This text outlines the professional work of Advait Sarkar, a researcher and lecturer affiliated with the University of Cambridge, University College London, and Microsoft. His research focuses on the intersection of human-computer interaction, artificial intelligence, and data programming. Key areas of his work include expanding the capabilities of spreadsheets for tasks like machine learning, critically examining the design of human-AI interactions, and developing novel data visualization tools. A significant portion of his recent work is dedicated to Generative AI, where he studies its cognitive impact on knowledge workers, proposes new interaction models, and analyzes the limits of AI explainability. His academic research has translated into tangible product contributions at Microsoft (Excel, Teams, PowerPoint) and informs his roles as a government policy advisor and an educator.
Advait Sarkar's Research: Bridging Human Interaction with AI and Data
Spreadsheet Innovation: Sarkar's research expands the boundaries of spreadsheet technology. He has demonstrated how to use spreadsheets for machine learning and music composition, improve formula efficiency, integrate Python code, and enable spreadsheets to understand units like dollars and kilograms.
Critiquing AI Interaction Design: He challenges the term "human-AI collaboration" as a poor metaphor that overlooks the labor of data workers. His research also finds that users often prefer specifying rules to an AI rather than providing simple feedback (like/dislike) and that conversational language is not always the ideal way to interact with voice assistants.
Improving AI Training Data: He has shown that asking people to rank items (e.g., from best to worst) is a faster and more accurate way to measure human judgment than asking for numerical scores. This insight is crucial for training higher-quality machine learning systems.
Generative AI and Cognition: His recent research explores how Generative AI tools like ChatGPT affect knowledge work. He proposes that AI can be designed as a "provocateur" to challenge users and improve their critical thinking and intentionality.
Understanding and Defining Generative AI: Sarkar has developed a working definition for the term "Generative AI" to bring clarity to the field. He also argues that, in general, we cannot explain the behavior of large language models, and importantly, they cannot accurately explain their own behavior.
Microsoft Product Impact: His work has directly influenced Microsoft products. He contributed to Excel's dynamic arrays, rich data types, and Python integration; the "reuse slides" feature in PowerPoint; and research on remote work and creativity that informed the development of Microsoft Teams.
Policy and Education: Sarkar acts as an advisor to the UK and other governments on technology and automation. He also co-taught one of the world's first graduate courses on designing user interfaces for machine learning systems at the University of Cambridge.