Belief Updating and Delegation in Multi-Task Human-AI Interaction: Evidence from Controlled Simulations

TL;DR

The paper investigates how users form and update beliefs about a multipurpose AI across three tasks with fixed accuracies, and how these beliefs guide delegation. It demonstrates robust cross-task belief spillovers, conservative within-task updating relative to Bayesian normative steps, and delegation that is primarily driven by subjective AI accuracy beliefs rather than self-confidence. Dispositional trust and AI literacy shape initial priors, indicating stable traits influence onboarding of AI capabilities. The findings have practical implications for designing calibrated, task-specific cues and onboarding strategies to mitigate belief spillovers and reliance gaps in multi-tool AI interfaces.

Abstract

Large language models (LLMs) increasingly support heterogeneous tasks within a single interface, requiring users to form, update, and act upon beliefs about one system across domains with different reliability profiles. Understanding how such beliefs transfer across tasks and shape delegation is therefore critical for the design of multipurpose AI systems. We report a preregistered experiment (N=240; 7,200 trials) in which participants interacted with a controlled AI simulation across grammar checking, travel planning, and visual question answering, each with fixed, domain-typical accuracy levels. Delegation was operationalized as a binary reliance decision: accepting the AI's output versus acting independently, and belief dynamics were evaluated against Bayesian benchmarks. We find three main results. First, participants do not reset beliefs between tasks: priors in a new task depend on posteriors from the previous task, with a 10-point increase predicting a 3-4 point higher subsequent prior. Second, within tasks, belief updating follows the Bayesian direction but is substantially conservative, proceeding at roughly half the normative Bayesian rate. Third, delegation is driven primarily by subjective beliefs about AI accuracy rather than self-confidence, though confidence independently reduces reliance when beliefs are held constant. Together, these findings show that users form global, path-dependent expectations about multipurpose AI systems, update them conservatively, and rely on AI primarily based on subjective beliefs rather than objective performance. We discuss implications for expectation calibration, reliance design, and the risks of belief spillovers in deployed LLM-based interfaces.

Figures (10)

• Figure 2: Observed vs. normative belief updates under the strict CF policy (H2). Each point is one trial-level update (demeaned within participant–task). The slope ($\hat{\sigma}\approx 0.52$) indicates conservatism: participants updated in the correct direction, but only about half as much as Bayes’ rule prescribes. Appendix \ref{['app:h2_robust']} shows analogous plots for lenient and hybrid policies.
• Figure 3: Appendix H2. Observed vs. normative belief updates under alternative CF policies. All slopes $\hat{\sigma}\approx 0.47$–0.49, indicating conservatism.
• Figure 4: Appendix H2. Distribution of individual updating slopes ($\hat{\sigma}$) across CF policies. Most participants under-react relative to the Bayesian benchmark ($\sigma=1$).
• Figure 5: Appendix H5. Distributions of trial-1 priors (0--100) by task (grammar, travel, VQA). Points show medians; boxes show IQR.
• Figure : (a) Grammar Error Detection