand on, which is the whole point. A refuge that shrinks every quarter is not a refuge, it is a queue.

Universal basic income solves it. This conflates two different things. Redistribution can keep people alive. It cannot give them economic participation, and participation is what the circuit needs. A population kept on a stipend is not buying with earned income, it is being managed, and the political economy of that arrangement is not democratic agency, it is dependence. UBI answers survival. It does not answer the severance, because it does not put wages back into the loop, it routes around the loop entirely.

Abundance flips the whole thing. This is the strongest objection, and worth taking seriously. If production costs genuinely fall toward zero, perhaps scarcity stops mattering and the wage circuit becomes irrelevant rather than broken. I do not dismiss it. But it is a bet on a phase transition we have not seen, against a mechanism we are watching operate now. The honest version of the falsification test is this. Show AI driving output growth that decouples cleanly from human wages while broad living standards rise. If that arrives, the thesis is wrong. Until it does, the abundance case is a hope, and the severance is a measurement.

What would actually falsify this

A thesis that cannot be wrong is not worth holding, so here is what it would take. You would need scalable, wage-sustaining, AI-resistant, value-creating work for a majority of working-age adults, not in theory but in practice. You would need a coordination mechanism that lets firms escape the prisoner’s dilemma without being undercut. You would need a stable task boundary that does not erode. And you would need real economic agency for the people on the receiving end, not a managed allowance. All of these, at once, not one of them. The reason the thesis is hard to dislodge is not that it forbids a good outcome. It is that the good outcome requires every one of those conditions to hold together, and each is individually under pressure right now.

A diagnosis, not a warning

The framing I keep coming back to is medical. A warning says here is something that might happen if you are not careful. A diagnosis says here is the mechanism already running in the patient. The wage-demand circuit is the patient, Unit Cost Dominance is the agent, and the prisoner’s dilemma is the reason nobody can administer the cure unilaterally. None of this is certain in the way a sum is certain. It is structural and probabilistic, a strong tendency in an uncoordinated system rather than a fixed appointment with collapse. But the burden of proof has quietly shifted. The comforting story now has to explain which specific escape route holds, and against the mechanics, none of the usual answers do.