The integration of generative artificial intelligence (AI), such as GPT-based systems, into educational contexts has generated both enthusiasm and apprehension. On the one hand, students gain unprecedented access to instant feedback, diverse perspectives, and customized explanations that can scaffold their cognitive growth. On the other hand, concerns have emerged regarding overreliance on AI-generated content, diminished critical thinking, and the erosion of independent learning capacities.
This study situates itself at the intersection of educational behaviorism and AI-driven pedagogy, asking a central question: how do students exhibit patterns of dependence and autonomy when interacting with GPT? By analyzing real-world dialogue data and student reflections, this research offers empirical insights into how AI reshapes the learning process, and where its promise and pitfalls lie.
The research is guided by three interrelated questions:
What behavioral indicators signify dependence in student-GPT interactions?
Dependence in this study refers to behaviors where students delegate core cognitive tasks—such as problem-solving, idea generation, or critical evaluation—entirely to the AI. Indicators may include direct content adoption without scrutiny, frequent reliance on GPT for basic conceptual clarification, or repeated requests for final answers rather than intermediate reasoning steps. Such patterns reflect a passive stance toward learning, wherein GPT acts less as a partner and more as a surrogate thinker.
What behavioral indicators signify autonomy in student-GPT interactions?
Autonomy in this study refers to active and reflective engagement, where students leverage GPT as a tool to expand rather than substitute their intellectual work. Indicators may include students reformulating GPT outputs in their own words, challenging or questioning AI responses, and strategically employing prompts to seek diverse or contradictory perspectives. Autonomous learners exhibit metacognitive awareness, using GPT as a springboard for deeper inquiry.
How does task type, motivation, and digital literacy influence the balance between dependence and autonomy?
The balance between dependence and autonomy is not static. For instance, when facing unfamiliar topics, students may lean toward dependence as a means of overcoming knowledge gaps, whereas in reflective or creative tasks, they may exert autonomy by blending AI insights with personal reasoning. Motivation and digital literacy further complicate this dynamic: students with strong intrinsic motivation may use GPT more critically, while those lacking digital literacy may misinterpret or uncritically accept AI responses.
The relevance of these questions lies in their alignment with educational behaviorism’s concern with observable patterns of action, as well as their implications for pedagogy in AI-rich learning environments. By identifying concrete markers of dependence and autonomy, educators can design interventions that foster critical, autonomous learning while mitigating risks of cognitive outsourcing.
The study adopts a mixed-methods design, integrating quantitative coding of GPT-student dialogues with qualitative interviews. This triangulated approach ensures both breadth and depth, capturing not only behavioral frequencies but also students’ subjective interpretations of their interactions with AI.
Sixty students from a comprehensive university were recruited, representing both undergraduate and graduate cohorts across disciplines (humanities, social sciences, and STEM). The sample was diverse in gender, age, and digital literacy levels. Participation was voluntary, and all data were anonymized to protect privacy.
Dialogue Corpus: Over 500 GPT-student dialogue sessions were collected across assignments such as essay writing, problem-solving, and exam preparation. Each dialogue consisted of at least 10 turns.
Questionnaires: Students completed surveys assessing their digital literacy, self-regulated learning strategies, and perceived reliance on AI.
Semi-structured Interviews: A subsample of 20 students was interviewed to explore how they perceive GPT’s role in their learning.
Conversation Coding
Dependence Codes: Copy-paste adoption, one-way information requests, minimal paraphrasing.
Autonomy Codes: Prompt diversification, reflective questioning, integration of external knowledge.
Hybrid Codes: Alternation between dependence and autonomy within the same task.
Grounded Theory Analysis
Using open and axial coding, themes were extracted regarding motivational drivers, trust in AI, and strategies for managing uncertainty.
Quantitative Measures
Frequency of dependent vs. autonomous codes per dialogue.
Correlations between digital literacy scores and reliance tendencies.
Comparative analysis across disciplines and academic levels.
To ensure coding reliability, two independent coders analyzed 20% of the corpus, achieving a Cohen’s kappa of 0.87. Member checking during interviews further enhanced interpretative validity.
The data revealed strong evidence of dependence in contexts of high cognitive load, such as essay drafting and technical problem-solving. Many students defaulted to requesting “ready-made” answers, often copying AI-generated texts with minimal edits. In some cases, GPT became a substitute teacher, with students consulting it more frequently than course instructors. Dependence was exacerbated among students with lower digital literacy, who tended to conflate AI fluency with accuracy.
Conversely, a significant subset of students demonstrated autonomous behaviors. They employed GPT as a critical partner, deliberately requesting multiple perspectives, testing counterarguments, and synthesizing AI responses with peer-reviewed sources. For these students, GPT functioned as a catalyst for self-directed exploration rather than a crutch. Interviews highlighted that autonomous learners often viewed GPT as “a brainstorming partner rather than an authority,” reflecting a mature epistemic stance.
Most students exhibited hybrid behaviors, oscillating between dependence and autonomy within the same session. For instance, a student might initially ask GPT to summarize a text (dependence) but later refine their own interpretation by challenging GPT’s summary (autonomy). This hybrid pattern underscores that dependence and autonomy are not binary states but part of a continuum.
Task Type: Dependence was more prevalent in time-pressured tasks, while autonomy surfaced in exploratory or reflective assignments.
Motivation: Intrinsically motivated students displayed higher autonomy, while extrinsically motivated students leaned toward dependence.
Digital Literacy: Students with advanced digital literacy demonstrated stronger abilities to critically interrogate GPT outputs.
The findings suggest that educators cannot simply categorize AI as either beneficial or harmful. Instead, pedagogical strategies should aim to cultivate critical AI literacy, teaching students to distinguish between appropriate reliance and productive autonomy. Possible interventions include:
Embedding AI literacy modules into curricula.
Encouraging students to document their reasoning process when using GPT.
Designing assignments that reward critical evaluation of AI outputs rather than blind adoption.
This study demonstrates that GPT-student interactions embody a dynamic interplay between dependence and autonomy. While some students risk outsourcing cognitive labor to AI, others leverage it as a scaffold for deeper inquiry. The findings underscore that the key determinant is not the technology itself but the learner’s approach, shaped by motivation, literacy, and task context.
From an educational behaviorist perspective, observable patterns of student behavior can inform interventions that encourage reflective, autonomous engagement with AI tools. The challenge for educators is to design environments that promote autonomy while acknowledging the practical utility of AI. Future research should expand cross-cultural comparisons, integrate multimodal data (e.g., eye-tracking), and examine how teacher mediation shapes the dependence-autonomy balance.
Bandura, A. (1986). Social Foundations of Thought and Action: A Social Cognitive Theory. Prentice-Hall.
Deci, E. L., & Ryan, R. M. (2000). Self-determination theory and the facilitation of intrinsic motivation. American Psychologist, 55(1), 68–78.
Gee, J. P. (2017). Teaching, Learning, Literacy in Our High-Risk High-Tech World. Teachers College Press.
Henri, F. (1992). Computer conferencing and content analysis. In A. Kaye (Ed.), Collaborative Learning Through Computer Conferencing. Springer.
Lai, C., & Li, G. (2011). Technology and learner autonomy. Innovation in Language Learning and Teaching, 5(2), 95–110.
Selwyn, N. (2019). Should Robots Replace Teachers? AI and the Future of Education. Polity.
Sweller, J. (1988). Cognitive load during problem solving. Cognitive Science, 12(2), 257–285.
Wertsch, J. V. (1991). Voices of the Mind: A Sociocultural Approach to Mediated Action. Harvard University Press.
