Belief Update Modeling for AI Agents by Human-Derived Dual-Thresholds Evidence Accumulation Process

Autonomous drones play an increasingly critical role in dynamic environments, where effective decision-making depends on accurate and timely belief updates. However, the processes by which artificial intelligence (AI) systems revise their internal beliefs and adapt their behavior remain poorly understood, especially under conditions of uncertainty and partial observability. This study addresses these challenges by proposing a comprehensive framework to model and analyze the belief updating process in autonomous systems. The methodology integrates Bayesian inference, evidence accumulation theory, and dynamic time warping to investigate the relationship between environmental changes and AI decision-making. Key findings demonstrate that the stability of environmental score coefficients ensures robust prioritization of critical factors such as wind conditions, while the inverse relationship between environmental change intensity and belief update duration highlights the AI’s adaptability. A dual-threshold approach, informed by human cognitive patterns, captures scenarios where belief updates occur independently of observable behavior, providing deeper insights into AI cognition. Experimental validation within a simulated drone environment underscores the framework’s potential for applications such as urban search and rescue, where reliability and responsiveness are essential. This research advances the understanding of AI decision-making, offering a foundation for developing transparent, adaptive, and human-aligned intelligent systems.