Source-linked AI summary
Reclaiming Epistemic Agency: A Critical Framework for Human-Generative AI Co-Agency in Education
Biranchi Poudyal
TL;DR
The paper examines how GenAI redistributes epistemological agency and responsibility in education, a shift existing integration frameworks do not adequately address. It evaluates relevant theories and introduces the Ecological Co-Agency Framework, which structures agency relationally, regulatorily, and pedagogically under human epistemic accountability.
Problem
Existing literature does not adequately address GenAI-related agency, power dynamics, data management, or ethical responsibilities in education.
Method
The paper conceptually evaluates distributed agency, Self-Determination Theory, Society 5.0, and technology-integration paradigms before developing the Ecological Co-Agency Framework.
Results
The framework defines agency as relational, regulatory, and pedagogical, constrained by human epistemic accountability through contestability, provenance, and non-delegation of responsibility.
Takeaways & Limitations
The framework gives educators, institutions, and policymakers specific questions for deliberating responsible GenAI use, including task roles, learning-cycle placement, and learners’ ability to challenge outputs.
Takeaways & Limitations
Theoretical analysis provides no empirical evidence that the framework improves student achievement, equity, or teachers’ time, and the targeted search may have missed relevant articles.
Abstract
from arXiv · showhide
Generative artificial intelligence (GenAI) has been primarily framed as an impartial educational tool. However, this framing overlooks an even larger shift: the reassignment of epistemological authority from teachers to students to machines. This paper presents a conceptual evaluation of the extent to which GenAI redistributes students' and teachers' ability to act in classrooms to produce knowledge, validate each other's claims, and create evidence of student learning while collaborating with and competing against humans. This evaluation draws on various theoretical paradigms, including Distributed Agency, Self-Determination Theory, Society 5.0, and Technology Integration Paradigms, including TPACK and SAMR. While all of the theoretical paradigms evaluated are relevant to the role of agency within education mediated by AI, none of them address the ongoing disparity regarding equitable distribution of power, ownership of the data used to mediate interaction, and accountability in relation to human-mediated interactions. As such, this paper introduces the Ecological Co-Agency Framework, which defines agency in terms of relational, regulatory, and pedagogical processes, conditioned by a defined commitment to human accountability for epistemological claims.
1. Introduction
GenAI is reshaping how students and teachers produce and evaluate knowledge, redistributing agency toward non-human systems. This creates epistemological, ethical, and accountability questions that existing educational framing does not adequately address.
- GenAI enables customized learning, collaborative knowledge development, and expanded opportunities for student creativity.
- Educational GenAI introduces concerns about overreliance, authorship, and accountability alongside expanded knowledge production and accessibility.
- GenAI systems can generate, evaluate, and adapt content without direct human involvement, contributing to a redistribution of agency from humans to machines.
- Existing research has focused mainly on interest, participation, teacher workload, and assessment validity rather than GenAI’s epistemological role.
- The paper asks how GenAI affects student and teacher agency and what mechanisms could support equitable, responsible integration into instruction and learning.
2. Conceptual and Methodological Approach
The paper uses conceptual research to synthesize established theories into an original model of GenAI and agency in education. Its literature synthesis was purposeful and iterative, relying on citation chaining rather than formal inclusion and quality-appraisal criteria.
- The study is conceptual research that synthesizes, organizes, and builds on established theory rather than generating new empirical data.
- The paper develops an original model using distributed and relational agency, self-determination theory, self-regulated learning, and technology-integration models.
- The literature search was purposeful and iterative across Scopus, Web of Science, ERIC, and Google Scholar.
- Citation chaining was used to identify relevant studies in a fragmented and rapidly expanding literature.
- The selection process applied no formal inclusion or exclusion criteria and no quality-appraisal tools.
3. Literature Review
The literature review examines GenAI through learner agency, contested intelligence, Society 5.0, and technology-integration frameworks. It finds that existing approaches do not adequately address agency, power, data management, and ethical responsibility in GenAI-supported education.
- Learner and Teacher Agency After GenAI: AI may raise students’ perceived creativity while making their creative work more similar, increasing individual novelty but reducing diversity across the class.
- Learner and Teacher Agency After GenAI: Creativity is framed as guiding creative processes, while autonomy, competence, and relatedness remain necessary despite changed conditions under GenAI.
- Contested Intelligence: Co-intelligence describes joint human-machine cognitive production, whereas consciousness and subjective awareness are presented as uniquely human.
- Contested Intelligence: Technical scalability does not support delegating moral judgment to systems lacking subjective interest in decision outcomes.
- Society 5.0 and the Politics of Human-AI Symbiosis: GenAI tutoring outcomes depend on developer-embedded defaults, while bias, transparency, and intellectual-property ownership are prerequisites for Society 5.0 integration.
