Core Concepts
Large Language Models (LLMs) facilitate human-AI co-creation of research questions, with breadth-first and depth-first approaches impacting creativity and trust differently.
Abstract
The content discusses the development of the CoQuest system for human-AI co-creation of research questions using Large Language Models (LLMs). It explores two interaction designs: breadth-first and depth-first RQ generation. The study evaluates user perceptions, behavior, and outcomes under each condition through a within-subject user study. Participants rated RQs for creativity and trust, with results showing differences between the two conditions. Breadth-first led to stronger perceived creativity and trust, while depth-first yielded higher-rated novelty and surprise in generated RQs.
Structure:
Introduction to CoQuest system for human-AI co-creation.
Experiment design with 20 HCI researchers.
Findings on user perceptions under breadth-first vs. depth-first conditions.
Impact on creativity, trust, novelty, and surprise in generated RQs.
User feedback highlighting preferences for each condition.
Stats
"Participants created 504 RQs throughout the study."
"Breadth-first condition resulted in 276 RQs; Depth-first condition resulted in 228 RQs."
"Survey results showed significantly stronger creativity (p=.015) and trust (p=.011) under breadth-first condition."
"Depth-first condition yielded higher-rated novelty (p=.002) and surprise (p=.017) in generated RQs."
Quotes
"AI-generated RQs were more innovative under the depth-first condition."
"Participants found the breadth-first approach easier to interpret and less cognitively demanding."