Hope, Signals, and Silicon: A Game-Theoretic Model of the Pre-Doctoral Academic Labor Market in the Age of AI

Abstract

Generative AI can make early research work easier to produce and harder to interpret. This paper develops a compact game-theoretic model of this production evaluation tension in the pre-doctoral academic labor market. In the model, PIs organize RA labor, allocate AI between routine and novel tasks, and choose mentoring intensity. RAs choose effort, while admissions committees infer research potential from noisy task-level signals under fixed admissions capacity. A mechanism-preserving simulation examines whether the model's qualitative mechanisms continue to hold when RAs and PIs are heterogeneous, research outcomes partly depend on luck, admissions evaluation is noisy, and elite Ph.D. capacity is fixed. The analysis yields three implications. First, routine task AI can increase observable routine output while reducing the diagnostic precision of routine evidence. Second, heterogeneous PI objectives and task complementarity can lead laboratories to adopt different AI strategies, with some emphasizing scalable routine production and others emphasizing mentoring and novel-task augmentation. Third, when elite Ph.D. capacity is fixed, broad improvements in visible records can raise admissions cutoffs rather than expand access proportionally. The simulation reinforces these mechanisms by showing that AI can raise routine output while weakening the link between evaluated scores and latent ability, increasing the risk that high-ability or high-realized-merit candidates are missed. The paper suggests that as routine evidence loses diagnostic content, evaluation should place greater weight on less easily automated forms of contribution, including judgment, interpretation, research design, and process-based evidence.