The standard automation model treats labour as a single factor that capital substitutes for.
Acemoglu and Restrepo broke labour into tasks and asked which tasks automation displaces,
which it complements, and which entirely new tasks the technology creates. The long-run labour
share depends on the balance. Plotted here: eight task families, scored on displacement (how much
AI substitutes) and reinstatement (how much new work AI creates around it).
Hand-tuned composites synthesised from Acemoglu & Restrepo (2020), MIT FutureTech and OECD work-task taxonomies.
Reading the colours
Net positive — AI creates more new tasks than it displaces. Labour share rises locally.
Neutral — Roughly balanced. Long-run effect depends on institutional response.
Net negative — Displacement dominates. Labour share falls, wages compress.
The thesis
AI's task profile is unusual: it threatens cognitive, white-collar, mid-skill tasks that previous
waves of automation left untouched. The distributional consequences for South Africa, where the
formal middle-skill labour market is already small relative to the informal sector, are genuinely
unclear and worth research investment, not casual extrapolation from US data. The middle is being
hollowed in the rich world; what hollowing means in an economy without the middle to begin with
is the open question.