The Transparency Paradox in Explainable AI: A Theory of Autonomy Depletion Through Cognitive Load

TL;DR

The paper tackles the transparency paradox in explainable AI by modeling human autonomy as a dynamic stochastic process $A_t$ whose drift is modulated by information $I_t$ and cognitive load. It develops a stochastic-control framework, solving via Hamilton-Jacobi-Bellman equations, and validates predictions with Monte Carlo simulations. Five testable predictions on disengagement timing, working-memory moderation, and optimal information levels are derived, showing that adaptive transparency policies outperform static approaches. The findings yield design principles for real-time, capacity-aware AI explanations, enabling personalized and context-sensitive transparency that preserves autonomy while improving decision quality.

Abstract

Objective: This paper develops a theoretical framework explaining when and why AI explanations enhance versus impair human decision-making. Background: Transparency is advocated as universally beneficial for human-AI interaction, yet identical AI explanations improve decision quality in some contexts but impair it in others. Current theories--trust calibration, cognitive load, and self-determination--cannot fully account for this paradox. Method: The framework models autonomy as a continuous stochastic process influenced by information-induced cognitive load. Using stochastic control theory, autonomy evolution is formalized as geometric Brownian motion with information-dependent drift, and optimal transparency is derived via Hamilton-Jacobi-Bellman equations. Monte Carlo simulations validate theoretical predictions. Results: Mathematical analysis generates five testable predictions about disengagement timing, working memory moderation, autonomy trajectory shapes, and optimal information levels. Computational solutions demonstrate that dynamic transparency policies outperform both maximum and minimum transparency by adapting to real-time cognitive state. The optimal policy exhibits threshold structure: provide information when autonomy is high and accumulated load is low; withhold when resources are depleted. Conclusion: Transparency effects depend on dynamic cognitive resource depletion rather than static design choices. Information provision triggers metacognitive processing that reduces perceived control when cognitive load exceeds working memory capacity. Application: The framework provides design principles for adaptive AI systems: adjust transparency based on real-time cognitive state, implement information budgets respecting capacity limits, and personalize thresholds based on individual working memory capacity.

Figures (4)

• Figure 1: Monte Carlo Validation of Theoretical Predictions. (A) Mean trajectories match theoretical exponential growth within 2% error. (B) Variance growth is monotonically increasing as predicted. (C) Disengagement timing matches Proposition 3 within 7% across information levels. (D) Working memory effects show predicted monotonic relationship, with approximately 0.76 time units per WM item. (E) Q-Q plot confirms log-normal distribution of autonomy ($R^2 = 0.99$). Validation summary confirms all five predictions passed.
• Figure 2: Optimal Control Policy $u^*(A, I, t=5)$ Exhibits Threshold Structure. High-autonomy and low-information regions (upper left) indicate maximum information provision ($u^* = u_{\max}$). Low autonomy and high information regions (lower right) recommend ceasing information provision ($u^* = 0$) to preserve cognitive resources.
• Figure 3: Value Function $V(A, I, t=0)$ Exhibits Inverted-U Shape in Information Dimension. The value function peaks at moderate information levels ($I \approx 1.5$), consistent with the transparency paradox. The maximum value occurs at high autonomy with moderate information; the minimum value occurs at boundary autonomy with excessive information.
• Figure 4: Mean Autonomy Trajectories Under Three Transparency Policies. Optimal control (solid line) maintains autonomy while achieving the highest decision quality. Maximum transparency (dashed) depletes autonomy rapidly, leading to high disengagement rates. No transparency (dotted) preserves autonomy but fails to provide decision-relevant information.