From Gatekeeper to Architect: Operationalizing AI as a Cognitive Partner in Higher Education

In higher education, debates about Generative Artificial Intelligence (GenAI) often polarize around academic integrity risks versus efficiency gains. Missing from both accounts is a core instructional imperative. Large Language Models (LLMs) must be deliberately engineered to support higher-order cognitive development at scale rather than replace it. We present a simulator-based framework that operationalizes LLMs as constrained instructional instruments grounded in cognitive science and learning theory. The architecture decomposes the instructional process into three discrete phases (i.e., structured comprehension, schema-driven application and analysis, and synthesis under epistemic constraint), each instantiated via a dedicated, task-specific simulator. Central to this design is the instructor’s role as the sovereign epistemic regulator, defining knowledge granularity and uncertainty thresholds through precise parameterization and text selection. Rather than asserting a solution to Bloom’s 2-Sigma Problem, this framework demonstrates how LLMs can scale the specific instructional mechanisms associated with individualized tutoring (e.g., contingent feedback, error diagnosis, and iterative refinement). This approach provides a theoretically grounded, pragmatically constrained protocol for integrating GenAI into higher education, ensuring scalable cognitive development without compromising pedagogical sovereignty.