Research
My research is broadly focused on the relationships between people and AI systems. I’m especially interested in how we can design, understand, and evaluate these systems in ways that better support people, and in how these questions play out in practice. These questions lead me to combine my background in machine learning with perspectives and methods from HCI and the social sciences.
My research is deeply collaborative, and the themes described below have grown out of interactions and discussions with my amazing colleagues at Microsoft, as well as students and collaborators outside Microsoft. Being at a technology company naturally shapes the questions I ask: I get to see up close how AI is developed and used in practice, and how organizational realities shape efforts to build AI responsibly.
AI & Meaningful Human Engagement
Much of my current work focuses on what I've come to call AI & meaningful human engagement, exploring what it means for AI to empower and enhance people. Rather than focusing only on what an AI system can produce or accomplish, I’m interested in how interacting with it shapes people’s agency, skills, knowledge, and understanding, and how we can design systems with these broader effects in mind.
Transparency & Interpretability
My research on transparency and interpretability takes a human-centered view of what it means to understand an AI system. I think broadly about transparency, including model and dataset documentation, interpretability techniques, and ways of communicating uncertainty. Rather than treating transparency as an end in itself, I’m interested in what different people need to know about AI systems, how that understanding shapes their interactions with them, and how we can design transparency approaches that better support their goals.
Evaluating AI Systems
My research on AI evaluation focuses on the choices that turn things we care about into things we can measure. I’m interested in how we decide what to measure, whose perspectives shape those decisions, and whether the resulting evaluations support the conclusions we want to draw. My collaborators and I draw on ideas from measurement and the social sciences to think about how we can evaluate not only what AI systems can do, but also how interacting with them affects people.
Responsible AI in Practice
My work on responsible AI in practice explores the gap between what we want responsible AI practices to accomplish and how they actually work in organizations. I’m interested in practitioners’ needs and constraints, how tools and processes get adapted to particular contexts, and what studying these practices can teach us about how to better support responsible AI.
Foundations & Earlier Work
My earlier work spanned machine learning theory and algorithmic economics, including crowdsourcing and human computation, information aggregation and prediction markets, online learning, and learning across data sources. I often used theoretical tools to design algorithms and understand their behavior. A thread running through much of this work was an interest in the interplay between people and computational systems. Over time, I became increasingly interested in studying that interaction from the human side, and began drawing on methods from HCI and the social sciences to do so. That shift led to many of the questions that drive my research today.