Exploring Collaboration Models for Supporting Ideation with AI Agents

Workshop Recommended citation: La Scala, J., & Gillet, D. (2025). Exploring Collaboration Models for Supporting Ideation with AI Agents. In Kangaslampi, R., Langie, G., Järvinen, H.-M., & Nagy, B. (Eds.), SEFI 53rd Annual Conference. European Society for Engineering Education (SEFI), Tampere, Finland. DOI: 10.5281/zenodo.17631593. This Conference Paper is brought to you for open access by the 53rd Annual Conference of the European Society for Engineering Education (SEFI) at Tampere University in Tampere, Finland. This work is licensed under a Creative Commons Attribution-NonCommercial-Share Alike 4.0 International License. Exploring Collaboration Models for Supporting Ideation with AI Agents J La Scala a,1, D Gillet b a École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, 0000-0002-8057-7787 b École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, 0000-0002-2570-929X Conference Key Areas: Digital tools and AI in engineering education Keywords: Generative Language Models, Human-AI Collaboration, Ideation, Innovation, Design-based Learning ABSTRACT This workshop explores Human-AI collaboration in ideation within design-based learning, a pedagogical approach widely used in engineering education. With the rise of Generative AI and generative language models (GLMs), new possibilities have emerged to support both divergent and convergent thinking in creative problemsolving. However, while AI agents have shown promise in generating ideas and guiding the ideation process, their integration into educational contexts remains underexplored, particularly from the perspective of educators. Participants engaged in a structured, hands-on exploration of AI-assisted ideation. The workshop began with an introduction followed by an interactive demonstration, where participants experienced an AI-supported ideation activity from a student’s perspective. This was followed by a guided tutorial on implementing AI-generated feedback in a collaborative online environment (Graasp.org), allowing participants to tailor the behavior of the AI agent for their own educational scenario. The final segment consisted of a structured discussion, where participants reflected on the benefits, limitations, and pedagogical implications of AI feedback for ideation, as well 1 Corresponding Author J La Scala jeremy.la[email protected] as shared their perspectives and expectations regarding human-AI collaboration for ideation activities. By the end of the workshop, participants had firsthand experience designing and integrating AI feedback in ideation activities, and contributed to a collective report capturing educator insights—informing ongoing research and practice at the intersection of AI, design-based learning, and engineering education. 1 BACKGROUND AND RATIONALE 1.1 Ideation in Design-based Learning Design-based learning is an approach used in engineering education to teach students the integration of their disciplinary knowledge and skills towards the design of an artifact. Within this approach, design thinking is the core metacognitive and managerial process that students practice. At its core, the creativity of the students is solicited to combine their knowledge to devise potential solutions to their problem statement through ideation. Ideation comprises two complementary cognitive processes: divergent thinking, which generates a broad array of ideas, and convergent thinking, which refines and selects the most suitable ones (Frich et al., 2018, 2019). 1.2 Human-AI Collaboration for Ideation With advancements in Generative AI and more specifically, generative language models (GLMs), new possibilities have emerged for supporting the ideation process. GLMs can generate a wide variety of ideas (Haase & Hanel, 2023; Memmert et al., 2024; Stevenson et al., 2022) and thus, contribute to divergent thinking. In the form of chatbots, these models can aid individuals in the ideation process, providing valuable suggestions or guidance (Memmert & Bittner, 2024). In this workshop, we introduced and discussed new AI agents and their corresponding interaction models that we designed for supporting students during collaborative ideation. We classify these agents in two categories: (1) the artificial peers and (2) the artificial facilitators. The artificial peers enact the role of a participant to the ideation activity. They can propose ideas and provide feedback on these ideas. Studies of artificial peers as contributors to the pool of ideas have reported a positive reception by participants to the ideation but were inconclusive regarding the effects on the creativity of the groups (La Scala, Bartłomiejczyk, et al., 2025; Schwabe et al., 2025). On the other hand, the facilitators provide guidance during the activity, without directly contributing to the ideation. This type of agent may be more relevant to educational contexts as their task do not overlap with the role of the students and do not provide an opportunity for the students to offload their thinking to the agent. This workshop focuses on the concept of artificial facilitator and, in particular, AI feedback. To date, few investigations have examined teachers’ perceptions in designing and implementing systems that facilitate human-AI collaborative ideation. 