Newcomb's Paradox Under Epistemic Uncertainty: A Two-Stage Policy Analysis

Newcomb’s paradox is still debated despite its deceptively simple setup. Here we introduce a two-stage policy framework parametrized by pre-entry and in-room 1-boxing propensities (α,β), predictor error rate ε, and uncertainty about the predictor’s timing (λ). An expected utility analysis shows that 1-boxing dominates across the joint uncertainty space of these parameters (except at the extremes), even under high risk aversion. Thus, we argue that the paradox is fundamentally epistemic: the rational response depends on the agent’s beliefs about predictor accuracy and about which stage of deliberation the predictor targets. We suggest that 2-boxers either believe that the predictor Omega is unreliable, or that the choice in-room is completely uncaused (i.e. unpredictable). If such a belief is due to a valuation of one’s own free will equivalent to or above $1,000, then we argue that even under maximin reasoning, 1-boxing is optimal as being wrong about one’s own deterministic nature is far more costly for 2-boxers than 1-boxers.