Sep 21 (Mon) @ 9:30am: "Designing Interactive Decision Support Systems: From Perceptual Aids to Shared Decision-Making," Avinash Ajit Nargund, ECE PhD Defense

Location: Engineering Science Bldg (ESB), Room 1001

Abstract

Humans make decisions every day, from routine choices to consequential strategic ones. Doing so requires identifying relevant information, interpreting it in light of one’s goals and prior experience, and determining how to act. Intelligent interactive systems can support this process by making task-relevant information easier to perceive and interpret and, increasingly, by taking on parts of the decision process themselves.

This dissertation investigates these two forms of support across sensorimotor and strategic human–AI collaboration settings. In sensorimotor settings spanning movement learning, motor control, and immersive navigation, we examine how computational systems can make otherwise difficult-to-perceive information available to users, such as subtle aspects of movement during exercise or location of nearby physical obstacles while immersed in virtual reality. We further investigate how these systems can help users develop the perceptual skills needed to recognize task-relevant features independently, reducing their reliance on continued external guidance. In strategic human–AI collaboration, we examine how users decide which parts of a demanding decision process to perform themselves and which to delegate to an AI, particularly when evaluating many possible actions and their consequences. We investigate how these preferences evolve through interaction and whether behavioral and task-state signals can anticipate changes in the allocation of work.

Taken together, this dissertation contributes empirically grounded guidelines for designing interactive systems that seek to address specific challenges users face when making decisions, rather than treating an increase in information conveyed or in system involvement as inherently beneficial. Across the studies, the effectiveness of computational support depended in part on how well it aligned with user’s decision-making strategies, prior experience, and evolving mental models. The findings also reveal recurring tradeoffs across performance, cognitive effort, and user preferences, suggesting that effective support cannot be defined along a single dimension. As physical and agentic AI systems become increasingly capable of participating in human decisions, these findings provide a starting point for designing support that accounts for both the demands of the task and the characteristics of the user.

Bio

Avinash Ajit Nargund is a ECE PhD candidate advised by Prof. Misha Sra. His research focuses on designing and evaluating AR/VR systems that support sensorimotor performance and learning, as well as decision-making and control dynamics in human–AI collaboration. He received his B.E. from R.V. College of Engineering and his M.S. from the Georgia Institute of Technology, and has previously worked at NIO USA Inc. and Argo AI.

Hosted By: ECE Professor Misha Sra

Submitted By: Avinash Nargund | Email