EAGER: Pilot Study on Bias and Trust in AI Systems

机器人和其他自主系统正在迅速扩散，并被用于公共领域，如陪伴、促进互动学习和向用户提供建议。然而，当人类与这些系统互动时，信任（包括不信任和过度信任）和偏见是如何形成的，人们知之甚少。该项目的研究人员将使用基本的现实场景来理解信任和偏见背景下的人际互动，然后将这些数据转化为量化属性。这些数据将用于开发通用算法，这些算法将有助于为机器人编程，以确保人类在与机器人交互时的安全和福祉。该项目将进行一系列试点测试，探索基础研究问题，以找到最小化人机交互潜在负面影响和后果的方法。此外，研究人员将把研究与研究生和本科生的教学和培训结合起来。该项目的目的是通过了解信任是如何建立的，以及偏见是如何影响人类感知和算法性能的，从而获得知识，为未来机器人系统的设计提供信息。作为这项工作的一部分，该团队还旨在建立基线算法，以减轻自主系统中偏见的影响。该项目的具体研究目标是量化人机交互场景中信任和偏见的影响，其中算法是基于人类专家的学习数据设计的；并开发客观减轻偏见的方法，同时仍然优化机器人的性能。试点测试的目的是促进对人机交互中的社会认知的理解。随着机器人和其他自主系统在公共领域的激增，将获得直接的社会效益。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。

Robots and other autonomous systems are proliferating rapidly, and are being used in the public domain for such purposes as companionship, facilitating interactive learning, and making recommendations to users. However, little is understood about how trust, including mistrust and over-trust, and bias develop as humans interact with these systems. The investigators of this project will use basic real-life scenarios to understand human-human interaction in the context of trust and bias, and then will translate these data into quantized attributes. This data will be used to develop general algorithms that will be useful in programming robots in ways that ensure human safety and well-being when interacting with robots. This project will conduct a series of pilot tests to explore fundamental research questions to find ways to minimize potential negative impacts and consequences of human-robot interaction. Additionally, the investigators will integrate the research with teaching and training of graduate and undergraduate students.The aim of this project is to gain knowledge that informs the design of future robot systems by understanding how trust is established and how bias impacts human perception and algorithmic performance. As part of this work, the team also aims to establish baseline algorithms that mitigate the impacts of bias in autonomous systems. The specific research objectives of this project are to quantify the impact of trust and bias in human-robot interaction scenarios where the algorithms are designed based on learned data from human experts; and to develop methods for objectively mitigating bias, while still optimizing for robot performance. The pilot tests are designed to contribute to understanding of social cognition in human-robot interaction. Direct societal befits will be gained as robots and other autonomous systems proliferate in the public domain.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.