Automation risks human capital traps, needs policy

Entry-level tasks are crucial for workers to accumulate human capital and advance their careers, a process the authors term 'learning-by-doing.'

Generative AI and similar technologies present a dual challenge: they can automate these foundational tasks, thereby narrowing the skill acquisition pipeline for junior workers.

Simultaneously, these technologies can complement senior managers, enabling them to expand the task frontier and potentially create new learning opportunities.

This duality generates a pecuniary externality, as individual firms and workers do not internalize the broader impact of automation and task creation decisions on aggregate human capital accumulation.

The study highlights that the effect on career dynamics and welfare depends on how the economy coordinates its learning capacity.

The study develops a continuous-time general equilibrium model incorporating endogenous automation, an endogenous task frontier, and endogenous career choice with human capital accumulation.

These components jointly determine the extent of learning-by-doing.

For economies with low learning capacity, a unique stationary equilibrium emerges, following a standard task-based automation pattern.

However, for high-capacity economies, stationary equilibria can exist in pairs.

These include a high-learning equilibrium, featuring limited automation and lower worker wages, and a low-learning equilibrium, which often results in extensive automation and a 'human-capital trap' with zero on-the-job learning.

The high-learning equilibrium consistently yields higher aggregate welfare.