EAGER: Toward Ethical Intelligent Autonomous Systems, A Case-Supported Principle-Based Behavior Paradigm

智能自主系统(IAM)即将被广泛部署在与人们密切互动的领域(例如个人助理、医疗保健、无人驾驶汽车、搜索和救援)，它们将被期望负责任地驾驭这一充满伦理色彩的景观。由于正确的道德行为不仅涉及不做某些事情，而且还涉及做某些事情以实现理想的事务状态，因此与IAM行为有关的伦理问题很可能无法得到简单、静态的解决方案，并超出其设计者的掌握。PI认为，这种制度的行为应该以明确的伦理原则为指导，这些原则是从伦理学家存在共识的特定案例中抽象出来的。为此类技术奠定基础是本次探索性研究的重点。项目成果将有助于缓解对智能自主系统的担忧，因为在现实环境中，由伦理原则指导的此类系统的行为可能比没有这一维度的此类系统的行为更容易被接受。事实上，能够以更大的自主性运行的伦理智能自主系统可能被允许在更广泛的领域帮助人类。国际和平研究所预计，这项工作的一个重要副产品是，随着这项工作的具体化，很可能会对所涉及的伦理理论有更彻底的理解。为了确保正确的道德行为，IAM应该权衡其他可能的行动，以确定在任何给定时刻哪种行为是道德上更可取的。PI将利用他之前的研究来开发和部署衡量行动的伦理偏好的原则，并证明跨多个领域的自治系统的行动选择的合理性，以开发案例支持的基于原则的行为(CPB)范例，其中，从伦理困境的特定案例中抽象出内涵定义的伦理行动偏好，并用于对具有伦理意义的行动进行排序。PI将使用老年护理机器人领域作为试验台来精炼和编码CPB，首先使用基于模拟的虚拟部件，然后在实验室环境中实现该模拟。特别是，PI将为老年护理机器人定义一套全面的具有伦理意义的行动。然后，他将通过伦理图灵测试开发和验证伦理原则，该伦理原则可用于根据伦理偏好对这套设备进行排序，并使用该原则指导模拟的Unbound Robotics UBR-1机器人的行为，该套行为位于Gazebo辅助生活设施的模拟中。接下来，他将在实验室环境中用实际的UBR-1机器人来实现这种模拟。最后，他将细化和编纂案例支持的基于原则的行为范例的需求、方法、实现细节和测试方面。除了开发和实施确保智能自主系统中伦理行为的一般方法外，这项研究还将提供证据，证明伦理原则和决策可以在机器可能与人类互动的领域进行计算和有效运作。

Intelligent autonomous systems (IAMs) are on the verge of being widely deployed in domains in which they will interact closely with people (e.g., personal assistance, healthcare, driverless cars, search and rescue), and they will be expected to navigate this ethically-charged landscape responsibly. As correct ethical behavior not only involves not doing certain things, but also doing certain things to bring about an ideal state of affairs, ethical issues concerning the behavior of IAMs are likely to elude simple, static solutions and exceed the grasp of their designers. The PI argues that the behavior of such systems should be guided by explicit ethical principles abstracted from particular cases where a consensus of ethicists exists. Laying the foundations for such technology is the focus of this exploratory research. Project outcomes will help alleviate concerns with intelligent autonomous systems, since the behavior of such systems guided by ethical principles is likely to be more acceptable in real-world environments than that of such systems without this dimension. Indeed, ethical intelligent autonomous systems capable of functioning with more autonomy might be permitted to assist human beings in a wider range of domains. The PI expects that an important byproduct of this work is that a more thorough understanding of the ethical theory involved is likely to result, as it is made concrete. To ensure correct ethical behavior, IAMs should weigh alternative possible actions against each other to determine which is ethically preferable at any given moment. The PI will leverage his previous research in developing and deploying principles that weigh the ethical preference of actions, and justify action choices for autonomous systems across multiple domains, to develop a paradigm of case-supported principle-based behavior (CPB) in which intensionally defined ethical action preference is abstracted from particular cases of ethical dilemmas and used to order ethically significant actions. The PI will refine and codify CPB using the domain of eldercare robots as a test bed by, first employing a virtual component based upon simulation and then reifying this simulation in a laboratory setting. In particular, the PI will define a comprehensive set of ethically significant actions for an eldercare robot. He will then develop and validate via an Ethical Turing Test an ethical principle that can be used to order this set by ethical preferences, and use this principle to guide the behavior of a simulated Unbounded Robotics UBR-1 robot with this set of behaviors situated in a Gazebo simulation of an assisted-living facility. He will next reify this simulation with an actual UBR-1 robot in a laboratory setting. Finally, he will refine and codify the requirements, methods, implementation specifics, and testing aspects of the case-supported principle-based behavior paradigm. Besides developing and implementing a general methodology for ensuring ethical behavior in intelligent autonomous systems, this research will provide evidence that ethical principles and decision-making can be computed and function effectively in domains where machines are likely to interact with human beings.