Taking Control: How Students Navigate AI in Learning

by

Author: Denis Avetisyan

New research identifies the key ways students actively shape their learning experiences when using artificial intelligence tools.

A grounded theory approach reveals four core dimensions of student agency in AI-assisted learning environments: initiating, adopting, seeking help, and reflecting.

While generative AI holds considerable promise for education, realizing its potential hinges on students’ active, responsible engagement – a quality fundamentally linked to the complex concept of student agency. This study, ‘A Theoretical Framework of Student Agency in AI- Assisted Learning: A Grounded Theory Approach’, addresses a gap in understanding how agency manifests in AI-assisted learning environments through a grounded theory investigation of higher education students. Findings reveal that student agency is characterized by four key, interconnected aspects – initiating & redirecting, mindful adoption, external help-seeking, and reflective learning – representing a proactive and iterative process. How might these empirically-derived insights inform the design of educational experiences that cultivate robust student agency in an age of increasingly sophisticated AI tools?

The Shifting Sands of Learning: AI and the Rise of Student Agency

The educational landscape is undergoing a swift transformation with the burgeoning integration of Generative AI tools, promising a new era of personalized learning experiences. These technologies, capable of tailoring content and pace to individual student needs, are moving beyond simple tutoring systems to offer dynamic, adaptive educational pathways. Current implementations range from AI-powered writing assistants that provide immediate feedback on student compositions, to platforms generating customized practice problems in mathematics and science. While early applications focused on automating tasks like grading and providing basic explanations, the current wave of AI aims to create truly individualized curricula, adjusting in real-time based on a student’s demonstrated understanding and learning style. This shift represents a move away from a ‘one-size-fits-all’ approach, potentially unlocking each student’s unique potential through targeted support and challenge.

The process of learning extends far beyond the passive absorption of facts and figures; true understanding blossoms when students become architects of their own educational journeys. Research demonstrates that individuals retain information more effectively and develop deeper conceptual understanding when they actively participate in defining learning goals, selecting resources, and evaluating their own progress. This active construction of knowledge – shaping one’s learning pathway – fosters critical thinking skills, problem-solving abilities, and a sense of ownership over the material. Consequently, educational approaches that prioritize student direction and self-regulation are proving increasingly vital, enabling learners not just to know information, but to truly understand and apply it.

The increasing prevalence of artificial intelligence in education necessitates a renewed focus on cultivating student agency – the ability of learners to direct their own learning journey. While AI excels at delivering information and even providing solutions, it fundamentally lacks the capacity for genuine understanding, critical thinking, or the ability to connect knowledge in meaningful ways. Consequently, the role of the student is shifting from passive recipient to active constructor of knowledge; students must learn to formulate questions, evaluate information provided by AI, identify gaps in their understanding, and pursue learning paths that align with their individual goals. This proactive approach ensures that learning extends beyond mere memorization and fosters the development of higher-order cognitive skills essential for navigating an increasingly complex world, where the capacity to learn how to learn is as important as the knowledge itself.

Dimensions of Agency: Navigating the AI Landscape

Student agency, when considered in the context of artificial intelligence integration, is not a unitary characteristic but rather a collection of observable capacities. These capacities manifest as specific behaviors, notably the ability to initiate interactions with AI tools – posing questions, requesting information, or assigning tasks – and, crucially, to redirect those interactions. Redirection encompasses modifying prompts, refining queries based on AI responses, and switching between different AI tools or approaches to achieve a learning goal. The frequency and sophistication with which a student demonstrates these initiating and redirecting behaviors serve as indicators of their agency; a student who passively accepts AI outputs exhibits lower agency than one who actively shapes the interaction to suit their needs and learning objectives.

Mindful Adoption, as a dimension of student agency, involves the active and critical assessment of content produced by artificial intelligence systems. This necessitates evaluating AI outputs for accuracy, bias, and relevance to the learning task at hand, rather than accepting them at face value. Integration of AI-generated material requires students to synthesize this content with their existing knowledge, appropriately cite sources, and transform it into original work that demonstrates understanding. Successful Mindful Adoption is characterized by a student’s ability to discern the limitations of AI, identify potential errors, and exercise independent judgment in utilizing AI as a tool for learning and knowledge creation.