2 WORKSHOP OBJECTIVES AND DESIGN The purpose of this workshop was to initiate the conversation in the engineering education community on Human-AI collaboration and the corresponding interaction models to support creativity in ideation. More specifically, we collected the perception of the teachers regarding these technologies, discuss new collaboration models with their benefits, risks and limitations. The participants had the opportunity to experiment with the configuration and the use of a feedback agent through an online collaborative environment integrated to Graasp.org, a learning experience platform (LXP) (Gillet et al., 2022). 2.1 Target Audience This workshop has been designed for science and engineering educators interested in exploring the benefits of Generative AI in education, particularly through its application in project-based and design-based learning approaches. The workshop was also intended to be relevant to educators working in interdisciplinary contexts, where students engage with design projects that involve collaboration with social sciences, humanities, or other non-technical disciplines. 2.2 Structure Table 1. Time plan of the workshop. Run time Activity 15 min Introduction and presentation of Human-AI collaboration models for ideation with focus on AI feedback. 5 min Interactive demo - Participants tried a collaborative ideation activity with AI feedback. 25 min Agent implementation tutorial - Participants were introduced to prompt design for feedback generation and configured an agent for a scenario of their choice. 15 min Structured discussion – Participants shared their experience with prompt design and their vision for integrating AI feedback in collaborative activities. The workshop started with the presentation of possible agent roles in the context of collaborative ideation. Then, we opened a short discussion followed by a questionnaire on AI feedback and the risk and benefits associated to it. Following this introduction, the participants were provided with an example ideation activity in which they could try to propose ideas and receive automated feedback. Afterwards, they were guided in the preparation of their own ideation activity on a topic of their choice. In this step, the participant used a web application that we developed to integrate AI agents in collaborative scenarios. The application has been designed to support the teachers in the configuration of the agents and the activity through an intuitive interface, with the aim to considerably reduce the time and skills required to design and implement such activities. The main task was to design a prompt that was automatically used by the system to generate feedback when a new idea was submitted. The participants were given two templates they had to discuss and tailor. The app relies on a template engine1 to build the requests that are submitted to the AI model. When generating feedback, special markup is replaced by the problem statement, the idea (labeled response in the system), and previous ideas with their corresponding feedback. See Prompt 1 for a detailed example and see (La Scala, Bartłomiejczyk, et al., 2025; La Scala, Sahli, et al., 2025) for a detailed description of the whole application. After this tutorial, we opened a group discussion on the prompt design, AI feedback, and more generally the concept of AI feedback in collaborative ideation. We collected the answers on an online discussion board where participants could submit their opinion in parallel to the discussion. We finally opened to discussion to the expectations of the participants regarding human-AI collaboration in the context of collaborative ideation, and, more generally, collaborative learning. 