AI-Augmented Inquiry and Regulation in Hybrid Systems: A Control Allocation Architecture for Preserving Epistemic Agency

Synthesis: Generative AI systems are increasingly integral to epistemic processes such as hypothesis generation, explanation construction, and decision-making, yet emerging evidence reveals a metacognitive dilemma: as external generative capacity increases, internal monitoring, calibration, and cognitive engagement may decline. Kuhn and colleagues propose the AIRIS (AI-Augmented Inquiry and AI Regulation in Education in Hybrid Systems) framework, a multi-level control allocation architecture specifying the conditions under which epistemic agency can be preserved in hybrid generative systems. Drawing on distributed cognition, cognitive load theory, multimedia learning, and self-regulated learning, it identifies seven interacting mechanisms through which hybrid cognition may become destabilized, from delegation and calibration drift to motivational-affective drift, and five regulatory operators (Anticipate, Interrogate, Reflect, Integrate, Synthesize) to counteract them.

Key Findings

1. A metacognitive dilemma arises when external generative capacity increases while internal monitoring, calibration, and cognitive engagement decline.
2. AIRIS is a multi-level control allocation architecture specifying conditions under which epistemic agency can be preserved in hybrid generative systems.
3. The framework draws on distributed cognition, cognitive load theory, multimedia learning, and self-regulated learning.
4. It identifies seven interacting mechanisms through which hybrid cognition may become destabilized, from delegation and calibration drift to motivational-affective drift.
5. Five regulatory operators (Anticipate, Interrogate, Reflect, Integrate, Synthesize) are specified to counteract destabilization.

What this means for practice

• Designers. Make uncertainty visible: surface genAI confidence estimates, alternative reasoning pathways, and the availability and quality of training data so learners can calibrate reliance instead of trusting a seamless interface.
• Designers. Build reversibility into the interface — let users temporarily disable assistance or periodically require unaided performance, countering the gradual, unnoticed accumulation of delegation and calibration drift.
• Instructors. Route learners through the five regulatory operators (Anticipate, Interrogate, Reflect, Integrate, Synthesize) so that genAI output is interrogated and reconciled with domain conventions rather than passively retained.
• Instructors. Judge genAI-supported work by the AI-withdrawal transfer test — what learners can do once the tool is removed — rather than by the quality of tool-assisted output.
• Researchers. Treat the seven destabilizing mechanisms and five operators as a program of testable propositions, not established findings, and design studies that isolate each mechanism.

Limitations

• This is a conceptual framework paper with no participants, intervention, or dataset; it integrates distributed cognition, cognitive load theory, multimedia learning, and SRL into an architecture, but presents no empirical test of the seven mechanisms or the five operators.
• The framework's outcomes are operationalized through an "AI-withdrawal transfer test," yet the paper supplies no instrument, reliability, or validity evidence for that measure, so its central success criterion is not yet a validated construct.
• Its supporting empirical claims are drawn from other researchers' work (for example the 61%/27% usage split in Liu et al. and degradation findings from Bastani et al.), never generated by the authors, so the architecture's fit to real learning contexts remains untested.
• Because it is a normative account of what must remain under learner control, it can specify design principles and propositions but cannot establish effect sizes or boundary conditions for the destabilization it describes.

Connected Concepts

• Metacognition
• Generative AI
• Self-Regulated Learning
• Cognitive Offloading
• Learner Agency
• Human-in-the-Loop
• Learning Theories

Connected Articles

• Towards a philosophy of ensemble cognition: Reconceptualising agency and mind in AI-mediated educational environments — Toward a philosophy of ensemble cognition
• Learning with machines: Toward a theory of epistemic co-agency — Learning with machines: Toward a theory of epistemic co-agency
• Artificial intelligence as a cognitive partner: a developmental framework for human-AI co-regulation in learning — Artificial intelligence as a cognitive partner
• From AI Anxiety to Strategic Regulation: How University Students Transform Generative AI into a Strategic Learning Resource — From AI Anxiety to Strategic Regulation
• Metacognitively Discordant Completion and the Aware Pass-Through of Non-Understanding in Generative AI Learning — Metacognitively Discordant Completion
• From Enhancement to Over-Reliance: A Mixed-Method Study of Generative AI and Sustainable Learning Performance — From Enhancement to Over-Reliance

Citation

Kuhn, Gerjets, Trautwein, Greene, Malone, Vogt, & Fütterer (2026). AI-Augmented Inquiry and Regulation in Hybrid Systems: A Control Allocation Architecture for Preserving Epistemic Agency. (physics.ed-ph).