Attention in the Loop. Active coupling for preserving metacognitive control in hybrid intelligence workflows.

Generative AI tools are reshaping how people study and work, raising concerns that cognitive offloading is eroding attention and metacognitive control. This thesis synthesizes literature on cognitive offloading, the extended mind, and trust in automation, and proposes the Attention in the Loop framework. The framework distinguishes passive offloading, where users consume fluent outputs with minimal verification, from active coupling, where the interface scaffolds planning, verification, and reflection. It is operationalized through five evidence-based design principles (Commit Before Consult, Decomposition, Verification, Retrieval First, Germane Load) and discussed in relation to education, knowledge work, and safety-critical professions. The thesis closes with an experimental agenda for testing whether active coupling preserves metacognitive control in hybrid intelligence workflows.