3 OUTCOME The workshop involved eight participants who were organized into three groups for the hands-on activities. It was facilitated by the first author. Data collected, field notes and additional material are available in the supplementary material: https://doi.org/10.5281/zenodo.17200231. Following the introduction, we briefly discussed the benefits, limitations, and risks associated with human-AI collaboration, particularly regarding AI-generated feedback. In groups, participants completed a survey exploring their perceptions of 1 LiquidJS: https://liquidjs.com/ [archive] Figure 1: Example of an idea proposed by one participant with the AI-generated feedback. the main benefits and risks of incorporating AI-generated feedback in collaborative ideation. The primary benefit that emerged was the potential to obtain feedback more quickly and at a larger scale. One group noted that AI cannot become fatigued by providing repetitive feedback, unlike teachers or teaching assistants. They also recognized that using AI feedback creates opportunities to discuss AI and human-AI collaboration in the classroom while fostering critical thinking about AI-generated responses. Another group emphasized the greater objectivity of AI, which is not influenced by student characteristics or interpersonal relationships. Regarding drawbacks and risks, two groups identified the danger of over-reliance on AI feedback, which presents a twofold concern. First, the feedback may be inaccurate or completely incorrect, potentially due to hallucinations. Second, students may use AI feedback to avoid meaningful interactions with teachers. Participants mentioned sycophantic AI (Sharma et al., 2025) as a potential driver of this risk, which could impair feedback quality by making it insufficiently critical and more appealing to students than teacher or peer feedback. After the interactive demonstration, we solicited participants’ opinions on the AIgenerated feedback. All participants agreed on two critical issues: the feedback was excessively long and verbose, and it was overly positive, reinforcing the previously identified problem of sycophantic AI. With these concerns in mind, we proceeded to the tutorial on prompt design and AI agent configuration. During the session, all groups successfully configured their activities and tested them within the allocated timeframe. One group inquired about the possibility of having the AI generate and includes images in its feedback, which was not technically feasible at the time. By examining the prompts developed by participants, we made two key observations. First, very few modifications were made to the templates, which may Problem statement: {{ problem_statement }} Participant's idea: {{ current_response }} {% if previous_responses.size > 0 %} Previous ideas and feedback: {% for item in previous_responses %} - Idea {{ forloop.index }}: {{ item.response }} Feedback: {{ item.feedback }} {% endfor %} {% else %} No previous ideas have been submitted yet. {% endif %} Please provide constructive feedback on the participant's idea with a focus on its relevance to energy usage and sustainability. Use the following structure: 1. Strengths: [<DESCRIBE WHAT MAKES STRONG IDEAS>] 2. Opportunities and questions: [suggest one or two ways the idea could be further developed, and pose openended questions that help the participant elaborate, clarify assumptions, or explore new angles] {% if previous_responses.size > 0 %} 3. Relation to previous ideas: [explain how this idea complements, improves, or differs from earlier ideas and feedback] 4. Make sure the feedback is short, with a maximum of thirty words. 5. Give a picture related to the feedback. 6. If student's input is shorter than ten words, reply: Please use a full sentence to describe your idea. 7. Give all output in Dutch. {% endif %} Prompt 1: Prompt designed by group 1. Additions to the template are in bold. Double brackets ( {{ }} ) and percentage brackets ( {% %} ) are used for variable rendering and control structures respectively. indicate that offering ready-to-use templates significantly reduces the barrier to entry for educators. Second, one group failed to recognize that most of their prompt would be ignored due to the conditional rendering system used for prompt generation (see Prompt 1). Conditional rendering conditions parts of the prompt based on the presence of previous responses in the discussion thread. During the workshop’s concluding discussion, participants reflected on the activity and shared their experiences both orally and through the online collaboration board. The first observation was that preparing the activity and designing the prompt proved easier than expected. One participant specifically highlighted the benefit of having a functional template as a starting point. However, another participant countered that the system’s user experience was initially unclear and required guidance. All participants agreed that the teacher’s ability to precisely configure AI behavior represents significant value. Nevertheless, they emphasized the difficulty of designing effective prompts and stressed that guidance and training would be essential for educators. One participant suggested that predefined templates, similar to those provided in this workshop, combined with clear guidelines integrated into the user interface, could serve as simple and effective strategies to support teachers. Building on this point, another participant