Essays on the Economics of Digitisation and AI

Abstract

Digital technologies and artificial intelligence are reshaping decision-making, work, and wellbeing. Across three essays, this dissertation examines a common question: how should digital tools be designed and deployed so that they improve outcomes for the people and organisations that use them? The first essay studies how AI-generated information should be presented to human decision-makers who deviate from Bayesian updating. It introduces calibrated coarsening as a framework and tests it in a randomised experiment with professional loan specialists, showing that carefully coarsened AI signals can improve decisions relative to both human-only and uncoarsened-AI benchmarks. The second essay examines the rollout of generative-AI pre-labelling tools inside a large data annotation firm. It documents that productivity gains concentrate on simpler tasks, human time reallocates from labelling (lower layer in hierarchy) to quality control (higher layer in hierarchy), with implications for organisational policy and reallocation of tasks. The third essay uses a natural field experiment with 13 million social media users to study digital-wellness commitment devices, documenting low take-up at scale relative to earlier small-scale studies. Together, the essays show that the benefits of digital tools depend on the design of the interfaces, organisations, and interventions through which they are deployed.