Jul 2026· Technology, Knowledge and Learning· 0 citations· 86 references
TL;DR
The findings suggest that GenAI tools like LLAMA may be associated with different patterns of critical thinking during debugging, yet their association with individual skills seems limited in the absence of well-structured instructional support.
Abstract
The growing presence of generative AI (GenAI) has raised questions about its role in supporting or hindering critical thinking in programming education. This study examined the influence of GenAI tools, particularly LLAMA, on students’ critical thinking skills during structured debugging tasks in a Java-based CS1 course. In a quasi-experimental design, 132 students were divided into an experimental group that received GenAI-assisted debugging instruction and a control group that received traditional instructor-led instruction. Both groups completed structured debugging tasks, a closed-book post-evaluation, and a program writing task, while students in the experimental group completed a survey. Students who used LLAMA for debugging tended to link lower- and higher-level thinking skills through stronger conditional associations between understanding code, applying solutions, and extending them into creative solutions, while students who used manual debugging showed stronger connections between analyzing problems and evaluating solutions. Permutation tests confirmed significant differences in skill associations between groups. However, these network-level differences did not always translate into higher performance scores on individual skills, as the control group achieved higher scores in core skills with lower-level cognitive demands. Creativity was also one of the differences observed between the groups. In the LLAMA group, Creativity was connected to Application, with a weaker connection to Analysis, whereas in the control group, it was disconnected from other skills. However, in some cases, students’ solutions in the control group showed structural changes that went beyond what was taught. The findings suggest that GenAI tools like LLAMA may be associated with different patterns of critical thinking during debugging. Yet, their association with individual skills seems limited in the absence of well-structured instructional support. To properly harness GenAI in debugging education, educators must adopt pedagogical approaches that guide students toward reflection, independent reasoning, and balanced cognitive engagement.
Programming students are no longer only learning to write code; they are also learning in environments where AI tools can explain, debug, and generate code alongside them. This shift creates a tension for programming education: the same tools that can make learning more accessible may also encourage dependence when students use them as substitutes for their own reasoning. Using a conceptual and narrative review of recent literature, this paper examines student dependence on generative AI tools in programming education. Central to this review is an examination of learning outcomes, independent programming ability, self-regulated learning, critical thinking, problem solving, learner characteristics, and instructional design in AI-supported programming environments. Students learning to code are increasingly using AI tools to answer questions, explain concepts, help debug, and make information easier to access, but their use can also create problems. Students who rely heavily on them, especially when instructors provide little instruction, may spend less time thinking through problems on their own or reflecting on their solutions. From the literature, we see that the impact of AI is more dependent on the learner, the learning environment, and the use of the tools than on the technology. Existing studies also have important weaknesses, including heavy use of self-reported measures, small sample sizes, correlational research, and limited evidence about what sustained AI use could mean for independent thinking, problem solving, and programming development over time. More data is needed to understand these longer-term effects.
Kazeem Babatunde Abioye· Journal of Education, Learni...· 0 citations
Generative AI has an overall positive short-term impact on higher-level thinking skills, with the most significant improvements in problem-solving, moderate improvements in critical thinking, and relatively low improvements in creativity.
Fareen Wahid· International Journal of Inn...· 0 citations
This study examined the effects of generative artificial intelligence as instructional scaffolding on secondary school students’ critical thinking and classroom engagement. A quantitative quasi-experimental method with a nonequivalent pretest–posttest control group design was applied to 64 eighth-grade students at SMP Negeri 1 Kota Bima, comprising 32 students in the experimental group and 32 students in the control group. The experimental group participated in eight generative artificial intelligence-supported learning sessions involving problem identification, information exploration, response verification, collaborative discussion, problem-solving, and reflection, while the control group received conventional instruction. Data were collected through a 20-item critical thinking test, structured classroom observations, and documentation. The data were analyzed using descriptive statistics, normality and homogeneity tests, normalized gain analysis, independent-samples testing, and effect-size analysis. The experimental group achieved a higher posttest mean score than the control group, with scores of 93.75 and 78.30, respectively. The normalized gain was 0.57 in the experimental group and 0.22 in the control group. The differences in posttest and normalized-gain scores were statistically significant, with a large effect size of 0.87. Classroom engagement also reached 92.11 percent in the experimental group compared with 71.80 percent in the control group. These findings demonstrate that generative artificial intelligence effectively supports critical thinking and classroom engagement when used as guided instructional scaffolding rather than as a provider of final answers.
Herman Herman, Muh. Nasir, A. Putra et al.· Journal La Edusci· 0 citations
This paper presents the development and initial validation of an instrument to measure self-efficacy while using GenAI to learn programming, and finds strong support for the validity of the existing Steinhorst instrument in a new context, specifically an introductory programming course that fully integrates GenAI.
J. Prather, Lauren E. Margulieux, Yekaterina Kharitonova et al.· International Computing Educ...· 0 citations
To investigate the collaborative mechanisms of generative artificial intelligence in programming instruction, this study conducted a 12-week quasi-experiment with 80 higher vocational students. The experimental group used AI assistance, while the control group relied solely on conventional search engines. The results revealed heterogeneous effects of AI usage, with significantly greater intra-group variance than inter-group variance. Three interaction patterns were identified—low-engagement copying, passive debugging, and active constructing—among which only the active-constructing pattern facilitated the internalization of programming thinking. Learning outcomes were optimal when the independent modification ratio was maintained within the $40 \%-60 \%$ range. Students who engaged in active error attribution demonstrated significantly better independent programming performance. Based on these findings, this paper proposes an “attribution- first” human- AI collaborative teaching framework to inform programming education reform.
Xue Wang, Zhi-Yuan Hou, Si-Miao Lang et al.· 2026 IEEE International Conf...· 0 citations
Fractions are a foundational mathematical concept that many elementary students struggle to
understand (Piaget, 1952). Traditional instruction often relies on memorization and repetition,
which reduces engagement and limits conceptual understanding. The platform introduces an
interactive learning platform that utilizes adaptive difficulty and an AI-powered chatbot to
support students. While artificial intelligence tools are becoming increasingly common in
education, most general systems focus on providing answers rather than ensuring students
understand the reasoning behind them (Burns, 2026). The chatbot encourages students to
explain concepts, promoting understanding instead of answer retrieval (Vygotsky, 1978). The
platform includes instructional units, quizzes, and final assessments to measure learning
outcomes. To evaluate the system’s effectiveness, a study was conducted. It involved students in
grades four through six. Fifteen students tested the platform, with eight using the adaptive
chatbot and seven using the standard chatbot. Results showed that students using the adaptive
chatbot achieved an average of 93.5% on the final assessment, whereas those using the general
model achieved only 83.0%. Results indicate that explanation-based AI tools can strengthen
conceptual understanding in mathematics.
Kriti Regandla, Anika Dubey, Dia Rai et al.· International Journal of Soc...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
