Artificial intelligence as a cognitive partner: a developmental framework for human-AI co-regulation in learning

Synthesis: This conceptual analysis by S, Joseph, Jose, S. M, N, & Joseph (2026) proposes that AI should be understood not merely as an external tool but as a cognitive partner in the co-AI Regulation in Education of thinking, learning, and self-control. Drawing on executive function, Metacognition, distributed cognition, and sociocultural development, it frames human-AI interaction as co-regulated cognition where AI acts as a scaffold, metacognitive support, external memory system, and decision partner. The framework is argued to be most relevant in middle childhood, adolescence, and adulthood, weighing the benefits of cognitive offloading against the risks of over-reliance.

The shift from tool to cognitive partner

The paper argues that developmental psychology has lacked a theoretical framework for the role of AI in cognition regulation across the lifespan. It proposes that AI participates in the co-regulation of thought and action rather than merely providing information. Grounded in learning theory, the authors integrate executive function, Metacognition, distributed cognition, and sociocultural (Vygotskian) development to describe a developmental paradigm of human-AI co-regulation.

Four roles of AI in cognitive regulation

1. Scaffold for cognitive performance — AI systems (e.g., intelligent tutoring systems) enable learners to work beyond their independent capability, offering 24-hour support, real-time Feedback, and dynamically adjusted difficulty, extending traditional Scaffolding.
2. Metacognitive support — conversational AI prompts users to rethink answers, generate alternative explanations, and validate their reasoning, supporting self-regulated learning.
3. External memory and cognitive offloading system — AI stores, retrieves, and generates information, offloading working-memory load in line with Cognitive Offloading research.
4. Decision partner — recommendation and conversational systems participate in planning and decision-making, framing AI as a collaborator rather than a tool.

Developmental stages and co-regulation

The framework is stage-sensitive: it is most directly applicable to middle childhood, adolescence, and adulthood, where metacognitive and self-regulatory capacities are developed enough for meaningful human-AI collaboration. In early childhood, AI's role is limited to structured external regulation of behavior. Across later stages, AI shifts from metacognitive partner (middle childhood and adolescence) to collaborator in complex cognition (adulthood).

Benefits and risks

AI-assisted co-regulation can improve performance, reduce cognitive load, and extend memory through beneficial offloading. However, the authors warn of the risks of excessive Cognitive Offloading, Over-Reliance on external advice, and reduced effortful processing and independent reasoning — outcomes that can undermine the development of self-regulation and Critical Thinking skills, especially in younger learners.

Key Findings

• Proposes a developmental framework of human-AI co-regulation in which AI is conceptualized not merely as an external tool but as a cognitive partner in the co-regulation of thinking, learning, and self-control.
• Integrates executive function, metacognition, distributed cognition, and sociocultural development theories to frame AI-mediated cognition as co-regulated between learner and intelligent system.
• Identifies four roles for AI in cognitive regulation: scaffold, metacognitive support, external memory / cognitive offloading system, and decision partner.
• Argues the framework is most relevant in middle childhood, adolescence, and adulthood; early childhood AI support is limited to structured external regulation.
• Highlights both benefits (performance gains via offloading, scaffolding, reduced cognitive load) and risks (over-reliance, excessive cognitive offloading, reduced independent reasoning and self-regulation development).
• Calls for longitudinal research, common conceptual definitions, and learning environments that balance external support with independent problem solving.

What this means for practice

• Instructors. Make support fade against demonstrated performance instead of holding it constant: intelligent tutors should give fewer hints as learners improve and ask learners to justify an answer before supplying it.
• Instructors. Divide the curriculum by what AI may carry: let it handle information processing, feedback provision, scaffolding, and routine problem solving, while critical thinking, ethical reasoning, decision making under uncertainty, and self-regulation stay sustained human work.
• Instructors. Match the AI role to the developmental stage the framework assigns: structured external regulation of behavior in early childhood, a metacognitive partner in middle childhood and adolescence, and a collaborator on complex cognition in adulthood.
• Researchers. Treat co-regulation with an intelligent system as a measurable construct: the paper's own call is for longitudinal work and common conceptual definitions before the four roles can be tested rather than asserted.

Limitations

• This is a conceptual analysis rather than a study: no learners were observed, so the four roles and the stage-sensitive framework are argued from existing theory and remain proposals.
• The theories it builds on — Vygotskian scaffolding and guided participation, executive function research, and distributed cognition — were developed where the guiding partner was another human; the authors note that AI responds instantly, runs continuously, and lacks the social limits governing human communication, so AI-mediated regulation may not reproduce human-scaffolding findings.
• The developmental boundaries are asserted rather than measured: early childhood is confined to structured external regulation and adulthood to collaboration on complex cognition, but no age-graded evidence is presented for where those transitions fall.
• The framework names excessive cognitive offloading as the central risk while crediting offloading's benefits, and supplies no measure or threshold for where the balance tips.

Citation

S, P., Joseph, J., Jose, M., S. M, A., N, R., & Joseph, J. (2026). Artificial intelligence as a cognitive partner: a developmental framework for human-AI co-regulation in learning. Frontiers in Developmental Psychology, 4, 1835258.