CICI: UCSS: Maximizing Data Utility and Participant Privacy through Usable, Secure Data Workflows for Human-Centered AI Research

点击翻译按钮获取中文摘要

A good way to improve privacy is to collect less data. However, all systems using artificial intelligence (AI), many of which people rely upon everyday (e.g., search engines, navigation apps, fraud detection, etc.), need data to work. This project balances the competing needs of obtaining data to improve AI technologies and maximizing the privacy of people contributing their personal information to research studies. Working in conjunction with the AI research community, this project develops a prototype system to help AI researchers secure data collected about people while at the same time improving their ability to make new scientific discoveries. Specifically, the prototype system will generate privacy-enhanced demographic questions and suggest data collection plans that ensure data collected about people balances needs for statistical rigor and community representativeness. In doing so, this system improves the quality, security, and efficiency of human-centered AI research by reducing the amount of data collected about people and helping to create AI systems that are usable, fair, accurate, and trustworthy.Designing human subjects studies that preserve research participants' privacy and security while still generating robust results is tricky. This project leverages cybersecurity techniques such as data-minimization to help human-centered AI researchers better protect research participants' privacy while ensuring their studies' statistical power and generalizability. In collaboration with the human-centered AI research community, this project builds a usable toolchain for generating data-minimizing demographic survey questions and determining statistically well-powered study sample size and demographically diverse composition to ensure the integrity of research results.The system maximizes the privacy of human subjects and recommends a sample size and composition which balances statistical power and representativeness. This approach promotes the quality of scientific discoveries at the point of study design. Through a process of need-finding, iterative toolchain refinement, and usability testing with AI researchers, this project builds a system for more secure, efficient, and robust human-centered AI research.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.