EAGER: A Mathematical Framework for Increasing Trust in Human-Machine Interactions

1548616（Jain）几十年来，自动化在航空和航天飞行以及制造业和加工厂等工业环境中发挥了关键作用。今天，人类每天都与自动化系统互动，包括在家中（例如机器人真空吸尘器和家庭监控系统）和在车辆中（例如导航系统）。自动化的作用是改善人类的状况，提高我们的生活质量。然而，当人类由于对机器根本缺乏信任而推翻自动决策时，自动化的好处有时就会丧失。研究表明，近80%影响生命和环境的重大事故与错误有关，其中一些可能源于对自动化的信任问题。因此，我们有必要研究人机交互，以便未来的自动化系统不仅更安全，而且更有效地促进人类的健康和繁荣。这个研究项目的目标是让机器具备情商。也就是说，我们相信，如果机器有能力感知和处理与之互动的人的情感，特别是信任，那么人与机器就能更好地合作，实现共同的目标或完成给定的任务。为了实现这一愿景，机器有必要能够1)测量人类的情绪状态，2)通过改变用户界面（UI）对人类做出相应的反应。我们将研究使用不同的心理生理学工具来进行这些测量，并利用数学工具来开发模型，以捕捉人类如何对机器ui中的视觉变化做出情感反应。这项研究的结果将使我们能够重新设计显示情商的机器ui。这个研究项目是两个教员之间的合作，一个是控制和自动化方面的专家，另一个是设计和心理学方面的专家。通过弥合这些学科之间的差距，pi将能够促进动力系统和控制、设计和社会科学领域的研究人员之间的对话和未来的研究参与。这是至关重要的，因为人机交互的研究本质上需要传统工程学科和社会科学的工具和方法。对于年轻的工程师来说，理解跨学科研究的价值和好处，并在其他领域的背景下理解自己的学科也很重要。为此，pi将向工程本科生展示该项目的成果，让他们接触到通过看似不同的研究领域的合作来改善人类状况的前沿研究。这也符合通过吸引包括女性和少数民族在内的更广泛的学生群体来提高STEM领域参与度的国家目标。随着社会各方面自动化程度的提高，人类越来越多地被取代为各种角色（如飞机飞行员和工厂操作员）的主要决策者。然而，人类仍然有能力推翻自动决策，当人类由于对机器缺乏信任而推翻自动决策时，就会出现一个重大问题。在被广泛称为“人类代理集体”（Human-Agent Collectives）的情况下，我们期望看到人类和机器之间对合作和信任的需求日益增长，以完成大量任务。因此，机器感知和响应人类的方式是特别重要的。本研究的目的是了解自动化系统中机器用户界面（ui）与人类信任之间的动态关系。通过这项研究获得的新知识将使我们能够通过最终重新设计用户界面来改善人机交互。这个修改后的用户界面将包括一个情感智能系统，让机器实时对人类做出反应。本研究有两个具体目标。第一个目标是对信任的实时测量进行动态表征。这样的测量目前还不存在，但为了让机器感知与它们互动的人类的信任程度，这种测量是必要的。我们建议确定一个动态的信任模型，该模型依赖于实时心理生理测量，如皮肤电反应（GSR）和脑电图（EEG），以及眼动追踪。第二个目标是定义一个数学框架来模拟人类对机器的情绪反应。机器通过其用户界面（UI）中的各种设计特征与人类交流。我们建议进行一项人类受试者研究，以数学方式表征机器如何使用特定的机器UI特征来动态改变人类对机器的信任。由于使用了心理生理学传感器，与以前开发的信任模型相比，这些模型将更广泛地适用于收集数据的人类受试者研究的特定背景。这个跨学科的项目将促进动态和控制、设计和社会科学社区的知识和理解。通过提出的研究，我们将能够设计一个闭环情商系统，以实现改善人与机器之间关系的总体目标，从而使一系列自动化系统更加可靠和高效地运行。此外，pi将在动力学/控制和机械设计会议上共同组织以人机交互为中心的活动，以鼓励这些不同领域的对话和研究合作。最后，在教育推广方面，pi将向机械工程本科生传播这项研究，使他们了解跨学科研究的兴奋和好处；通过扩大机械工程的品牌，以吸引包括女性和少数民族在内的更广泛的学生，这直接支持了ASME对2030年机械工程师的愿景。

1548616(Jain)For decades automation has played a critical role in air and space flight as well as in industrial settings such as manufacturing and processing plants. Today, humans interact with automated systems every day including in their home (e.g. robotic vacuums and home monitoring systems) and in their vehicles (e.g. navigation systems). The role of automation is meant to improve the human condition and improve our quality of life. However the benefits of automation are sometimes lost when humans override an automated decision due to a fundamental lack of trust in the machine. Research has shown that close to 80% of the major accidents that have affected lives and the environment are linked to errors, some of which may have originated from trust issues in automation. Therefore, it is necessary for us to study human-machine interactions so that the automated systems of tomorrow are not only safer but more effective in promoting the health and prosperity of humans. The goal of this research project is to enable emotional intelligence in machines. That is to say, we believe that if a machine has the ability to sense and process the feelings, particularly, trust, of the human interacting with it, the human and machine can better cooperate to achieve a common goal or complete a given task. In order to realize this vision, it is necessary for machines to be able to 1) measure a human's emotional state and 2) respond accordingly to the human through changes in its user interface (UI). We will study the use of different psychophysiological tools to make these measurements as well as utilize mathematical tools to develop models that capture how humans respond emotionally to visual changes in machine UIs. The results of this study will then enable us to redesign machine UIs that exhibit emotional intelligence. This research project is a collaboration between two faculty, one with an expertise in control and automation and another with an expertise in design and psychology. By bridging the gap between these disciplines, the PIs will be able to facilitate a dialogue and future research engagement between researchers in the fields of dynamical systems and controls, design, and the social sciences. This is critical because the study of