Agency in Educational Technology: Interdisciplinary Perspectives and Implications for Learning Design

📝 Paper Summary

Personalization (P13N) User modeling

The paper synthesizes perspectives from philosophy, education, and psychology to propose an adaptive 'agency personalization loop' where EdTech dynamically adjusts learner control based on their skills and progress.

Core Problem

Designers of educational technology (EdTech) struggle to determine the right balance of agency: how much control should be automated by AI vs. given to the learner?

Why it matters:

Personalized EdTech often automates choices to optimize efficiency, potentially sacrificing learner agency and motivation.
Young learners (children) have immature executive functions and may make ineffective learning choices if given full autonomy too early.
Existing evaluation rubrics for EdTech quality rarely consider agency design, leaving practitioners without guidance on how to support learner autonomy.

Concrete Example: A student using reading software might be automatically assigned the next text based on an algorithm (zero agency), whereas an open storytelling app lets them choose everything (full agency). Neither extreme is optimal: the former inhibits self-regulation skills, while the latter can overwhelm a novice learner with poor choices.

Key Novelty

Agency Personalization Loop

Conceptualizes agency not as a binary on/off switch, but as a dynamic parameter that EdTech should adaptively assign to learners.
Proposes a feedback loop where the system assesses learner characteristics (like prior knowledge and executive function) to set an initial agency level, then updates it based on performance.
Integrates interdisciplinary views: Philosophy's normative types of agency, Education's focus on teacher-student dynamics, and Psychology's findings on self-regulation and motivation.

Breakthrough Assessment

7/10

Provides a strong theoretical synthesis and a novel design framework (the personalization loop) for a critical, under-discussed aspect of EdTech, though it remains a conceptual proposal without empirical validation.

⚙️ Technical Details

Problem Definition

Setting: Designing adaptive educational technology systems for children that balance automated personalization with learner autonomy.

Inputs: Learner characteristics (prior knowledge, executive functions, metacognition) and real-time performance data.

Outputs: Adaptive assignment of agency levels (e.g., who controls task selection, content, or sequence).

Pipeline Flow

Assessment of Learner Characteristics (Student Model)
Algorithmic Assignment of Agency Level
Task Execution & Performance Monitoring
Student Model Update

System Modules

Student Model

Stores and updates learner state including prior knowledge, executive function skills, and metacognitive abilities.

Agency Assignment Algorithm

Determines the optimal level of agency (who controls what) for the current step based on the Student Model.

Learning Interface

Presents the learning activity with the assigned constraints and choices.

Novel Architectural Elements

Explicit parametrization of 'Agency' as an adaptive variable within the personalization loop, distinct from just content difficulty or pacing.

Comparison to Prior Work

vs. Standard ITS: Proposes varying the *control mechanism* itself (agency), not just the content difficulty.
vs. Pure Shared Control: The Agency Personalization Loop is dynamic over time—agency can increase as the learner demonstrates improved self-regulation skills (developmental trajectory).
vs. Static EdTech Design: Treats agency as a continuous, adaptable variable rather than a fixed feature of the software [not cited in paper, general practice comparison].

Limitations

The proposed framework is theoretical and has not been implemented or empirically tested in a fully functioning system within this paper.
Measuring prerequisites like 'agency beliefs' or 'metacognitive skills' in real-time for the loop is technically challenging.
Does not address the ethical implications of algorithms deciding when a human is 'ready' for agency.

Reproducibility

No replication artifacts mentioned in the paper. The paper is a theoretical review and framework proposal; it does not present a specific software implementation or release code/data.

📊 Experiments & Results

Evaluation Setup

The paper is a conceptual review and does not contain an empirical evaluation section with quantitative results.

Metrics:

Statistical methodology: Not explicitly reported in the paper

Main Takeaways

Agency is multi-dimensional across disciplines: Philosophy emphasizes types (epistemic, moral), Education focuses on teacher-student-tech dynamics, and Psychology highlights the link between control, motivation, and self-regulation.
Agency supports learning only when learners have the prerequisites (knowledge, executive functions) to use it effectively; otherwise, too much agency can hamper learning.
Effective EdTech should use an 'Agency Personalization Loop' to adaptively grant control: starting with more scaffolding for novices/younger children and releasing control as skills develop.
Designers can manipulate agency via four dimensions: What is chosen (content, sequence, appearance), Who chooses (student, system, shared), When (start, during, end), and Where (dashboard, activity).

📚 Prerequisite Knowledge

Prerequisites

Basic understanding of Self-Regulated Learning (SRL)
Familiarity with Adaptive Learning Technologies or Intelligent Tutoring Systems
Concepts of Executive Functions (inhibition, working memory, cognitive flexibility)

Key Terms

Agency Personalization Loop: A proposed theoretical cycle where EdTech systems adaptively assign the level of control (agency) to a learner based on their current skills and performance history.

Epistemic Agency: Agency sensitive to norms regarding knowledge, justification, and evidence assessment (e.g., deciding when a belief is justified).

Executive Functions: Cognitive processes including inhibition, working memory, and cognitive flexibility that enable goal-directed behavior and self-regulation.

Shared Control: A design paradigm where control over learning decisions (like task selection) is distributed between the system and the learner, often by having the system pre-select a subset of suitable options.

Self-Regulated Learning: A process where learners consciously set goals, plan strategies, and monitor their progress; often requires support or scaffolding for younger learners.

Expertise Reversal Effect: The phenomenon where instructional methods that are effective for novices (high guidance) become ineffective or detrimental for experts, who benefit from more autonomy.

Metacognition: Awareness and understanding of one's own thought processes, crucial for monitoring learning progress and adjusting strategies.