What AI Cannot Do

The Research Case for Human-Advantage Domains

Most AI-and-employment research focuses on what machines might replace. The EPOCH framework, developed by Loaiza and Rigobon at MIT Sloan (2024) through analysis of 19,000 tasks across 950 occupations, asks the opposite question: what human capabilities does AI not reliably replicate, and what does the labor market data say about those capabilities' economic value? The answer — grounded in historical employment data, current hiring rates, and forward BLS projections — is more specific and more hopeful than most coverage suggests.