AI Institute for Collaborative Assistance and Responsive Interaction for Networked Groups (AI-CARING)

人们在工作、家庭和社会环境中相互协作，这些互动会随着时间的推移而变化，这取决于群体中的能力、角色、责任、规范和人际关系。人类-人工智能交互（HAI）系统可以通过提供关于小组成员的状态、背景和需求的及时信息，并代表他们与其他此类人工智能系统进行交互，从而在管理小组协作方面提供帮助。老年人的家庭护理领域是一个复杂的辅助设置激发本研究的主要例子。老年人、家庭照顾者、医疗专业人员、朋友和邻居经常合作应对不断变化的需求。为了在这种情况下提供帮助，人工智能系统需要：(a)通过整合许多感官模式的数据，对人们的身体、心理和社会能力和需求进行建模；(b)检测用户能力和需要在生理、认知、社会和心理方面的变化；(c)了解整个支助网络的动态关系和能力；(d)调整交互行为，以便最有效地帮助用户。该项目将开发人类与人工智能交互的方法，学习人类行为的个性化模型，以及它们如何随着时间的推移而变化，并利用这些知识更好地协作、沟通和帮助用户。为了推动这些创新，该研究所将成为学术界和工业界合作努力的连接点。除了先进的研究外，这些合作将通过广泛的劳动力发展、教育、推广、扩大参与和知识转移计划，积极培养下一代人才，培养多样化、训练有素的劳动力，这些计划旨在传播有关交互式人工智能系统开发的知识和热情。网络群体协同援助和响应式交互AI研究所（AI- caring）将开发一门专注于个性化、纵向、协作式AI的学科，其特点是设计、开发和部署嵌入用户社区的交互式智能AI系统，并持续一段时间（数月或数年）。设想中的人工智能系统将采用嵌入普通消费设备（如手机、智能扬声器）的虚拟助手的形式，通过语音、手势、视觉、听觉和混合现实界面与用户进行交互。人工智能系统将建立用户能力、目标、价值观和人际关系的个性化纵向模型，该模型基于汇总的传感器观察和过去互动的历史。在这些模型的基础上，联网的代理团队将根据用户的个人和社会规范，通过个性化和价值驱动的互动提供协调一致的援助。计算、社会科学和医疗保健领域的研究人员将合作设计、开发和部署AI系统，其中包括用于用户建模和个性化的样本高效技术、用于纵向人类-人工智能团队的稳健方法、具有社会意识和尊严保护的人工智能方法、可解释的系统、实验设计的新指南以及这些领域的新基准和指标。共同设计方法、研究示范和长期实地评估将涉及家庭（配备不同类型的传感器），其中包括有认知和身体障碍的老年人、他们的家人、非正式照顾者、专业保健提供者和社区伙伴。人工智能护理系统将加强日常生活，识别行为变化，为护理人员提供团队支持，为与专业人员的互动提供支撑计划，并就个人的不同能力提供道德鼓励和反馈。这些基本功能将支撑响应性和个性化的人机交互，从而改变我们与人工智能系统的日常体验。这项工作的长期影响将超越护理，扩展到任何包括通过语音、手势、视觉和混合现实界面进行长期人机交互的应用程序。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。

People collaborate with one another in work, home, and social settings, and these interactions change over time based on the capabilities, roles, responsibilities, norms, and interpersonal relationships of those in the group. Human-AI Interaction (HAI) systems can provide assistance in managing group collaborations by providing timely information about the status, context, and needs of group members, and by interacting on their behalf with other such AI systems. The area of home care for aging adults is a prime example of a complex assistive setting inspiring this research. Older adults, family caregivers, medical professionals, friends and neighbors often collaborate to respond to changing needs. To assist in such settings, HAI systems need to: (a) model the physical, mental, and social capabilities and needs of people by integrating data across many sensory modalities; (b) detect physical, cognitive, social and psychological changes in user capabilities and needs; (c) understand the dynamic relationships and capabilities across the support network; and (d) adapt interactive behaviors in order to assist the user most effectively. This project will develop approaches in human-AI interaction that learn personalized models of human behavior and how they change over time, and use that knowledge to better collaborate, communicate, and assist the user. To drive these innovations, the Institute will serve as a nexus point for collaborative efforts across academia and industry. In addition to advanced research, these collaborations will actively build the next generation of talent for a diverse, well-trained workforce through a wide range of workforce development, education, outreach, broadening participation, and knowledge transfer programs designed to disseminate knowledge about, and enthusiasm for, the development of interactive AI systems.The AI Institute for Collaborative Assistance and Responsive Interaction for Networked Groups (AI-CARING) will develop a discipline focused on personalized, longitudinal, collaborative AI -- characterized by the design, development, and deployment of interactive, intelligent HAI systems embedded within communities of users over extended periods of time (months and years). Envisioned HAI systems will take the form of virtual assistants embedded in common consumer devices (e.g., cell phones, smart speakers) that will interact with users via speech, gesture, visual, auditory, and mixed reality interfaces. HAI systems will establish personalized longitudinal models of user abilities, goals, values, and interpersonal relationships based on aggregated sensor observations and the history of past interactions. Building on such models, networked teams of agents will provide coordinated assistance through personalized and value-driven interactions that operate in accordance with users’ personal and social norms. Researchers in computing, social sciences, and healthcare will collaborate to design, develop, and deploy HAI systems that include sample-efficient techniques for user modeling and personalization, robust methods for longitudinal human-AI teaming, socially-conscious and dignity-preserving AI methodologies, explainable systems, novel guidelines for experimental design, and novel benchmarks and metrics for these areas. Co-design approaches, research demonstrations and long-term field evaluations will involve households (instrumented with different types of sensors) that include older adults with cognitive and physical impairments, their family, informal caregivers, professional health providers and community partners. AI-CARING systems will reinforce daily routines, recognize changes in behavior, provide team support for caregivers, scaffold planning for interactions with professionals, and provide ethical encouragement and feedback regarding an individual's varying abilities. These fundamental capabilities will scaffold responsive and personalized Human-AI Interaction that will transform our day-to-day experiences with AI systems. The long-term impact of this work will go beyond caregiving, extending to any application that includes long-term Human-AI Interaction through speech, gesture, visual and mixed reality interfaces.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.