Designing and Evaluating Human-AI Interaction Strategies to Facilitate Collaborative Idea Generation

Collaborative ideation is a central activity of design-based and project-based learning scenarios. Despite its critical importance, the effectiveness of group-level idea generation is frequently hindered by socio-cognitive barriers, including production blocking, evaluation apprehension (fear of judgment), and idea fixation. While the emergence of generative artificial intelligence and generative language models offers transformative potential for supporting creative work, existing research has predominantly focused on individual-AI interactions, leaving a significant gap in understanding how these tools can support, facilitate the generation of innovative ideas in collaborative settings. This thesis addresses this research gap by designing and evaluating Human-AI interaction strategies specifically intended to facilitate collaborative ideation in educational contexts, with an emphasis on large-scale blended learning environments. This research employs a mixed-methods approach across seven empirical studies-including randomized controlled trials, participatory design workshops, and qualitative field evaluations-involving a total of 465 participants. To conduct this research, a web application-the Collaborative Ideation Facilitator-was developed and integrated into a learning experience platform, serving as a flexible research instrument for testing automated facilitation mechanisms. The thesis conceptualizes and implements a digital intervention model structured around three primary Human-AI interaction strategies: AI Editor, AI Peer, and AI-generated feedback. The AI Editor was designed to reinforce participant anonymity by acting as a linguistic filter that automatically reformulates submissions to strip away stylistic idiosyncrasies. Aside from strengthening anonymity, this mechanism proved particularly useful for non-native English speakers who felt better supported in articulating their thoughts. The study of AI Peers-autonomous generative agents that contribute their own ideas-reveals that generative language models systematically outperform human participants in generating original ideas, and that they provide the greatest creative stimulus in "cold-start" scenarios or for participants with limited prior knowledge of the topic. Moreover, this thesis introduces the concept of epistemic modulation, which posits that AI Peers actively reshape the epistemic frame of the group-the shared understanding of what kinds of ideas are valuable for the task. The study of automated formative feedback highlights a distinct paradox of utility. While AI-generated feedback was found to be highly actionable and specific, it simultaneously resulted in a measurable decrease in user enjoyment and a decrease of the sense of authorship. Educators involved in the research also identified sycophancy-the tendency of AI to be unhelpfully positive-as a critical limitation, emphasizing the need for more critical AI feedback to foster genuine learning and productive cri