- The Proliferation of Integration Frameworks: TPACK, SAMR, and Their Limits: Existing technology-integration approaches include TPACK-based self-regulation models, teacher-development extensions, student interaction categories, and teacher involvement stages.
- The Proliferation of Integration Frameworks: TPACK, SAMR, and Their Limits: Reviewed literature did not treat GenAI integration primarily as an agency issue or address its distinctive power dynamics, data management, and ethical responsibilities.
4. Towards an Ecological Framework of Co-Agency
The Ecological Co-Agency Model integrates relational, regulatory, and pedagogical dimensions within a non-negotiable boundary of human epistemological accountability. It specifies how AI-mediated learning can distribute activity while retaining human responsibility for educational outcomes.
- GenAI-mediated learning involves an external agent, learner planning, action and reflection, and teacher positioning toward technology.
- 4.1 Relational Co-Agency: Relational co-agency treats agency as interactional and requires transparent allocation of tasks between AI systems and students.It also depends on ethical co-agency and human accountability for AI-supported educational outcomes.
- 4.2 Regulatory Co-Agency: Regulatory co-agency maps GenAI support onto forethought, performance, and reflection while keeping students in control of objectives and decisions.GenAI can provide objectives, sources, feedback, prompts, and progress reports, but students and teachers evaluate development using those outputs.
- 4.2 Regulatory Co-Agency: Strategic cognitive offloading can support transformative learning when intentional, whereas routine offloading can prevent resourceful and reflective engagement.
- 4.3 Pedagogical Co-Agency: Pedagogical co-agency positions teachers along an observer-to-innovator continuum shaped by time, training, technology, and administrative support.Without structural support, the continuum can identify already-resourced teachers rather than function as a developmental tool.
- 4.4 The Ethical Boundary Condition: Human epistemic accountability bounds all three dimensions through contestability, provenance, and responsibility for evaluating AI-generated outputs.The framework treats accountability as an ethical boundary condition because high-quality outputs may still produce bias and reduce human responsibility.
5. Discussion: Implications for Policy and Practice
The framework calls for intentional changes in student and teacher roles, AI timing within self-regulated learning, and institutional governance of transparency and responsibility. Policy guidance should convert responsible AI principles into context-specific questions about labor, learning cycles, preparation, contestability, and provenance.
- Role redesign: Students become collaborative knowledge creators, while teachers design methods that cultivate responsible GenAI use rather than merely provide answers.The model anticipates intentional assignment design and systemic cultural change because this transition will not occur automatically.
- Regulatory design: GenAI’s effect on student agency depends on when it enters the self-regulated learning cycle.Information before an attempt may eliminate forethought and increase perceived lack of agency; information after an attempt may support performance and reflection.
- Institutional governance: Transparency requirements make GenAI procurement an exercise of epistemological governance, including disclosure of training-data origins and inherent biases.This requirement may be especially difficult for low-income districts with limited access to resources.
- Policy operationalization: Responsible AI guidance should ask whether task labor, SRL timing, institutional preparation, learner challenge, and system provenance are explicitly addressed.The answers will vary by educational context, so institutions should replace vague appeals to balance with deliberative questions.
6. Limitations and Future Research
The paper is a theoretical contribution rather than an empirical evaluation, and its targeted literature search may have omitted relevant work. Future research is positioned to test the framework across educational settings and examine relationships among its dimensions.
- Limitations: The framework has not yet been empirically evaluated for effects on student achievement, student equity, or teachers’ work time.The study is theoretical and does not provide empirical evidence for these outcomes.
- Limitations: The literature review used a targeted rather than systematic search, potentially omitting articles in other languages, regional journals, or related fields.The paper describes the reviewed literature as substantial but acknowledges limits in search coverage.
- Future research: Future research should measure efficiency and epistemic agency and test relationships among task division, SRL intervention timing, and the framework’s other dimensions.The paper presents these directions as part of an interdisciplinary research agenda.
7. Conclusion
The paper argues that GenAI reallocates agency and responsibility between humans and machines, creating an epistemological concern about whose judgment underlies educational claims. It responds with the Ecological Co-Agency Framework and identifies empirical testing across educational settings as a next step.
- Conclusion: GenAI reassigns agency and responsibility between humans and machines, a shift that much existing technology and adoption literature does not adequately capture.The paper characterizes the central concern as epistemological rather than primarily technical.
- Conclusion: The Ecological Co-Agency Framework defines agency as relational, regulatory, and pedagogical, constrained by human epistemic accountability.Its boundary conditions are contestability, provenance, and non-delegation of moral or intellectual responsibility.
- Conclusion: The framework offers educators, institutions, and policymakers more precise language for addressing human and artificial contributions through deliberation rather than chance.It does not resolve the tensions between AI’s pedagogical utility and the problems discussed in the paper.
- Future research: Researchers plan to develop theoretical models based on the framework and test it empirically across multiple educational settings.The proposed empirical work extends the conceptual contribution beyond theoretical exploration.