In the past 15 years a `task-based' literature has emerged, exploring the consequences of technological change on the labour market. This literature relies on a particular understanding of the capabilities of machines -- known as the `ALM hypothesis'. However, this hypothesis has often led the literature to underestimate these capabilities. Tasks that were believed to be out of reach of automation can now be automated. In this paper I try to make two contributions. First, I set out two distinct explanations for why these capabilities were underestimated -- one that is explored in the recent literature and maintains the ALM hypothesis, and a new explanation that challenges it. I propose a new hypothesis that nests the ALM hypothesis as a special case. Secondly I build a new task-based model based on new reasoning about how machines operate. This leads to a pessimistic account of the prospects for labour. In a static model, increasingly capable machines drive down relative wages and the labour share of income and force labour to specialise in a shrinking set of tasks. In a dynamic version of the model, labour is driven out the economy at an endogenously determined rate, forced to specialise in a shrinking set of types of tasks, and wages steadily decline to zero. In the limit, labour is fully immiserated and `technological unemployment' follows.
This is still a working paper, and all comments and reflections are gratefully received.