Date of Award

Spring 2026

Language

English

Embargo Period

4-15-2028

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

College/School/Department

Department of Educational and Counseling Psychology

Program

Educational Psychology and Methodology

First Advisor

Tianlin Wang

Second Advisor

Gabriel Schlomer

Committee Members

Tianlin Wang, Gabriel Schlomer, Jianwei Zhang

Keywords

AI-based scaffolding, critical thinking, academic writing, Generative AI, feedback uptake

Subject Categories

Educational Psychology | Educational Technology | Instructional Media Design | Language and Literacy Education

Abstract

The rapid integration of generative artificial intelligence (AI) into higher education has fundamentally reshaped the context under which academic writing is produced and evaluated. Large language model (LLM)-based AI systems now offer immediate, adaptive, and interactive feedback, creating new possibilities for supporting writing revision and student learning. Nonetheless, the widespread adoption of AI in academic writing also raises significant concerns regarding students’ potential overreliance on AI-generated feedback. When students accept AI outputs without engaging in evaluative reasoning, justification, or metacognitive regulation, the development of essential higher-order thinking skills, particularly critical thinking (CT), may be undermined.

Such risks highlight the urgent need for pedagogical approaches that move beyond simply permitting or restricting AI use and instead provide structured guidance for critical engagement with AI-generated feedback. Without explicit instructional design, AI may function as a shortcut rather than a cognitive scaffold. Thus, it is necessary to establish pedagogically grounded frameworks that guide students to interrogate, evaluate, and strategically integrate AI feedback into their own reasoning processes. Although national and institutional policies increasingly call for ethical AI use in teaching and learning, empirically validated pedagogical models that systematically align AI-based feedback with the cultivation of CT in academic writing remain limited.

To address these gaps, this dissertation develops and empirically evaluates an AI-based scaffolding framework that repositions generative AI as a Socratic instructional mediator to support students’ critical uptake of writing feedback. Building on scaffolding theory and the Interactive–Constructive–Active–Passive (ICAP) framework, AI-based scaffolding conceptualizes CT as a dynamic, reciprocal process enacted through structured student–AI Socratic dialogue during revision.

This dissertation comprises three interrelated studies. Chapter II employed design-based research to develop and iteratively refine a six-step AI-based scaffolding framework: (1) Preparation, (2) Guided Self-Evaluation, (3) Socratic Dialogue with AI, (4) Critical Evaluation and Inquiry, (5) Reflection-Based Synthesis, and (6) Independent Revision. The study establishes the theoretical coherence, design validity, and usability of the framework, identifying five core critical thinking processes embedded within AI interaction: conceptualizing, inquiring, evaluating, synthesizing, and reflecting. Chapter III examined the causal impact of AI-based scaffolding through a randomized controlled experiment with repeated measures, comparing three conditions: scaffolding with AI, AI without scaffolding, and no AI use. Results indicated that AI-based scaffolding was associated with stronger gains in CT-related writing performance, as well as potential support in CT dispositions, compared to unstructured AI use or no-AI conditions. These findings provide essential internal validity evidence that pedagogically structured AI integration may enhance higher-order thinking beyond performance gains alone. Chapter IV investigated the ecological validity and process-level mechanisms of the scaffolding framework within an authentic undergraduate online course. Using mixed methods, including text analyses of student reflection journals, and thematic analysis of student oral reflections, the study traced how students’ CT trajectories evolved over iterative writing cycles. Findings reveal that AI-based scaffolding can support sustained cognitive agency, facilitating students’ capacity to critically evaluate, negotiate, and transform AI-generated feedback rather than passively adopt it.

Collectively, this dissertation advances theoretical, methodological, and pedagogical contributions by (a) extending scaffolding theory to human–AI interaction, (b) providing longitudinal and causal evidence for AI-integrated instructional design, and (c) offering classroom-grounded guidelines for embedding generative AI as a structured cognitive scaffold in academic writing. By reconceptualizing AI as a pedagogical partner rather than a content substitute, this work contributes to a process-oriented vision of critical human–AI interaction that supports students’ independent reasoning and thinking in AI-assisted learning environments.

License

This work is licensed under the University at Albany Standard Author Agreement.

Download

Available for download on Saturday, April 15, 2028

Share

COinS