The findings matter as generative AI becomes increasingly embedded in university life. Students now use tools such as ChatGPT and other AI systems to explain concepts, prepare assignments, solve problems, and check academic answers. While these functions can support learning, Lopulalan, Muhibi, and Wikanso warn that AI becomes problematic when it replaces rather than supports students’ own thinking processes.
AI Creates a New Challenge for Independent Learning
Artificial intelligence has transformed how university students obtain information and complete academic tasks. Generative AI can provide immediate explanations, summarize complex material, suggest ideas, and offer feedback on written work.
However, the convenience of instant answers can also encourage students to outsource important cognitive processes. Instead of analyzing a problem, evaluating evidence, or deciding whether an answer is correct, students may simply ask AI to perform those tasks for them.
The researchers describe this pattern as cognitive dependency. It occurs when students increasingly hand over core thinking activities—including problem-solving, answer validation, and learning decisions—to AI systems. Learning autonomy, by contrast, requires students to set goals, choose strategies, monitor their understanding, and evaluate their own learning outcomes.
This distinction is particularly important in Indonesia, where the adoption of AI in education has accelerated alongside broader digitalization following the COVID-19 pandemic. The researchers note that students often adopt AI independently, while institutional guidance on responsible and pedagogically appropriate use may not always keep pace.
Study Examined 312 AI-Using Students
Lopulalan, Muhibi, and Wikanso used a quantitative, cross-sectional survey to examine the relationship between AI cognitive dependence and students’ learning autonomy.
The study involved 312 undergraduate students in Education and Social Sciences programs at state universities in Central Java. Participants were active users of generative AI for academic activities, using tools such as ChatGPT at least two to three times per week during the previous three months.
Students completed a questionnaire using a five-point scale. The researchers measured cognitive dependence through students’ reliance on AI for problem-solving, validating answers, and making learning decisions. Learning autonomy was assessed through self-regulation, critical thinking, and control over the learning process.
The responses were then analyzed statistically to identify patterns and determine whether cognitive dependence on AI was associated with students’ ability to manage their learning independently.
Most respondents were sixth-semester students, representing 42.3 percent of the sample, followed by fourth-semester students at 31.1 percent and eighth-semester students at 26.6 percent. Women accounted for 62.5 percent of respondents and men 37.5 percent.
AI use was widespread among participants. Nearly four in five students, or 78.8 percent, reported using AI more than three times per week, primarily for written assignments, explanations of concepts, and validation of academic answers.
Answer Validation Is the Biggest Warning Sign
The researchers found that students’ average level of cognitive dependence on AI was 3.67 on a five-point scale, placing it in the medium-to-high range.
The strongest form of dependence involved using AI to validate answers.
- Answer validation: 3.82
- Problem-solving: 3.69
- Learning decision-making: 3.50
- Overall cognitive dependence: 3.67
Meanwhile, students’ average learning autonomy score was 3.41, classified as medium. Self-regulation and learning planning scored lower than learning resource management, suggesting that students were relatively capable of accessing resources but faced greater challenges in maintaining reflection and metacognitive control.
The relationship between the two variables was statistically significant. The researchers found a correlation of r = −0.53, p < 0.001, meaning that higher cognitive dependence on AI was associated with lower learning autonomy.
Regression analysis produced a similar result. Cognitive dependence had a significant negative effect on learning autonomy, with β = −0.53. The statistical model indicated that cognitive dependence on AI explained approximately 28 percent of the variation in students’ learning autonomy.
The most concerning behavior was again answer validation. Reliance on AI to determine whether an answer was correct was the strongest predictor of declining learning autonomy, with β = −0.41. Reliance on AI for problem-solving followed with β = −0.29.
The Problem Is Not AI, but How Students Use It
The findings from Politeknik Pelayaran Banten, STAI Syarif Muhammad Raha, and Universitas PGRI Madiun do not suggest that universities should eliminate AI from education.
Instead, the researchers distinguish between two different roles for AI.
AI can act as a cognitive scaffold when it helps students understand difficult concepts, receive feedback, improve drafts, or explore alternative approaches. In this role, the student remains responsible for thinking and making decisions.
AI becomes more problematic when it acts as a substitute for thinking. When students routinely ask AI for final answers and accept those answers without verification, important stages of learning—including planning, reflection, evaluation, and critical reasoning—can be bypassed.
The researchers particularly highlight answer validation because treating AI as a “truth determinant” may reduce students’ willingness to monitor their own understanding and evaluate information independently.
Universities Need Responsible AI Learning Policies
The findings have direct implications for lecturers and higher education institutions.
Universities can encourage students to use AI as a learning assistant rather than an automatic answer generator. Assignments, for example, can require students to explain their reasoning, verify AI-generated information, identify errors in AI responses, and describe how they reached their conclusions.
AI literacy should therefore go beyond knowing how to write effective prompts. Students also need to understand the limitations of AI, question its outputs, verify information, and retain responsibility for academic decisions.
Lopulalan of Politeknik Pelayaran Banten, Muhibi of STAI Syarif Muhammad Raha, and Wikanso of Universitas PGRI Madiun argue that the central challenge is not the existence of AI itself but the way AI is integrated into learning. Their findings support learning designs and institutional policies that encourage critical, reflective, and responsible AI use without replacing student autonomy.
The researchers also recommend further studies using longitudinal or experimental designs to determine whether AI dependence directly causes declines in learning autonomy. Future research could examine the roles of AI literacy, cognitive load, and instructional design in determining when AI strengthens learning and when it undermines student agency.
Author Profiles
Pierre Marcello Lopulalan — Politeknik Pelayaran Banten, Indonesia. He is the corresponding author of the article and focuses on issues related to AI-based learning, cognitive dependence, and student learning autonomy.
La Muhibi — STAI Syarif Muhammad Raha, Indonesia. His contribution addresses AI-supported higher education and the relationship between technology use and independent learning.
Wikanso — Universitas PGRI Madiun, Indonesia. His research contribution is associated with education, learning autonomy, and the implications of emerging technologies for higher education.
The article lists the three authors and their institutional affiliations but does not provide academic degrees or detailed individual fields of expertise, so those details cannot be reliably added beyond the information stated in the source.
Research Source
Article title: Cognitive Dependency in Artificial Intelligence–Based Learning: Its Impact on Students' Learning Autonomy
Authors: Pierre Marcello Lopulalan, La Muhibi, and Wikanso
Journal: Indonesian Journal of Interdisciplinary Research in Science and Technology (MARCOPOLO)
Volume: 4, No. 2, 2026, pp. 91–104
Publication year: 2026
DOI: https://doi.org/10.55927/marcopolo.v4i2.11
Official journal: https://journalmarcopolo.my.id/index.php/marcopolo
0 Komentar