does not replace occupations; it replaces tasks, and every occupation is a of them.
Whether a job disappears depends on what proportion of its is and what remains.
The tasks that resist automation are not the ones people expect.
Arithmetic went first, then pattern recognition, then the drafting of routine prose.
What remains is largely physical in settings and the handling of other people.
Both were long regarded as low-skilled, which is a judgement about pay rather than about difficulty.
A machine can pass a professional examination and cannot reliably fold a towel it has not seen before.
The effect on the distribution of earnings is therefore not a simple movement in one direction.
Work in the middle of the distribution, which is routine and , has been consistently for thirty years.
Work at both ends has grown, and the two ends are paid very differently for reasons unrelated to automation.
Retraining is the standard policy response and has a record that is honestly described as poor.
The difficulty is not that adults cannot learn but that the programmes are designed around the wrong unit.
They train for occupations, which move, rather than for tasks, which transfer.
A worker retrained for a specific job is in the same position when that job changes again.
Programmes built around tasks have better outcomes and are harder to advertise.
A course promising a named job recruits well and a course promising a capability does not.
Policy has followed recruitment rather than outcomes, which is an ordinary failure and not a technological one.
The technology has been doing something legible for decades while the response has been aimed at the wrong thing.