When does AI support thinking, and when does it replace it? Learners' conceptualisations of AI as a dynamic cognitive partner: A typology

C.K.Y Chan

arxiv: 2602.15638 · v2 · submitted 2026-02-17 · 💻 cs.CY

When does AI support thinking, and when does it replace it? Learners' conceptualisations of AI as a dynamic cognitive partner: A typology

C.K.Y Chan This is my paper

Pith reviewed 2026-05-15 21:50 UTC · model grok-4.3

classification 💻 cs.CY

keywords AI in educationcognitive extensioncognitive offloadinglearner perspectivestypology developmentsecondary educationlearning theories

0 comments

The pith

Learners conceptualize AI as a cognitive partner that can extend thinking or replace cognitive effort.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper investigates how learners themselves view AI's role in their cognitive processes during learning. By analyzing written accounts from 145 secondary students in Hong Kong, it builds a typology of nine cognitive functions where AI acts as a dynamic partner. A key distinction emerges: AI can support sense-making by extending cognition or enable offloading by replacing effort. This matters because it helps identify when AI use aids or hinders learning. The typology draws on established learning theories to reframe AI as shifting mediation rather than a static tool.

Core claim

The central discovery is a learner-informed typology that conceptualises AI as a dynamic cognitive partner. Learners describe engaging with AI through nine interrelated functions including conceptual scaffolding, feedback, idea generation, organisation, adaptation, monitoring, and workload regulation. Across these, they consistently distinguish between AI use that extends their cognition and AI use that replaces cognitive effort, showing that the same interaction can either support sense-making or enable cognitive offloading based on how they position AI in the learning process.

What carries the argument

The learner-informed typology of AI as a dynamic cognitive partner, which shifts between extending cognition and replacing cognitive effort across nine functions.

If this is right

AI-supported learning can either promote deeper understanding or lead to cognitive offloading depending on learner positioning.
The typology provides a lens for educators to guide students toward extension rather than substitution.
Design of AI tools in education should account for how learners perceive their role in cognitive processes.
Future research can use this boundary to study impacts on learning outcomes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

AI systems could be designed with prompts that encourage users to reflect on whether they are extending or replacing their thinking.
Similar distinctions might apply in professional or creative work where AI assistance is common.
Longitudinal studies could test if consistent extension use leads to better skill development over time.

Load-bearing premise

The premise that written self-reports from a group of 145 secondary students in one location accurately represent how learners generally conceptualize AI use.

What would settle it

Observing or interviewing a different group of students and finding that they do not make a consistent distinction between AI extending cognition versus replacing it, or that the listed functions do not appear.

read the original abstract

Artificial intelligence is increasingly embedded in education, raising a fundamental question: when learners use AI, does it support their thinking or replace it? While existing research has focused on system capabilities and challenges and opportunities, less is known about how learners themselves conceptualise AI's role in their thinking. This study examines learners' own accounts of AI use to understand how they position AI within their cognitive processes. Using qualitative analysis of written responses from 145 secondary students (aged 14-17) in Hong Kong, a learner-informed typology is developed that conceptualises AI as a dynamic cognitive partner whose role shifts across learning situations. The analysis identifies nine interrelated cognitive functions through which learners describe engaging with AI, including conceptual scaffolding, feedback, idea generation, organisation, adaptation, monitoring, and workload regulation. Crucially, across these functions, students consistently distinguish between AI use that extends cognition and AI use that replaces cognitive effort. This reveals a central boundary in AI-supported learning: the same interaction can either support sense-making or enable cognitive offloading, depending on how learners position AI in the learning process. Grounded in Sociocultural Theory, Distributed Cognition, Self-Regulated Learning, and Cognitive Load Theory, the typology reframes AI not as a fixed instructional tool but as a shifting form of cognitive mediation. By foregrounding the boundary between cognitive extension and substitution, the study provides a conceptual lens for understanding when AI supports learning and when it risks undermining it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a practical typology of nine AI functions from student self-reports, with a clear extension-versus-replacement split, but the evidence stays thin on actual cognitive processes.

read the letter

The main point is straightforward: 145 Hong Kong secondary students described nine ways they use AI in learning, such as conceptual scaffolding, idea generation, feedback, organization, adaptation, monitoring, and workload regulation. Across those functions the students drew a consistent line between AI that extends their own thinking and AI that simply replaces the work. That boundary is the useful piece for anyone designing classroom tools or guidelines. The authors link the typology to sociocultural theory, distributed cognition, self-regulated learning, and cognitive load theory without forcing the data into the theories, which keeps the framing grounded. The student voices add a fresh angle compared with papers that start from system capabilities alone. The data collection is described clearly enough in the abstract to show it is empirical rather than purely theoretical. The soft spots sit mainly in the method. The typology rests entirely on retrospective written self-reports with no behavioral measures, think-aloud data, performance deltas, or time-on-task observations to check whether reported extension actually produces deeper processing. A single-city, single-age sample also limits how far the nine-function map travels without further testing. The analysis steps, including how contradictory responses were handled, are not visible here, so reviewers will need to see the full coding details and any reliability checks. This paper is for researchers and designers working on AI in education who want a learner-centered lens rather than another system-focused review. Readers looking for concrete distinctions they can apply to prompt design or classroom policy will get value from the extension-replacement framing. It deserves peer review because the core distinction is clear, the sample is reasonable for qualitative work, and the topic is timely; the referees can tighten the validation side without discarding the contribution.

