With the ongoing innovation of intelligent systems that coordinate and collaborate with humans, it becomes increasingly important to understand how interactions should be designed to support effective communication, social norms, and appropriately calibrated trust. These intelligent systems are becoming less and less constrained to single embodiments: voice-activated agents that are typically embodied in smart speakers, for example, can interact with users through multiple platforms and control multiple devices in a shared space. As researchers and designers explore the potential for agents to serve as interactive interfaces to complex systems, they grapple with questions of how technical constraints and social context might impact aspects of agents' design and use. These aspects include possibilities for and effects of physical design, how agents should handle complex ethical and interpersonal constructs like social privacy (what happens when a smart home agent keeps secrets?), how they might be mentally modeled (are they tools, collaborators, or something else?), and what their roles and responsibilities are among genuine social players.

I argue that agent identities can play a mediating role in shaping the interactions that are situated in these complex and integrated contexts, as well as their outcomes. By integrating theoretical, empirical, and design work on agent identity, smart environments, and technology mediation, I formulate a preliminary conceptual model of agent identity as a mediating entity in relationships among individuals, embodiments, and services. In this dissertation, I discuss several studies that explored possible future designs for agent identities as service touchpoints, manipulated agent identity in human-robot collaboration settings, examined the role of embodiment in interactions between agent identities and ancillary users, and explored possibilities for future human-agent interaction in smart homes and other smart environments. The work that I have done with my colleagues to date has revealed novel insights that aid in mapping the space of human-agent interactions in complex social and physical environments and informing new frameworks for understanding and studying AI agent identities. In interpreting these insights, we have made several contributions to basic scientific knowledge in human-computer interaction (HCI), human-robot interaction (HRI), and human-agent interaction (HAI). We also generated design implications for agents and robots that interact socially with people in various domains. This work has both revealed and addressed a crucial need and timely opportunity to formulate new, informed approaches to human-AI interactions that take a bird's-eye view of entire AI-rich environments (including multiple people, multiple agents, and multiple embodiments).

194 pages

Thesis Committee:
Jodi Forlizzi (Co-Chair)
Aaron Steinfeld (Co-Chair)
Geoff Kaufman
Terrence Fong (NASA Ames Research Center)

Brad A. Myers, Head, Human-Computer Interaction Institute
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu