CAREER: SocioCulturally Competent Agents to Study and Improve Human-AI interaction

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。人工智能（AI）系统在社会中无处不在；人们不断地与人工智能系统互动和合作来完成任务。无论是在在线虚拟环境中与人工智能代理互动，还是在虚拟助手的帮助下决定下一步去哪里吃饭，人工智能系统正越来越多地融入我们的日常生活。在如此广泛的部署下，了解一个人的环境如何影响他们与人工智能系统的互动，对于设计和开发有能力和道德的人工智能系统至关重要。人工智能系统应该能够与那些有压力或被社会结构边缘化的人进行适当的互动，也应该能够与那些没有压力或不受社会文化结构系统性边缘化影响的人进行适当的互动。这项研究将促进人们对如何创建能够适应人类差异（以及可能调解这些差异的社会结构）的人工智能系统的理解，同时也促进了人们对一个人的环境如何影响他们感知和与人工智能系统合作的方式的理解。该项目的研究与多个教育重点相辅相成，其中包括一项针对传统边缘化群体的大一新生的定向前培训计划。该项目将帮助本科生批判性地反思他们自己的环境如何影响他们设计、开发和实施人工智能系统的方式。这项工作的目标是开发一个基于过程的多层次计算理论，描述人类社会文化知识如何积极影响人工智能代理的设计、开发和交互。该项目旨在了解如何开发能够解释不同社会文化观点及其影响的有能力的人工智能代理。这项工作的产品包括扩展的计算认知架构和新的认知人工智能代理，这将对其他人工智能系统的开发有用。这些产品对于模拟人类行为的一般计算模型也很有用。该项目将产生调节人类行为的生理、情感和认知系统之间相互作用的计算模型，以及在人类与人工智能交互过程中为这些系统和过程提供某些用途的社会文化知识的解释。计算模型将在几个时间尺度上连接与人类行为和人工智能相关的领域，使人类与人工智能交互的建模、仿真和研究更加容易处理。与计算模型和工具相一致，研究者将进行研究，以更深入地了解社会文化观点和知识如何影响人们，并在任务期间与人工智能代理合作。这些研究的结果将为计算模型提供信息。这项工作将导致对调解人类与人工智能交互的过程和知识的更深入的定性和定量和理解，开发更有能力的人工智能代理，以及开源计算工具，以继续扩展这种理解。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。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).Artificial Intelligence (AI) systems have become ubiquitous in society; people constantly interact and cooperate with AI systems to accomplish tasks. Whether it is interacting with an AI agent in an online virtual environment or deciding where to eat next with the help of a virtual assistant, AI systems are increasingly integrated into our everyday lives. With such wide-spread deployment, understanding how a person’s context affects their interaction with AI systems is vital to designing and developing competent and ethical AI systems. AI systems should be able to interact as appropriately with someone who is stressed or has been marginalized by societal structures as well as it does with someone free of stress or has not been subject to socio-cultural structures that systematically marginalize them. This research will advance the understanding of how one can create AI systems that can adapt to differences in people (and the social structures that might mediate these differences), while also advancing the understanding of how a person’s context affects the way they perceive and cooperate with an AI system. The research in this project is complemented with multiple educational thrusts, including a plan to develop a pre-orientation program that targets incoming first year undergraduates from traditionally marginalized groups. The program will help undergraduate students be critically reflective how their own context affects the way they might design, develop, and implement AI systems. The goal of this work is to develop a process-based multilevel computational theory that describes how human socio-cultural knowledge can positively affect the design, development, and interaction with AI agents. The project aims to understand how one can develop competent AI agents that can account for differing socio-cultural perspectives and their effects. Products of this work include an extended computational cognitive architecture and new cognitive AI agents that will be useful for the development of other AI systems. These products will also be useful for general computational models that simulate human behavior. The project will result in a computational model of interactions between physiological, affective, and cognitive systems that modulate human behavior, as well as an account for socio-cultural knowledge that affords certain uses of these systems and processes during human-AI interaction. The computational model will connect areas related to human behavior and AI on several time scales to make the modeling, simulation, and study of human-AI interaction more tractable. In concert with the computational models and tools, the investigator will conduct studies that provide a deeper understanding of how socio-cultural perspectives and knowledge affects and is used by people while cooperating with AI agents during tasks. Results from these studies will inform the computational model. The work will result in a deeper qualitative and quantitative and under-standing of the processes and knowledge that mediate human-AI interaction, the development of more competent AI agents, and open-source computational tools to continue to expand upon this understanding.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.