Session 9 - Strand 2 (Online)

Abstract:

Generative AI challenges how higher education identifies and evaluates understanding. When AI systems can rapidly produce essays, explanations, and arguments, traditional performance outputs no longer reliably indicate learning. This raises a central question: if AI can generate the products associated with thinking, what becomes of thinking itself in education? This presentation argues that the core issue extends beyond academic integrity to a deeper problem in how universities conceptualise understanding. Academic assessment often treats performance outputs as proxies fo learning, yet such outputs do not necessarily reflect the reasoning processes that produce understanding. The talk introduces the IAR Hypothesis, developed within a broader research project on metacognitive learning architectures. Prototypes deployed through LLM interfaces have generated over 1,000 recorded student interactions with positive early feedback, suggesting a possible pathway for AI to support the development of higher-order thinking.