Tasks given to Participants of an Adaptive Trust-Calibrated Human–AI Collaboration for Enhanced System Usability

The task dataset used in this study comprises 120 decision-support tasks designed to reflect realistic Human–AI collaboration scenarios under varying levels of complexity and uncertainty. Each task is characterized by a unique identifier, task type (risk assessment, classification, decision support, or what-if analysis), difficulty level (low, medium, high), simulated AI reliability condition (high, medium, low), and an associated ground-truth label to enable objective performance evaluation. The dataset was systematically constructed to expose participants to diverse interaction conditions, including cases where AI recommendations are reliable and others where uncertainty is deliberately introduced, thereby enabling observation of over-reliance and under-reliance behaviors. By explicitly modeling task difficulty and AI reliability, the dataset supports controlled analysis of trust calibration, cognitive workload, and decision accuracy, and provides a reproducible foundation for evaluating adaptive Human–AI collaboration frameworks without reliance on sensitive real-world data.本研究使用的任务数据集包含120个决策支持任务，旨在反映不同复杂度与不确定性水平下的真实人机协作(Human–AI collaboration)场景。每项任务均配有唯一标识符、任务类型（风险评估、分类、决策支持或假设分析(what-if analysis)）、难度等级（低、中、高）、模拟AI可靠度条件（高、中、低），以及用于开展客观性能评估的关联真实标签(ground-truth label)。该数据集经系统性构建，可让参与者接触到多样化的交互场景，涵盖AI推荐可靠的情形与刻意引入不确定性的场景，从而能够观测到过度依赖与依赖不足的行为表现。通过显式建模任务难度与AI可靠度，该数据集支持对信任校准、认知负荷(cognitive workload)与决策准确性开展控制性分析，并可为评估自适应人机协作框架提供可复现的基础，无需依赖敏感的真实世界数据。