Planning under Diagnostic Uncertainty: Question-Driven Learning in the Age of AI

Many important economic decisions depend not only on acquiring information, but on determining which questions are worth asking before committing to an action. Individuals and organizations must organize inquiry: they select questions, interpret answers, and update beliefs in ways that determine whether decisive distinctions are identified or missed. Despite its importance, the organization of inquiry is largely absent from formal economic analysis.
This paper studies decision-making under \emph{diagnostic uncertainty}, in which agents must learn not only about payoff-relevant states, but about which questions are informative. We develop a model in which agents choose questions sequentially prior to commitment, while facing uncertainty over a latent diagnostic structure that governs how questions generate answers. Before each inquiry step, agents may incur cognitive cost to refine beliefs about this diagnostic structure. The resulting problem is a finite-horizon dynamic program over inquiry, in which costly attention is allocated to belief transformations over diagnostic structure rather than directly to payoff-relevant states.
Two canonical diagnostic geometries isolate distinct planning margins. In one, value depends on locating a decisive question; in the other, on maintaining a correct sequence of questions. Both environments admit closed-form solutions and yield a common representation in which optimal inquiry increases the probability of success relative to uninformed search.
The framework identifies a distinct margin of economic behavior—planning under diagnostic uncertainty—that becomes increasingly important in environments where answers are abundant but the organization of inquiry remains scarce.