CHAPTER 01 / 03
Tasks before professions
When we talk about artificial intelligence, we tend to treat professions as indivisible units. Yet a job consists of different tasks: finding information, preparing drafts, making decisions, checking work and reaching agreements with other people. One of these becoming faster does not mean that all of them change to the same extent.
The ILO’s 2025 assessment examines potential occupational exposure to generative AI through tasks. It reports that one in four workers worldwide is in an occupation with some degree of exposure. That figure does not measure actual job losses. Taking the continuing need for human input into account, the assessment finds transformation more likely than complete replacement for many jobs. [1]
CHAPTER 02 / 03
Easy answers, difficult choices
When one stage of production becomes easier, value may also shift towards the question that precedes it and the judgement that follows. A well-written answer can conceal a faulty assumption. A quickly prepared analysis can circulate incorrect data more quickly.
Knowing how to use a tool and knowing how to assess its output should therefore be considered separate skills. Who retains control, how mistakes will be noticed, and where responsibility lies depend on the design of the work as much as on the tool’s features.
CHAPTER 03 / 03
Looking for another measure
If we measure intelligence only by the speed of answering, we miss these distinctions. Recognising uncertainty, consulting the right person, deciding not to do unnecessary work, or tracing a claim to its source are also forms of intellectual labour. Their value will not automatically rise; it will also depend on what institutions choose to measure.
Alongside asking how much a machine can produce, we need to ask what we count as work done well. We can discuss this without waiting for the next technology: how will we know that receiving more answers has helped us make better decisions?
25 September 2026 · Opening text completed and revised.