2003 theoretical framework explaining labor market polarization through the substitutability of routine tasks and the complementarity of non-routine cognitive tasks with technology.
The ALM model, formalized by David Autor, Frank Levy and Richard Murnane in a landmark paper published in the Quarterly Journal of Economics in 2003, proposes a theory of the effects of technological progress on labor market structure based on a task taxonomy rather than occupations. The model distinguishes two main axes: task nature (routine versus non-routine) and dimension (cognitive versus manual). Routine tasks, codifiable by explicit rules, are substitutable by technology: form processing, accounting data entry, repetitive assembly operations. Non-routine tasks resist automation: in the cognitive dimension they include analysis, judgment, communication and creativity; in the manual dimension they include adaptive dexterity (plumbing, nursing). This framework predicts employment polarization: middle-skill routine-task jobs disappear, while high-skill non-routine cognitive jobs and low-skill non-routine manual jobs persist. The model implicitly includes Polanyi's paradox as a barrier: the non-codifiable nature of tacit knowledge protects complex cognitive tasks from substitution. The emergence of LLMs and agentic AI in the early 2020s empirically circumvented this barrier, forcing Autor himself to reassess his conclusions in subsequent work.
According to the ALM model, an M&A lawyer (non-routine cognitive tasks: judgment, negotiation, strategy) should be less substitutable than a legal assistant handling standard contract drafting (routine cognitive tasks). Yet empirical studies from 2023-2025 show that LLMs have begun automating tasks at the top of the income distribution, reversing the model's historical prediction.
ALM model, modèle de polarisation de l'emploi, routine-biased technological change