Predictions at war: an inferential approach to decision-making in international politics

This thesis introduces the inferential actor model to International Relations to address the persistent puzzle of why leaders make sub-optimal foreign policy choices. Drawing on the neuroscience theory of active inference, it reframes decision-making as a dynamic process in which uncertainty is not simply a problem of missing information but an actor-managed variable with direct implications for the decision process. The model’s validity is tested through an extended case study of Lyndon Johnson’s Vietnam strategy (1964-65), comparing its expectations with those of two stylized alternatives: a realist model rooted in rationalist assumptions and a domestic politics model. Based on extensive archival research, the thesis finds that gradual escalation was less the product of firm conviction than a costly exercise in hypothesis testing. Johnson relied on predictions about the consequences of losing South Vietnam to sustain escalation despite low confidence in his policy options. His combination of high goal confidence and low policy confidence led him to tolerate prediction errors, attributing the South’s continued decline to tactical flaws rather than to the infeasibility of his objective. This produced a strategy characterized by exploration and experimentation in search of the elusive policy that would stabilize the South. Yet because his central goal was framed in negative terms - preventing an undesirable outcome - it became dependent on deteriorating conditions, effectively chasing an ever-shifting target. The case illustrates how actors use inferences not only to minimize uncertainty about future courses of action but also to conceptualize their goals and policies, underscoring the dual edge of inferential thinking: its necessity for strategy and its vulnerability to miscalibration of confidence.