In July 2026 more than 250 economists and AI researchers, sixteen of them Nobel laureates, signed a statement organised by Stanford's Digital Economy Lab. The judgement was that the change AI brings could be larger than the Industrial Revolution but unfolding over a vastly shorter time frame. Among the signatories were researchers who had spent years disputing displacement forecasts.
There is one question the statement does not answer. When a firm automates part of a job, someone has to fund whatever comes next for the person who did it: the months without income, the retraining, the search for different work, the lower wage in the new role. Who carries that cost?
Who Pays for the Journey?, finished in July, sets out to answer it. It is written by Jussi Kajala, CEO of 3DBear, Glyn J. Eggar, an actuary in Scotland, and Mitchell Weisburgh, an American education entrepreneur and investor.
The mistakes are being made now
Public argument about AI stays up in the clouds. How many jobs go, and whether we arrive at broad abundance or widespread hardship? End-state speculation is cheap, because nobody can be shown wrong for a decade. The truth is that the mistakes a society makes are being made right now.
The paper's comparison is the closure of Britain's coal mines in the 1980s. The pits were going to close on any plausible energy-policy path. The catastrophe was the absence of a transition mechanism. Forty years on, those communities still show median hourly earnings 6 to 7 per cent below the national average. Over 90,000 displaced miners were diverted to incapacity benefits, so joblessness became permanent. The damage was severe.
A single employer cannot solve this problem
Consider two logistics centres in the same city. One automates its sorting, cuts everyone's hours by a fifth, maintains full pay and holds its prices where they were. The other automates, lets a fifth of its staff go and passes the saving into lower prices. Within a quarter the first centre is losing contracts.
From this follows the paper's central claim. If a behaviour is competitively suicidal, appeals to responsibility will not produce it. Only a rule binding all competitors simultaneously will. Such a duty is priced into the market at once, as a cost of doing business, so no firm loses ground for being the decent one.
The limits of voluntarism are clearest in training. Germany's dual apprenticeship system, the most admired in the world, is voluntary for most firms, and the number of firms offering apprenticeships has fallen. Denmark obliges every employer to contribute whether it trains anyone or not, and the scheme has been evaluated and found to have positive employment effects.
When the pits closed, no such mechanism existed. This time it has to be built in advance, because the change ahead is larger and faster.
The bottom rungs have been taken off the ladder
The traditional path to expertise runs through the bottom rungs. The junior does the easy work badly, learns from doing it and becomes senior. AI automated the easy work, so the load-bearing rung went first.
In US payroll data there is a roughly 16 per cent relative employment decline for 22- to 25-year-olds in the most exposed occupations. It arrives through hiring rather than firing. Entry-level postings are down around 35 per cent in three years.
Meanwhile the value of the work AI cannot do is rising. AI-assisted consultants were 19 points worse outside the frontier of what the language model handles well.
Finland still has a chance
ETLA finds that AI exposure has not displaced young workers in Finland. Employment among young people in high-exposure occupations tracks almost identically with low-exposure occupations, and around 40 per cent of employed Finns already use generative AI at work. ETLA attributes the buffering to the Nordic labour-market model and strong dismissal protection.
The paper also makes proposals, addressed separately to workers, employers and policymakers.
The full paper is available here: Who Pays for the Journey?