PBL-Students as AI Designers:A PBL Workshop on Co-Creating Pedagogically Aligned Chatbots

Abstract

The rapid spread of generative AI (GenAI) in higher education has sparked reactions ranging from tech-euphoria to tech-dystopia, and integration in teaching and practice remains uneven and problematic. Most educational AI tools align with teacher-centred approaches and primarily support individual, closed learning formats (Xu and Ouyang, 2022) and consequently, the role of chatbots in open, collaborative environments is challenging and underexplored. Recent reviews highlight persistent misalignments between intended learning strategies and chatbot design and implementation, resulting in limited contextual precision, reduced transparency and the opaqueness of proprietary systems (Wollny et al. 2021; Hwang and Chang 2023; Labadze et al. 2023; Otto et al., 2025). These issues are especially visible in Problem-Based Learning (PBL) or similar educational approaches that leverage collaborative inquiry, shared knowledge construction, and engagement with situated, open-ended problems in project groups.<br/>Within such settings, general-purpose chatbots often fail to support the situated nature of students’ work, incentivize undisclosed or uneven GenAI use among group members that obscure collaborative processes, create shadow workspaces, and undermine trust and shared responsibility (Otto et al., 2025). The “black-box” character of many commercial systems further complicates efforts to ensure equitable participation and mutual understanding (Bozkurt et al., 2023).<br/>In an attempt to mitigate these issues, we designed a workshop for co-creation of community-owned custom chatbots (Lyngdorf et al., 2025), offering an alternative to commercial tools and a testbed for pedagogically aligned GenAI. Involving students directly in developing and shaping chatbots has the potential to embed the technology within the project group rather than have it act as an external, pre-structured tool, helping to open what would otherwise remain a black box (Ejsing-Duun et al., 2022). Such student-centred processes can enhance transparency around AI use, strengthen trust, and align chatbot practices with principles of networked learning, echoing co-creation research on engagement and ownership (Kaminskiene et al., 2020). In addition, involving students in developing system prompts, knowledge bases etc. for chatbots assists them in developing critical approaches to the implicit biases and underlying instructions inherent in the chatbots that are now so prevalent in their lives and education. Our workshop is designed for educational systems where collaboration and inquiry are key pedagogical values, such as the PBL model used at Aalborg University. Here, students typically work in project groups with 3-6 members for a whole semester. The project groups organise large parts of their work themselves and have a high level of ownership and agency in the project work. In this connection, AI-literacy, including reflections on the impact of AI technologies on collaboration and learning processes, is crucial.<br/>For now, the workshop has only been piloted with a limited amount of students, but from a pedagogical and research perspective, our workshop aims to provide a structured space to examine opportunities and constraints of involving student-designed, chatbots in collaborative learning, not to promote blanket adoption, but to critically explore how co-creation can address current pedagogical challenges, support collaborative practices, and enable more transparent, pedagogically grounded uses of GenAI in higher education.<br/>For our session, we will share the core ideas and setup we have used to facilitate the workshops for students, give the participants a brief rundown of the introduction and instructions used and give a sample assignment from the workshop, allowing for a hands-on testing of the development of system-prompts through our framework. By the end of this hands-on workshop, participants will have collaboratively instructed and tested alternatives to large, general-purpose chatbots to demystify how GenAI actually works in educational settings, learned what a system prompt is and how educators and students can co-create, locally and iteratively, pedagogically grounded, context-specific prompts that “un-black-box” LLM behavior. We also hope to further reflections on practical ways to balance agency and ownership when integrating AI into teaching and group work.

Ähnliche Arbeiten