Winter into Spring Doctoral School "Understanding, Controlling, and Designing human-AI interactions"

Elena Glassman, Harvard University

Title:

Leveraging Theories of Human Cognition to Build Reliable Tools from Unreliable AI

Abstract:

AI is powerful, but it can make choices that result in objective errors, contextually inappropriate outputs, and disliked options. This is especially critical when AI-powered systems are used for context- and preference-dominated open-ended AI-assisted tasks—like ideating, summarizing, searching, sensemaking, and the reading and writing of text or code. We need AI-resilient interfaces that help users notice and recover from AI choices that are not right, or not right for them given their goals and context. We have derived design implications from key theories of human cognition to help us build more AI-resilient interfaces and reliable tools from unreliable AI. This talk will walk through two new systems that demonstrate this approach: CorpusStudio, an AI-powered writing environment, and MOCHA, a tool for co-adaptive machine teaching.

Bio:

Elena L. Glassman is an Assistant Professor of Computer Science at the Harvard John A. Paulson School of Engineering & Applied Sciences, specializing in human-computer interaction. Prior to that, she was a postdoctoral scholar at UC Berkeley, and obtained a BS, MEng, and PhD in Electrical Engineering and Computer Science from MIT. She has been named a Stanley A. Marks & William H. Marks Professor at the Radcliffe Institute for Advanced Study and a National Academy of Sciences Kavli Fellow. Her work has been funded by the NSF, private industry, the Berkeley Institute for Data Science, and the Sloan Research Fellowship. This work has received Best Paper and Honorable Mention awards at top-tier human-computer interaction research venues.

Zana Buçinca, Harvard University

Title:

Value-Aligned Human-AI Interaction

Abstract:

The anticipated large-scale deployment of AI systems in knowledge work will impact not only productivity and work quality but also workers' values and workplace dynamics. I argue that how we design and deploy AI-infused technologies will shape people's skills and competence, their sense of agency, collaboration with others, and even the meaning they derive from their work. I design human-AI interaction techniques that complement people and amplify their values in AI-assisted work. My research focuses on (1) understanding how people make AI-assisted decisions and (2) designing novel interaction paradigms, explanations and systems that optimize human-centric outcomes (e.g., human skills) and output-centric outcomes (e.g., decision accuracy) in AI-assisted tasks. In this talk, I will present a suite of interaction techniques I have introduced to optimize AI-assisted decision-making. These include cognitive forcing interventions that reduce overreliance on AI, adaptive AI support that enables human-AI complementarity in decision accuracy, and contrastive explanations that improve both decision accuracy and users’ task-related skills.

Bio:

Zana Buçinca is a PhD candidate in Computer Science at Harvard working at the intersection of human-AI interaction and responsible AI. Her research integrates cognitive and social science theories to design novel human-AI interaction techniques that complement workers and amplify their values in AI-assisted tasks. Her work has been recognized with the IBM PhD Fellowship, a Siebel Scholarship, and a Best Paper Award at IUI 2020. Zana has been named a Rising Star in AI by the University of Michigan, a Rising Star in Management Science & Engineering by Stanford, and one of the Top 10 Most Inspiring Women in STEM by UNDP Kosovo.