Human–AI Augmented Cognition Framework: A Purpose-Led Framework for Cognitive Allocation and Augmentation in Human–AI Practice

Abstract

The Human–AI Augmented Cognition Framework (HAACF) is a purpose-led framework for designing how cognition should be distributed between humans and artificial intelligence. Rather than beginning with what AI can automate, HAACF begins with the cognitive intent of an activity and asks which cognitive operations should be retained by humans, augmented through Human–AI interaction, or delegated to AI. The framework introduces a seven-dimensional Cognitive Allocation Test based on capability, responsibility, consequence, context, evaluability, complementarity and efficiency. It distinguishes three principal allocation modes — Retain, Augment and Delegate — alongside an Unresolved state for situations in which an appropriate allocation cannot yet be established. Where augmentation is appropriate, HAACF provides five cognitive interaction patterns: FORM, EXPAND, CHALLENGE, EVALUATE and JUDGE. It also introduces Evaluability and Evaluation Escalation as mechanisms for determining whether AI contributions can be adequately assessed for their intended use. A central principle of HAACF is that Task Performance and Human Capability are distinct outcomes. AI-supported activity may improve immediate performance while strengthening, maintaining or reducing capabilities that matter beyond the immediate task. The framework therefore treats cognitive allocation as a dynamic design problem requiring metacognitive regulation, reflection and adaptation over time. HAACF is intended to support application across learning, professional knowledge work, organisational decision-making and Human–AI system design. It complements the CloudPedagogy AI Capability Framework, Capability-Driven Development and Human–AI Governance Engineering while remaining an independent framework. Version 1.0 is presented as a proposed conceptual and practical framework. It establishes the initial architecture, terminology, application methods, worked examples, limitations and research agenda for subsequent evaluation and development.