AI-Supported Problem-Based Learning for Enhancing Computational Thinking

Synthesis: This quasi-experimental study (N = 103 first-year undergraduates in Nigeria) tested whether AI-supported Problem-Based Learning (AI-PBL) outperforms conventional instruction in building computational thinking (CT) and academic achievement within computer robotics programming. Grounded in Vygotsky's Social Constructivism, the intervention cast AI tools (ChatGPT and Teachable Machine) as "more capable peers" providing adaptive Scaffolding inside students' Zone of Proximal Development across a 14-week, project-based robotics curriculum. The AI-PBL group significantly outperformed the control group on posttest CT and academic achievement after controlling for pretest scores, while gender did not significantly moderate the gains — evidence that AI-enhanced PBL can be both effective and equitable in under-resourced STEM contexts.

Key Findings

AI-PBL students significantly outperformed the conventional-PBL control group on posttest computational thinking, controlling for baseline scores — confirming AI-enhanced scaffolding meaningfully strengthens CT acquisition.
AI-PBL students also achieved significantly higher academic achievement in computer robotics programming, controlling for pretest performance.
Gender did not significantly moderate the instructional effects: both male and female students benefited roughly equally from the AI-PBL approach, countering persistent concerns about gendered outcomes in STEM.
The CT instrument captured five dimensions — abstraction/pattern recognition, algorithm design, decomposition, debugging, and critical reasoning/Metacognition — all of which were targeted by AI-scaffolded robotic tasks (e.g., line-following, assistive, recycling, and search-and-rescue robots).
Grounding AI tools in learners' Zone of Proximal Development (as digital "more capable peers") enabled adaptive Feedback, personalized scaffolding, and iterative problem-solving that traditional lecture-based teaching could not supply.

Study Design & Method

Design: Quasi-experimental, pretest–posttest non-equivalent groups (intact classes; coin toss assigned group); 2 × 2 factorial with instructional strategy and age group as factors, gender treated as covariate. Analyzed with ANCOVA (JAMOVI) plus N-Gain descriptive analysis.
Participants: 103 first-year undergraduates (59 female, 44 male; aged 16–25) across two purposively selected public universities in Southeastern Nigeria, in a compulsory Foundations of Robotics Programming course.
Groups: Experimental (n = 51) received a 14-week AI-PBL intervention (3 contact hours/week, 42 total hours); control (n = 52) received traditional lecture-based instruction on identical content.
AI tools: ChatGPT (idea generation, code explanation, iterative debugging, decision-logic formulation, reflective writing) and Teachable Machine (training image/color classifiers), embedded with Arduino robotics kits.
Authentic robotic tasks: design a line-following robot navigating a dynamic maze; a service robot assisting visually impaired users (HRI, voice recognition, obstacle avoidance); a recycling-sorting robot (supervised ML classification); and a search-and-rescue robot (computer vision + decision trees).
Instruments: 20-item Computational Thinking in Robotics Scale (CT-RS; α = 0.89) capturing abstraction, algorithm design, decomposition, debugging, and critical reasoning/metacognition; 50-item Computer Robotics Programming Skills Test (CRPST; α = 0.91) across motion control, ML for robotics, human–robot interaction, and computer vision.
Limitation noted: the raw source file truncates before the full statistical results tables, so exact effect sizes and N-Gain values could not be reproduced here.

Implications for AI in Education

AI-PBL is a scalable pathway in under-resourced contexts: where expert facilitators and robotics kits are scarce, intelligent tutoring, chatbots, and adaptive feedback can substitute for limited human scaffolding and extend PBL's reach.
AI tools should act as cognitive scaffolds, not answer engines: framing ChatGPT/Teachable Machine as "more capable peers" within the ZPD supports metacognition, iterative debugging, and transferable CT rather than passive consumption.
Equity benefits: the non-significant gender moderation suggests adaptive, inclusive AI features can help narrow traditional gender gaps in STEM, but only if access to AI tools is equitable — infrastructure and teacher readiness remain key barriers.
Call for institutional/policy adoption: the authors urge investment in teacher training, context-aware (localized/adaptive) AI-PBL frameworks, and further research on scalability, with implications for K–12 STEM pathways as well as higher education.

Connected Concepts

Connected Articles

Citation

(2026). AI-supported problem-based learning for enhancing computational thinking. Computers in Human Behavior: Artificial Humans, 7, 100263.