human-machine interactions inherently requires tools and methods from both traditional engineering disciplines as well as the social sciences. It is also important for young engineers to understand the value and benefits of interdisciplinary research and to understand their own discipline in the context of other fields. To that end, the PIs will present the results of this project to engineering undergraduates to expose them to cutting edge research that is rooted in improving the human condition through collaboration across seemingly disparate fields of study. This also aligns with national goals of increasing participation in STEM fields by appealing to a wider demographic of students, including women and minorities. With increasing automation in all aspects of society, humans are increasingly being displaced as the primary decision-maker in various roles (such as aircraft pilots and plant operators). However, humans still have the ability to override automated decisions, and a significant problem arises when the human overrides an automated decision due to a fundamental lack of trust in the machine. In what are broadly being called Human-Agent Collectives, we expect to see a growing need for cooperation and trust between humans and machines to accomplish a large range of tasks. Therefore, the way in which the machine senses and responds to the human is of particular importance. The objective of the proposed research is to understand the dynamic relationship between machine user interfaces (UIs) and human trust in automated systems. The new knowledge gained through this research will enable us to improve human-machine interactions by ultimately redesigning the user interface. This modified user interface will include an emotional intelligence system for the machine to respond to the human in real-time. Two specific aims guide this research. The first aim is to conduct a dynamical characterization of real-time measurements of trust. Such measurements do not currently exist and are necessary in order to allow machines to sense the trust level of the humans that they are interacting with. We propose to identify a dynamic model of trust that relies on real-time psychophysiological measurements such as galvanic skin response (GSR) and electroencephalography (EEG), as well as eye-tracking. The second aim is to define a mathematical framework for modeling human emotional response to machines. Machines communicate with humans through various design features in their user interface (UI). We propose to conduct a human subjects study to mathematically characterize how specific machine UI features can be used by a machine to dynamically change human trust in the machine. Due to the use of psychophysiological sensors, these models will be more widely applicable as compared to previously developed trust models that were specific to the context of the human subject studies from which the data was collected. This cross-disciplinary project will advance knowledge and understanding within the dynamics and control, design, and social science communities. Through the proposed research, we will enable the design of a closed-loop emotional intelligence system that achieves the overarching goal of improving the relationship between human and machine, thereby leading to more reliable and efficient operation of a range of automated systems. Additionally, the PIs will co-organize events centered on human-machine interaction at both dynamics/control and mechanical design conferences to encourage a dialogue and research collaborations across these otherwise disparate fields. Finally, in educational outreach, the PIs will disseminate this research to mechanical engineering undergraduates to expose them to the excitement and benefits of interdisciplinary research; this directly supports the ASME's vision for the mechanical engineer of 2030 by broadening the brand of mechanical engineering to appeal to a wider demographic of students, including women and minorities.