Architecting Uncertainty

How can we know whether students truly understand something in an age of generative AI, where essays and quiz answers can be machine-generated in seconds? This practical workshop explores how AI-powered simulations and role-plays can evidence understanding through skill application in novel contexts, particularly in asynchronous online learning. Rather than inferring understanding from performance on familiar tasks, we focus on scenario-based performance where learners must act, decide and justify under uncertainty. Drawing on research on transfer and productive failure (Kapur, 2025), we argue that understanding is best evidenced when students successfully apply ideas in new situations after grappling with initial difficulty. We also build on work in motivation and self-determination (Boschi, 2020), emphasising that AI-mediated assessment must protect learners’ autonomy, competence and relatedness,placing them on the “edge of chaos” in ways that are challenging, ethical, and psychologically safe.The workshop uses a recent pilot in the Digital Learning Institute’s Certified Digital Learning Professional programme, where working professionals completed an AI-enabled simulation created in Muzzy Lane's 'Author', applying digital learning design principles in a novel client scenario. We will share student perspectives from this pilot and introduce a simple design template that helps participants reimagine one of their own assessments as a short, AI-facilitated simulation or role-play. Over 60 minutes, participants will map an existing outcome to a novel context, identify productive-failure opportunities, and sketch feedback and debrief structures that make uncertainty safe and informative. They will leave with a draft assessment concept aligned to their own context and concrete ideas for embedding, using and critiquing AI in responsible, future-facing ways.