Human–AI co-learning loop: persistent learner-state memory, adaptive question generation, and a transparent evidence view

Design · All designs
Evidence · 2 claims (2 for) · 1 study (1 design), q2 · 0 of 1 report an effect size · 2 claims rest on one study

Description

CoLearn operationalises the ideal tutor as a loop in which the learner practises while the agent refines a persistent memory of the learner's mastery and misconceptions. It has "three components: (i) a persistent learner-state memory that updates per-topic mastery with a soft-evidence variant of Bayesian Knowledge Tracing, where a large language model acts as a continuous observation function", (ii) adaptive question generation targeting the weakest topic and recurring misconceptions, and (iii) an evidence view making personalisation visible and testable. Co-learning here is deliberately modest: the question strategy stays fixed while only the agent's memory adapts.

Design Implications

Context

Requirements

• A persistent per-(learner, subject) memory record so evidence accumulates across sessions rather than per-session logs
• An LLM grader emitting continuous mastery_evidence and evidence_strength values in [0,1] as the observation function

Constraints

• The strategy guideline is a fixed text consulted on every generation; automatic revision from accumulated outcomes is planned but not yet implemented
• A rising mastery estimate is evidence that the agent's belief has changed, not proof that the learner has mastered the skill

Target Learners

• secondary-school science learners (evaluated on Biology and Chemistry curricula)

Learning Goals

• topic-level mastery diagnosis
• misconception identification and remediation through targeted practice

Claims

• Collearn Ab Personalised Question Preference [+M]
• Collearn Adaptive Mastery Mae Convergence [+M]

Related Designs

Examples

-

Key Sources

• Kailai He, Zhihao Wu, Linhai Zhang, Runcong Zhao, Yulan He, Jiazheng Li. (2026). CoLearn: An Agentic Tutor that Learns its Learner in a Human–AI Co-Learning Loop. https://arxiv.org/abs/2609.21154

Page metadata

Field	Value
type	design
id	collearn-human-ai-co-learning-loop
title	Human–AI co-learning loop: persistent learner-state memory, adaptive question generation, and a transparent evidence view
description	CoLearn operationalises the ideal tutor as a loop in which the learner practises while the agent refines a persistent memory of the learner's mastery and misconceptions.
status	review
generated	by process:settle-candidates on 2026-10-10
trust	unverified — no one has confirmed this page's content
sources	1 — kailai-he-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.