External help-seeking, as a dimension of student agency, encompasses a student’s proactive identification of, and access to, supportive resources – including peers, instructors, and online materials – when facing challenges with AI-integrated tasks. This behavior is distinct from simply accepting assistance; it involves a deliberate assessment of need and a targeted search for relevant support. Complementary to this is reflective learning, which requires students to analyze their own learning processes, evaluate the effectiveness of AI tools used, and adjust their strategies accordingly. This includes considering the limitations of AI-generated outputs and identifying areas for personal improvement, ultimately fostering a more nuanced and self-directed approach to learning with AI.

Tracing the Threads of Agency: A Grounded Theory Approach

Grounded Theory is a systematic methodology used for developing theory from data, rather than testing pre-existing hypotheses. This inductive approach involves iterative data collection and analysis, where concepts and categories emerge directly from the observed phenomena. Data is coded line-by-line to identify key themes and patterns, and these codes are then grouped into broader conceptual categories. Theoretical sampling-a process of purposefully selecting data based on emerging concepts-guides further data collection to refine and develop the theory. Constant comparison, a core tenet of Grounded Theory, involves continuously comparing new data with existing codes and categories to identify similarities, differences, and relationships, ultimately leading to the development of a substantive theory grounded in the data itself.

Cognitive interviews were employed as a primary data collection method to understand the internal thought processes of students while they interacted with Generative AI tools. These interviews involved a structured, yet flexible, protocol where participants were asked to verbalize their thoughts, reasoning, and decision-making steps as they completed specific tasks using the AI. The technique focused on “thinking aloud” protocols, allowing researchers to capture not just what students did, but why they did it, revealing their understanding of the AI’s capabilities and limitations. Probing questions were used to clarify ambiguities and explore underlying assumptions, providing a detailed account of the cognitive load, mental models, and perceived agency experienced during the human-AI interaction.

Large Language Models (LLMs) function as the core technology behind a growing number of Generative AI applications, including chatbots, text summarization tools, and content creation platforms. These models, typically based on deep learning architectures with billions of parameters, generate text by predicting the probability of the next word in a sequence, given the preceding text. Our research leveraged LLMs to create interactive scenarios, allowing participants to engage in conversations and tasks mediated by the AI. This context was critical for observing perceived agency, as the LLM’s responses formed the basis for user interpretations of intentionality and control, even though the AI itself lacks consciousness or genuine agency. The ability of LLMs to produce coherent and contextually relevant text enabled a controlled environment to investigate how users attribute agency to these systems.

Prompt engineering was a critical component of this research, involving the careful design of text-based inputs to Large Language Models (LLMs) to elicit specific responses. This technique allowed for systematic manipulation of the LLM’s input, enabling researchers to observe how variations in prompting affected the AI’s output and, consequently, student interactions. By constructing prompts with controlled complexity and specificity, we established a degree of experimental control, minimizing extraneous variables and facilitating the isolation of agency-related behaviors. The prompts were iteratively refined throughout the study to ensure clarity, consistency, and relevance to the cognitive interview questions, thereby enhancing the reliability and validity of the collected data.

The Architect of Learning: Refining a Framework for Agentic Engagement

Recent investigations confirm and broaden the understanding of Agentic Engagement, revealing how students actively and purposefully participate when artificial intelligence is integrated into their learning process. This isn’t simply about students using AI tools, but rather about demonstrating intentionality – initiating tasks, strategically redirecting their efforts when facing challenges, and consciously choosing when and how to utilize AI assistance. The research demonstrates that students aren’t passive recipients of information, but instead exhibit proactive behaviors indicative of self-directed learning, carefully managing their cognitive load and seeking support as needed. This heightened level of agency, fostered by interactions with AI, suggests a shift toward more empowered and effective learning experiences, where students take ownership of their intellectual journeys.

