CRII: CHS: Experimental Studies of Human Trust in Machine Learning

机器学习(ML)已被应用于金融、医疗保健、城市运营以及定向新闻和广告等多个领域。他们取得了巨大的成功，从海量数据中发现了洞察力，并推动了决策。尽管应用广泛，但对ML方法中人们的信任或缺乏信任缺乏科学的理解，而人机协作的成功需要对信任的更深层次的理解。本课题通过大规模的随机人体实验，促进了外行人对ML信任的理解。该项目将产生关于人与ML系统之间信任的形成和维护的理论见解，以及设计人们可以接受的系统的实践见解。利用自动化中现有的信任理论模型，该项目将回答与人们在机器学习中的信任相关的三个基本问题。ML系统的性能(正确性、可靠性和可预测性)如何影响人们对它的信任？它的可解释性如何影响人们对它的信任？绩效和可解释性如何相互作用来影响信任？该项目将：(1)促进对ML系统的信任的发展和维护的理论和经验理解(并与传统自动化系统中的信任进行比较)，(2)提供有助于灌输对ML系统的信任的设计指南，以及(3)开发一个实验框架，用于以系统的方式评估和基准ML系统的可信性。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Machine learning (ML) has been applied to various domains including finance, healthcare, urban operations, and targeted news and advertising. They achieve great success uncovering insights from massive data and advancing decision making. Despite widespread applications, scientific understanding of peoples' trust or lack of trust in ML approaches is lacking, while the success of human-machine collaborations requires a deeper understanding of trust. This project advances the understanding of lay people's trust in ML through large-scale randomized human-subject experiments. The project will result in theoretical insights on the formation and maintenance of trust between humans and ML systems, and practical insights for designing systems acceptable to people.Leveraging existing theoretical models of trust in automation, this project will answer three fundamental questions related to lay people's trust in machine learning. How does the performance of an ML system (correctness, reliability, and predictability) affect people's trust in it? How does its interpretability affect people's trust in it? How do the performance and interpretability interact with each other to influence trust? The project will:(1) advance theoretical and empirical understanding about the development and maintenance of trust in ML systems (and compare with trust in conventional automated systems), (2) provide design guidelines that help instill trust in ML systems, and (3) develop an experimental framework for evaluating and benchmarking trustworthiness of ML systems in a systematic manner.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.