Discussion of automation and employment tends to fall into two positions. One holds that machines have always created more jobs than they removed, and that the present wave will prove no different. The other holds that this time the technology reaches into thinking itself, and that the comfort people draw from history no longer applies. Both sides produce large numbers, and neither is easy to test, because both are arguments about a whole economy over several decades.
The weakness they share is the level at which they are stated. A national employment figure is the sum of millions of separate decisions, and it moves slowly enough to hide almost anything. In the twenty years after the first factory robots were installed in car plants, total employment in most developed countries rose. That fact is true and nearly useless, because it says nothing about the particular workers whose skills lost their value in their particular towns.
A more useful question is not how many jobs exist but how the work involved in a job changes, and who pays for that change. Consider a hospital department that introduces a system to read images. The work of the doctor does not disappear. It divides. Part of the task, the routine comparison of one image against thousands of others, moves to the machine. What remains is the harder part: deciding which cases do not fit the pattern, and explaining a result to a frightened patient. The job title is unchanged. The daily experience of doing it, and the training required to do it well, are not.
This matters for policy because the two levels demand different responses. If the concern is the total number of positions, the natural answer is economic growth, and the argument becomes one about tax and investment. If the concern is the content of the work, the answer is training, standards and the terms on which technology is introduced — questions that are settled inside industries and companies, not in national budgets.
The evidence we collect reflects the level we argue at. Governments publish employment figures every month. Almost nobody publishes anything about how the content of ordinary work is changing, although the data exists inside firms and could be released in a form that protects individuals. Until that information is public, the debate will continue to be conducted with tools that cannot detect what people actually report experiencing.
There is a further difficulty, which is that the two effects arrive at different speeds. The change inside a job can be felt within a single year. The change in total employment, if it comes, appears over a decade or more. A public argument that only recognizes the slower measure will always conclude that nothing serious is happening, right up to the point where it becomes impossible to say so.
None of this determines whether the optimists or the pessimists are correct. It suggests that the question they are fighting over is the wrong one to fight over first. Before asking how many jobs there will be, it is worth establishing what those jobs will consist of, who decides, and who carries the cost of finding out.