Recent research delineates student agency when interacting with artificial intelligence through four interconnected behaviors. Initiating and redirecting involves students proactively starting tasks and adjusting their approach when AI responses aren’t quite right. Mindful adoption reflects a deliberate choice to integrate AI’s suggestions, rather than passively accepting them, demonstrating critical evaluation. Students also exhibit agency through external help-seeking, knowing when to consult human teachers or other resources to supplement AI assistance. Finally, reflective learning sees students analyzing their learning process-including how they’ve used AI-to improve future strategies. This framework isn’t simply about students having control over AI tools, but about observing these specific actions that demonstrate how they actively manage their learning journey with those tools.

The concept of student agency extends beyond merely providing access to tools or choices; true agency manifests in demonstrably active behaviors. This research clarifies that agency isn’t a passive state of having control, but a dynamic process of exercising it. Students don’t simply possess agency when interacting with AI; they actively demonstrate it through initiating new lines of inquiry, thoughtfully adapting AI-generated suggestions, proactively seeking external assistance when needed, and critically reflecting on their learning process. These observable actions – initiating and redirecting, mindful adoption, external help-seeking, and reflective learning – collectively define an agentic approach, showcasing a student’s deliberate management of their learning journey and a commitment to self-directed growth, rather than a simple reliance on technology.

Recognizing the specific behaviors that demonstrate student agency – initiating learning paths, thoughtfully integrating AI tools, proactively seeking support, and engaging in reflective practice – allows educators to move beyond simply providing access to artificial intelligence and instead cultivate genuinely empowering learning environments. Instructional design can be intentionally structured to prompt these behaviors, offering opportunities for students to define their learning goals, choose appropriate AI resources, ask clarifying questions when needed, and critically evaluate the outcomes of their efforts. This shift fosters a dynamic where students aren’t passive recipients of information, but active architects of their knowledge, ultimately strengthening not only their understanding of the subject matter, but also their metacognitive skills and capacity for lifelong learning.

The pursuit of understanding student agency within AI-assisted learning mirrors the inevitable entropy of any complex system. This study, identifying facets like initiating & redirecting and reflective learning, doesn’t seek to prevent change-the inherent dynamism of a learning environment-but to map its graceful unfolding. As G.H. Hardy observed, “The essence of mathematics lies in its elegance and its logical structure.” Similarly, this grounded theory approach doesn’t impose structure, but reveals the inherent logic of student engagement, acknowledging that even the most carefully designed educational system will require continuous adaptation and, ultimately, refactoring to remain relevant. The arrow of time points inexorably toward this need for iterative improvement.

What Lies Ahead?

This exploration of student agency within AI-assisted learning, while illuminating initial facets of proactive engagement – initiation, mindful adoption, help-seeking, and reflection – ultimately reveals the transient nature of ‘mastery’. The identified aspects of agency are not static attributes, but rather temporary equilibria within a dynamic system. Technical debt accumulates in pedagogical design, much like erosion, demanding constant re-evaluation of what constitutes ‘intentional’ interaction. Uptime, that rare phase of temporal harmony where tools function as intended, should not be mistaken for a fundamental state.

Future work must address the inevitable decay of these agentic behaviors. How do students negotiate agency when AI’s capabilities surpass their own understanding? The study offers a snapshot, but longitudinal investigations are crucial to map the shifting contours of student-AI relationships over time. Furthermore, a deeper consideration of contextual factors – institutional structures, assessment practices, and the evolving affordances of AI itself – is needed.

The field risks fixating on ‘agency’ as a property of the student, rather than recognizing it as an emergent property of the student-AI system. The true challenge lies not in maximizing agency, but in fostering resilience – the capacity to adapt and maintain meaningful engagement even as the technological landscape, and the very definition of ‘learning’, continues to change.

Original article: https://arxiv.org/pdf/2512.07143.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2025-12-10 01:30