Dual-process theories posit that fast, intuitive System 1 often produces bias, whereas slow, deliberative System 2 corrects it. This manuscript challenges that traditional view by synthesizing theory and evidence on motivated System-2 reasoning (MS2R)—the tendency for deliberation to be co-opted in defense of identity-relevant beliefs. Across domains (e.g., climate risk, vaccines, gun policy), polarization has been shown to increase with subject-matter knowledge or numeracy when issues implicate social identity, whereas the same abilities improve accuracy for identity-neutral problems. Six complementary mechanisms are reviewed to explain these effects: (a) confirmation bias and motivated skepticism; (b) identity-protective cognition; (c) belief as value-signaling (expressive utility); (d) cognitive dissonance and conflict rationalization; (e) the argumentative theory of reasoning; and (f) bounded/resource-rational approaches (e.g., BIASR) that yield selective trust as an efficient but polarizing heuristic. Boundary conditions and critiques are then considered. High-powered replications indicate a robust directional motivated-reasoning main effect but limited evidence that higher ability amplifies bias uniformly; design features (ambiguity, incentives, audience visibility) and affect-first gating appear pivotal. Identity-congruent responding is further reinterpreted as instrumentally rational signaling under reputational and career incentives, especially for elites. Finally, actionable countermeasures are proposed: accuracy prompts and incentives, self-affirmation and identity-safe framing, symmetric-scrutiny protocols (e.g., steel-manning), probabilistic forecasting with feedback, clarity-enhancing data displays and preregistered criteria, and prebunking. Research prospects include sharper operationalization separating ability, disposition, and state engagement; process-tracing and psychophysiological markers to adjudicate mechanisms; and preregistered, multi-lab designs that vary ambiguity, incentives, and audience. Educational applications emphasize embedding accuracy nudges, forecasting tournaments, and structured adversarial collaboration to realign System 2 from identity performance toward error detection.
Hongyang Liu (Sat,) studied this question.