CAREER: Virtual Reality Simulations of Moral Decision Making for Autonomous Vehicles

自动驾驶汽车（AVs）的事故证明，需要考虑自动驾驶汽车使用中固有的道德问题，以及它们将被编程遵循的道德目标函数。与自动驾驶汽车相关的工作一直被实用主义模型所主导，无论是在算法上还是在关于自动驾驶汽车的公众调查框架上。然而，由于方法论问题和公众对功利主义思想的明显抵制，有必要提供替代方法，充分捕捉人类道德判断和决策的灵活性。该项目利用代理-行为-后果（ADC）模型和对人类道德决策的研究，利用虚拟现实（VR）和模拟技术，彻底改变伦理人工智能（AI）技术的发展。该项目旨在通过与北卡罗来纳州立大学图书馆的虚拟现实实验室、北卡罗来纳州立大学交通研究与教育研究所（ITRE）、北卡罗来纳州专业工程师（PENC）和北卡罗来纳州科学与数学学院（NCSSM）合作，改善高中和大学教育、跨学科科学研究和交通工程。它有助于(i)提高公众的科学素养和公众对科学技术的参与；（二）妇女、残疾人和代表性不足的少数民族更多地参与研究；（iii）改善科学、技术、工程和数学（STEM）教育和教育工作者的发展；（四）提高个人在社会中的福利。ADC模型借鉴了伦理学、实验哲学、社会心理学和决策神经科学的见解，通过将道德判断和决策分解为关于代理人、行为和后果的积极或消极的直觉评估，并在任何特定情况下将其指定为“高风险”或“低风险”，来解释道德判断和决策。这些直观的评价结合起来产生对道德可接受性的积极或消极的判断。例如，如果行为人有良好的意图（例如，规避危险），并且后果良好（例如，避免了事故），那么在某人做出负面行为（例如，在整条线上改变车道）的情况下，总体道德判断将会减轻。ADC模型提供了表示这种情况的公式（例如，[A+], [D-] & [C+] = [MJ+]或积极的道德判断）。该项目开发了涉及交通（包括车辆和行人）的潜在有害决策的新道德情境，并对高中生、职业道德哲学家和未接受过正式道德培训的非专业人士的样本进行了道德判断测试。在沉浸式虚拟现实环境中，道德情境被编程并记录为动画，并在高中生、大学生和伦理学家中进行测试。该项目最终以基于ADC模型的道德算法原型告终，该模型在高保真交通模拟器的模拟视觉环境中与人类决策者进行交互。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Accidents with Autonomous Vehicles (AVs) attest to the need to consider the ethical issues inherent in AVs use, and the ethical goal functions they will be programmed to follow. Work with AVs has been dominated by utilitarian models both in algorithms and framing public surveys regarding AVs. However, due to methodological issues and a pronounced resistance of the general public towards utilitarian thinking, there is a need to offer alternative approaches that adequately capture the flexibility of human moral judgment and decision making. This project utilizes the Agent-Deed-Consequence (ADC) model and research on human moral decision making to revolutionize the development of ethical artificial intelligence (AI) technologies with the use of virtual reality (VR) and simulations. The project aims to improve high-school and college education, interdisciplinary scientific research, and transportation engineering by partnering with the NC State University Libraries’ Virtual Reality Lab, NC State’s Institute for Transportation Research and Education (ITRE), Professional Engineers of North Carolina (PENC), and the North Carolina School of Science and Mathematics (NCSSM). It contributes to (i) increased public scientific literacy and public engagement with science and technology; (ii) increased participation of women, persons with disabilities, and underrepresented minorities in research; (iii) improved Science, Technology, Engineering and Mathematics (STEM) education and educator development; and (iv) improved well-being of individuals in society.The ADC model draws on insights from ethics, experimental philosophy, social psychology and decision neuroscience to explain moral judgments and decisions by breaking them down into positive or negative intuitive evaluations concerning the Agent, the Deed, and the Consequence, and designating them as ‘high-’ or ‘low-stakes’ in any given situation. These intuitive evaluations combine to produce a positive or negative judgment of moral acceptability. For example, the overall moral judgment in a situation in which someone committed a deed that is judged as negative (e.g., changing lanes on a full line) would be mitigated if the agent had good intentions (e.g., circumventing a hazard) and the consequences were good (e.g., an accident was avoided). The ADC model provides formulas for representing such situations (e.g., [A+], [D-] & [C+] = [MJ+] or positive moral judgment). This project develops new moral situations about potentially harmful decisions involving traffic, including vehicles and pedestrians, and tests moral judgment on samples of high-school students, professional moral philosophers, and lay people with no formal training in ethics. The moral situations are programmed and recorded as animations in an immersive VR setting, and tested in populations of high school and college students as well as ethicists. The project culminates with a prototype ethical algorithm based on the ADC model which interacts with human decision makers in a simulated visual environment in the high-fidelity traffic simulator.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.