A System Dynamics Model for Human Trust in Automation under Speed and Accuracy Requirements
A System Dynamics Model for Human Trust in Automation under Speed and Accuracy Requirements
复制标题
速度和精度要求下人类对自动化信任的系统动力学模型
DOI：
10.1177/1071181319631167
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
H. Abbass
中科院分区：
文献类型：
--
作者：
Aya Hussein;S. Elsawah;H. Abbass
研究表明，人类对自动化的信任是人类对自动化依赖的关键预测因素。已经提出了几种模型来捕捉信任和依赖之间的相互作用，以及它们对任务绩效的综合影响。有些模型假设信任受自动化可靠性的影响，而另一些模型则假设信任受自动化速度的影响。事实上，速度和可靠性对于使命性能都是至关重要的，因此，这些模型不能代表自动化速度、自动化可靠性、人的决策以及对使命性能的后续影响之间的相互关系。为了解决这一差距，我们提出了一个系统动力学模型，该模型结合了自动化的速度和可靠性及其对信任的综合影响。我们的模型明确地代表了速度-准确性的妥协，通过受试者来衡量这些方面的相对重要性，同时评估的依赖性决定。该模型的校准和评估使用从人类实验中收集的数据，其中33名受试者互动的自动化援助群体监督觅食使命。仿真结果表明，该模型可以密切复制和预测的实验数据的可信度和收集的目标数。模型的局限性和模型扩展的进一步努力进行了讨论。
Research shows that human trust in automation is a key predictor of human reliance on the automation. Several models have been proposed to capture the interplay between trust and reliance and their combined impacts on task performance. Whereas some models assume that trust is affected by automation reliability, others assume that trust is affected by automation speed. In fact, both speed and reliability can be crucial for mission performance, therefore, these models do not represent the interrelationships among automation speed, automation reliability, human decision making, and subsequent effects on mission performance. To address this gap, we propose a system dynamics model which incorporates both the speed and reliability of automation and their combined effects on trust. Our model explicitly represents the speed-accuracy compromise adopted by the subjects to weigh the perceived relative importance of these aspects while evaluating the reliance decision. The model is calibrated and evaluated using data collected from a human experiment in which 33 subjects interacted with an automated aid for swarm supervision in a foraging mission. The simulation results show that the model can closely replicate and predict the experimental data in terms of the reliance rate and the number of targets collected. Model limitations and further efforts for model extension are discussed.