noted that substantial literature on feedback in educational contexts already exists, from which design principles for effectively prompting AI models to deliver feedback could likely be derived. Beyond the outcomes discussed above, we aimed for participants to gain understanding of multiple conceptual models for human-AI collaboration in collaborative ideation. Additionally, participants acquired experience in implementing AI feedback using our online collaborative ideation application, hosted within the LXP Graasp.org. 4 CONCLUSION This workshop demonstrated the potential of integrating configurable AI feedback within collaborative ideation platforms for engineering education. Participants recognized AI’s capacity to provide scalable, objective feedback while identifying critical implementation challenges. The primary barrier to adoption is the complexity of effective prompt design. While participants found the configuration process manageable with functional templates, successful implementation requires substantial support infrastructure, including ready-to-use templates and comprehensive educator training. A significant limitation emerged through the sycophantic behavior of current AI models, which produced overly positive and verbose feedback. This finding highlights a fundamental tension: while participants valued AI’s objectivity, current models tend toward unhelpfully accommodating responses that may impede genuine learning. This reinforces the need for AI models with enhanced capacity for productive criticism, such as antagonistic AI (Cai et al., 2024). These findings underscore the importance of developing configurable human-AI collaboration systems that can effectively enhance facilitation practices in engineering education and beyond. NOTE This article's language was refined with the assistance of generative language models. REFERENCES Cai, A., Arawjo, I., & Glassman, E. L. (2024). Antagonistic AI (arXiv:2402.07350). arXiv. https://doi.org/10.48550/arXiv.2402.07350 Frich, J., MacDonald Vermeulen, L., Remy, C., Biskjaer, M. M., & Dalsgaard, P. (2019). Mapping the Landscape of Creativity Support Tools in HCI. Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1–18. https://doi.org/10.1145/3290605.3300619 Frich, J., Mose Biskjaer, M., & Dalsgaard, P. (2018). Twenty Years of Creativity Research in Human-Computer Interaction: Current State and Future Directions. Proceedings of the 2018 Designing Interactive Systems Conference, 1235–1257. https://doi.org/10.1145/3196709.3196732 Gillet, D., Vonèche-Cardia, I., Farah, J. C., Hoang, K. L. P., & Rodríguez-Triana, M. J. (2022). Integrated Model for Comprehensive Digital Education Platforms. 2022 IEEE Global Engineering Education Conference (EDUCON), 1587– 1593. https://doi.org/10.1109/EDUCON52537.2022.9766795 Haase, J., & Hanel, P. H. P. (2023). Artificial muses: Generative Artificial Intelligence Chatbots Have Risen to Human-Level Creativity. Journal of Creativity, 33(3), 100066. https://doi.org/10.1016/j.yjoc.2023.100066 La Scala, J., Bartłomiejczyk, N., Gillet, D., & Holzer, A. (2025). Fostering Innovation with Generative AI: A Study on Human-AI Collaborative Ideation and User Anonymity. Proceedings of the 58th Hawaii International Conference on System Sciences, 337–351. https://doi.org/10.24251/HICSS.2025.041 La Scala, J., Sahli, S., & Gillet, D. (2025, April). Stimulating Brainstorming Activities with Generative AI in Higher Education. 2025 IEEE Global Engineering Education Conference, London, UK. https://doi.org/10.1109/EDUCON62633.2025.11016340 Memmert, L., & Bittner, E. (2024). Human-AI Collaboration for Brainstorming: Effect of the Presence of AI Ideas on Breadth of Exploration. Proceedings of the 57th Hawaii International Conference on System Sciences. Hawaii International Conference on System Sciences, Hawaii. https://hdl.handle.net/10125/106425 Memmert, L., Cvetkovic, I., & Bittner, E. (2024). The More Is Not the Merrier: Effects of Prompt Engineering on the Quality of Ideas Generated By GPT-3. Hawaii International Conference on System Sciences. https://doi.org/10.24251/HICSS.2023.903 Schwabe, G., Katsiuba, D., Specker, R., & Dolata, M. (2025). AI-ThinkLets for Brainstorming. Hawaii International Conference on System Sciences. https://doi.org/10.24251/HICSS.2025.043 Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S. R., Cheng, N., Durmus, E., Hatfield-Dodds, Z., Johnston, S. R., Kravec, S., Maxwell, T., McCandlish, S., Ndousse, K., Rausch, O., Schiefer, N., Yan, D., Zhang, M., & Perez, E. (2025). Towards Understanding Sycophancy in Language Models (arXiv:2310.13548). arXiv. https://doi.org/10.48550/arXiv.2310.13548 Stevenson, C., Smal, I., Baas, M., & Grasman, R. (2022). Putting GPT-3’s Creativity to the (Alternative Uses) Test (arXiv:2206.08932). arXiv. http://dx.doi.org/10.48550/arXiv.2206.08932