Human-AI Work Design: Skill Investment Under AI Progress

This research presents a two-period model examining how firms strategically manage human labor when deploying autonomous AI systems that may fail. The core decision revolves around how much work to allocate to AI versus keeping human workers engaged, recognizing that this choice impacts both current output and future human capital through learning and skill erosion. The model distinguishes between two dimensions of AI progress: its capability (output when functional) and its reliability (probability of functioning). In a single-firm scenario, worker engagement primarily serves as an investment in fallback capability, leading firms to engage the least-skilled workers most to close skill gaps. However, when worker mobility is introduced, the dynamic shifts. Engagement also influences labor-market sorting, as workers prefer jobs that foster valuable skill trajectories. This mobility motive can reverse the engagement pattern, directing investment towards higher-skilled workers closer to the AI frontier, where skill gains are more impactful. The paper also highlights that increased AI capability tends to raise engagement by enhancing the value of skill trajectories, while increased reliability has a more complex effect, potentially raising or lowering engagement depending on the balance between reduced fallback need and altered learning opportunities.