UC Davis

Beyond the vast repository of conceptual and descriptive knowledge on diverse topics shared by people online today, procedural knowledge on how to accomplish specific tasks or perform procedural skills shared online is increasingly vital in daily and professional life. Online How-To knowledge sharing, with advantages of transcending time, geographic and social boundaries over traditional face-to-face knowledge-exchange, provides people a valuable means of tackling unfamiliar tasks by referring archived How-To instructions. However, both the authors and recipients of How-To instructions face challenges in actionability due to the implicit nature of procedural knowledge, the gap between expert and novices, the static nature of archived tutorials and the accessibility of technological affordances.

Experts encounter significant actionable difficulties in externalizing their procedural skills, as articulating actionable steps tailored to specific application situations and addressing the tacit nature of these skills is often more challenging than composing general encyclopedic content. Novices, in turn, may struggle with these instructions when referring because experts may overlook novice perspectives, unintentionally using abstract forms, personal metaphors, or specialized terminologies. Furthermore, archived instructions represent static, predefined plans that may not align with the dynamic nature of real-world task execution dynamics, making it difficult for novices to plan or decide appropriate actions to take in practical situations.

In this dissertation, I explore how human-chatbot interaction can augment the procedural skills and agency of novice or common users, addressing the actionability gap in procedural work. It examines the chatbot’s role and interventions through three perspectives: in-situ intelligence, knowledge-based intelligence, and emotional intelligence. These dimensions aim to facilitate user action-taking, problem-solving, and sustained engagement during procedural work.

The dissertation presents four research projects. First, I present an exploration study in which I investigate how media choice in How-To instructions affect the collaborative authoring of procedural instructions and regular learners’ knowledge consumption. The result indicates that multimedia could be an effective means to communicate procedural knowledge among knowledge sharers and between knowledge sharers and consumers. However, challenges remain, as static archived instructions often fail to account for the practical situation dynamics of real-world task execution, leading to difficulties for users to follow instructions. As such, in the second study, I explore whether incorporating an interactive and contextual mechanism into the archived instruction via a conversational agent could increase the actionability of the archived instruction. Specifically, it aims to better assist users’ situational planning and sequential instruction-following. This opens a research avenue to investigate how to design a conversational agent that could collaborate with common users on procedural tasks, exploring the spectrum from structured, knowledgeless conversations to knowledge-intensive conversations. In the third study, I further delve into how a chatbot providing knowledge-intensive conversations influences users’ information-seeking and problem-solving behaviors during procedural work. It also discusses unconscious cognitive biases that contribute to overtrust and undertrust in the chatbot. In the fourth project, I present the understanding of how emotional expressiveness affects voice chatbot-human interaction. Finally, I discuss the general design implications derived from these research studies and summarize the contributions to the field.