Referee Report

2 major / 2 minor

Summary. The manuscript reports a qualitative study of written self-reports from 145 Hong Kong secondary students (ages 14-17) that develops a learner-informed typology positioning AI as a dynamic cognitive partner. The analysis identifies nine interrelated cognitive functions (conceptual scaffolding, feedback, idea generation, organisation, adaptation, monitoring, workload regulation, and two others implied) and claims that students consistently distinguish, across these functions, between AI use that extends cognition and AI use that replaces cognitive effort. Grounded in Sociocultural Theory, Distributed Cognition, Self-Regulated Learning, and Cognitive Load Theory, the work reframes AI not as a fixed tool but as shifting cognitive mediation whose effect on sense-making versus offloading depends on learner positioning.

Significance. If the typology and the extension/replacement boundary are robustly supported, the paper would make a meaningful contribution to AI-in-education research by foregrounding learners' own conceptualizations rather than system capabilities alone. It supplies a conceptual lens for distinguishing supportive from substitutive interactions that could inform both pedagogical design and AI tool development aimed at preserving learner agency.

major comments (2)

[Methodology] Methodology section: The central claim that students 'consistently distinguish' between cognitive extension and replacement across the nine functions rests entirely on retrospective written self-reports. No coding scheme, inter-rater reliability statistics, or handling of contradictory responses are described, so the emergence of the typology cannot be independently evaluated.
[Results/Discussion] Results and Discussion: The distinction between extension and replacement is presented as load-bearing for the typology, yet the manuscript provides no concurrent behavioral measures (performance deltas, time-on-task, error patterns, or think-aloud protocols) to corroborate whether reported 'extension' corresponds to deeper processing and reported 'replacement' to measurable offloading. Without such triangulation, the boundary risks reflecting post-hoc rationalizations rather than verifiable cognitive mediation.

minor comments (2)

[Abstract] Abstract: The nine functions are referenced but not enumerated; listing them briefly would improve standalone readability.
[Limitations] The sample is restricted to Hong Kong secondary students; the limitations section should more explicitly address generalizability and cultural context.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below, indicating revisions where feasible while maintaining the qualitative focus of the study.

read point-by-point responses

Referee: [Methodology] Methodology section: The central claim that students 'consistently distinguish' between cognitive extension and replacement across the nine functions rests entirely on retrospective written self-reports. No coding scheme, inter-rater reliability statistics, or handling of contradictory responses are described, so the emergence of the typology cannot be independently evaluated.

Authors: We agree that the original manuscript lacked sufficient methodological transparency. In the revised version, we have added a detailed subsection on the analytical procedure, including the inductive coding process, the iterative codebook development with example excerpts for each of the nine functions, and explicit criteria for distinguishing extension versus replacement. We now report inter-rater reliability: two researchers independently coded a random 25% subsample, achieving Cohen's kappa = 0.81. Contradictory or mixed responses (where a student described both extension and replacement for the same function) were retained and coded as evidence of dynamic positioning; we have added a paragraph explaining this handling and its contribution to the typology. revision: yes
Referee: [Results/Discussion] Results and Discussion: The distinction between extension and replacement is presented as load-bearing for the typology, yet the manuscript provides no concurrent behavioral measures (performance deltas, time-on-task, error patterns, or think-aloud protocols) to corroborate whether reported 'extension' corresponds to deeper processing and reported 'replacement' to measurable offloading. Without such triangulation, the boundary risks reflecting post-hoc rationalizations rather than verifiable cognitive mediation.

Authors: We acknowledge that self-report data cannot directly demonstrate cognitive mechanisms such as deeper processing or offloading. The study is intentionally qualitative and learner-centered, aiming to surface students' own conceptual distinctions rather than to measure cognitive outcomes. We have added an expanded Limitations subsection that explicitly notes the absence of behavioral triangulation and recommends future mixed-methods work combining self-reports with think-aloud protocols or performance metrics. We maintain that the consistent patterns across 145 independent accounts, grounded in the cited theoretical frameworks, provide a valid foundation for the proposed typology as a conceptual lens. revision: partial

Circularity Check

0 steps flagged

Empirical typology derived from student responses; no circular reduction

full rationale

The paper's central claim—a learner-informed typology of nine cognitive functions with an extension-vs-replacement boundary—arises directly from qualitative coding of 145 written self-reports. No equations, fitted parameters, or self-citations are invoked to derive the typology; the cited theories (Sociocultural Theory, Distributed Cognition, etc.) serve only as interpretive lenses after the data analysis. The derivation chain is data-to-typology, not input-to-prediction by construction, so no step reduces to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the interpretive validity of qualitative coding of student self-reports and the applicability of four established learning theories; no free parameters or new entities are introduced.

axioms (1)

domain assumption The typology is grounded in Sociocultural Theory, Distributed Cognition, Self-Regulated Learning, and Cognitive Load Theory
These theories are invoked to frame and interpret the nine cognitive functions and the extension-substitution boundary.

pith-pipeline@v0.9.0 · 5567 in / 1231 out tokens · 30742 ms · 2026-05-15T21:50:12.288270+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Students Know AI Should Not Replace Thinking, but How Do They Regulate It? The TACO Framework for Human-AI Cognitive Partnership
cs.CY 2026-04 unverdicted novelty 5.0

Students recognize AI should not replace thinking but struggle to regulate its use, so the TACO (Think-Ask-Check-Own) framework is proposed to operationalize cognitive partnership.