Future Everyday Interaction with the Autonomous Internet of Things

该项目旨在研究未来自主物联网（A-IoT）的交互机制和用户界面的设计：这是一个互联设备系统，超越了物联网的大多数当前化身，将自主或自动化方面作为关键特征。A-IoT的新生实例包括智能恒温器，它可以根据用户的存在及其日常生活来学习自动控制中央供暖系统，以及洗衣机，它可以在用完时订购洗涤剂。换句话说，这种A-IoT可以主动响应感测到的环境变化，有效地代表用户进行工作，并承诺更有效地利用资源（例如使用更少的能源进行加热）或增加便利性（例如始终提供清洁剂）。物联网产生的数据财富（或“洪水”）可能会不断增长，超出人类的能力，无法将其转化为可以采取行动的有意义的信息。因此，未来的交互系统将越来越多地支持将大型复杂数据的粒度决策委托给自主计算代理，使用户能够就其一般需求和舒适度做出明智的选择。在自主物联网中;数据和决策将部分由设备及其软件“主动”管理，利用机器学习技术和优化算法。然而，最近的研究调查了商业智能恒温器在现实世界中的接受程度，强调了错误、系统操作的有限易读性和过度的用户期望是如何导致沮丧并导致一些用户放弃这项技术的。我们之前的工作表明，人们不信任潜在的智能能源基础设施，因为缺乏对自主技术的所有权、意图和允许活动的问责。这些结果表明，A-IoT系统的设计需要解决几个挑战，以使其负责任;包括在系统方面，设计自主决策，以考虑到偶然人类行为的不确定性;在用户方面，需要使这些系统在日常生活中清晰和可用。事实上，在使系统的操作清晰和不使用户对人工智能算法的技术复杂性感到不知所措之间存在着内在的紧张关系。到目前为止，设计此类系统的方法相当稀疏，并且不特定于A-IoT系统（跨越HCI，AI和Ubicomp），因此需要更专注的方法来确定A-IoT系统实施的核心设计原则和方法。因此，我们的目标是在家庭日常环境中建立用户与A-IoT系统交互的科学基础，旨在阐明以下研究问题：在日常环境中，用户在多大程度上愿意将代理委托给A-IoT系统？如何设计与A-IoT系统的交互，以支持而不是阻碍用户的日常活动？智能代理管理这样的A-IoT系统所必需的功能是什么？我们怎样才能设计这样的系统，让用户既能委派控制权，又能轻松地重新获得控制权？除非这些问题得到充分解决，否则A-IoT系统很可能会让用户感到沮丧，导致时间和资源的大量浪费。因此，我们将通过多种技术的组合来应对这些挑战，包括研究现有实践，迭代开发新的A-IoT原型并对其进行野外评估。这种多学科的方法是由一个团队，汇集了国际领先的研究人员在人机交互，人工智能和设计民族志。

This project seeks to investigate the design of interaction mechanisms and user interfaces for a future Autonomous Internet of Things (A-IoT): a system of interconnected devices that reaches beyond most current incarnations of the IoT to include aspects of autonomy or automation as a key feature. Nascent instantiations of the A-IoT range from smart thermostats that learn to autonomously control central heating systems based on the presence of users and their routine, to washing machines that order detergent for delivery when it runs out. In other words, this A-IoT can proactively respond to sensed environmental changes, effectively doing work on behalf of users, with the promise of a more efficient use of resources (e.g. to use less energy for heating) or increased convenience (e.g. to always have detergent available). The wealth (or "deluge") of data produced by the IoT is likely to keep growing beyond human capacity to turn it into meaningful information that can be acted on. Therefore, it will require future interactive systems to increasingly support the delegation of granular decision making over large and complex data to autonomous computational agents, allowing users to make informed choices about their general needs and comfort. In an Autonomous IoT; data and decisions will be, in part, 'actively' managed by the devices and their software, drawing upon machine learning techniques and optimization algorithms.However, recent studies examining the real-world acceptance of a commercial smart thermostat highlighted how errors, limited legibility of the system operation, and excessive user expectations caused frustration and led to some users abandoning the technology. Our own prior work revealed people distrust a potential smart energy infrastructure due to lack of accountability of the ownership, intent, and permitted activities of the autonomous technology. These results suggest that the design of A-IoT systems needs to address several challenges to be made accountable; including, on the system side, designing autonomous decision-making to take into account the uncertain nature of contingent human behaviour; and on the user side, the need to make these systems legible and usable in everyday life. Indeed there is an inherent tension between making a system's operation legible and not overwhelming users with the technical complexity of artificial intelligence algorithms. To date, the methodologies to design such systems are rather sparse and not specific to A-IoT systems (spanning HCI, AI, and Ubicomp) and hence a more focused approach is required to determine the core design principles and methods for the implementation of A-IoT systems.Our goal is thus to establish the scientific underpinnings of user interactions with A-IoT systems, in a domestic everyday context, with the aim of elucidate the following research questions: to what extent may users be willing to delegate agency to A-IoT systems in everyday contexts? How should interactions with A-IoT systems be engineered to support rather than hinder users' daily activities? What capabilities are essential for intelligent agents to manage such A-IoT systems? How can we design such systems so that they allow users to delegate control, yet easily regain it? Unless such questions are fully addressed, A-IOT systems are likely to frustrate users, resulting in significant waste of time and resources.Hence, we will address these challenges through a combination of techniques, including the study of existing practices, the iterative development of novel A-IoT prototypes and their evaluation in-the-wild. Such a multidisciplinary approach is made possible by a team that brings together internationally-leading researchers in human-computer interaction, artificial intelligence and design ethnography.