Design-Based Research for Developing an AI-Assisted Collaborative Learning Model to Enhance Critical Thinking and Problem-Solving Skills in Higher Education

Synthesis: This Design-Based Research (DBR) study develops and evaluates an AI-Assisted Collaborative Learning (AACL) Model — an eight-week intervention with Indonesian undergraduates working in groups of four to five on authentic problem-based tasks. The model positions generative AI as an "intelligent learning partner" that supports idea generation, analysis, reflection, and collaborative decision-making while students remain the primary decision-makers. Results show substantial learning gains: critical thinking rose 24.1% (68.21 → 84.63) and Problem Solving rose from 70.14 to 88.72, alongside highly positive student perceptions.

Key Findings

• Design-Based Research (DBR) study developing and evaluating an AI-Assisted Collaborative Learning (AACL) Model across a single implementation cycle: an eight-week intervention with undergraduate students at an Indonesian public university, structured around authentic problem-based tasks with students organized into groups of four to five.
• Expert validation by three specialists (educational technology, instructional design, AI in education) rated the model highly valid with an overall score of 92.4% — Learning Design 94.1%, AI Integration 91.6%, Collaborative Learning 93.2%, and Activities 90.7% — with minor revisions focused on collaborative reflection, prompt-design guidelines, and ethical AI usage instructions.
• Critical thinking mean scores rose from 68.21 (SD = 7.34) to 84.63 (SD = 6.28), a 24.1% gain, with the largest improvements in evidence evaluation (66.91 → 85.27, +27.4%) and analysis (67.84 → 84.11, +24.0%).
• Problem-solving performance rose from an initial project average of 70.14 to 88.72 after model refinement; the strongest final criteria scores were team collaboration (90.02) and decision justification (89.41).
• Student perceptions (five-point Likert questionnaire) were highly positive: overall satisfaction 4.47, learning Motivation 4.54, collaborative learning support 4.51, problem-solving support 4.48, critical thinking enhancement 4.43, and ease of AI use 4.36.
• Qualitative analysis surfaced three recurring themes: students viewed AI as an effective brainstorming partner that stimulated idea generation rather than replacing independent thinking; AI-generated responses made collaborative interaction more active; and students became increasingly aware of the importance of verifying AI-generated information against academic literature before group decisions.

Study Design & Method

The study followed a Design-Based Research methodology with four iterative phases — needs analysis, model design, classroom implementation, and model refinement — combining iterative design cycles with continuous evaluation. The model was built on Constructivism learning principles, collaborative knowledge construction, and human–AI collaboration, positioning generative AI as an "intelligent learning partner" (a pedagogical agent) that supports idea generation, information analysis, reflection, and collaborative decision-making while students remain the primary decision-makers. The eight-week implementation used a four-stage learning cycle: (1) problem identification (exploring authentic cases and formulating research questions); (2) AI-assisted collaborative inquiry (groups gather information and compare viewpoints); (3) collaborative problem-solving (analyzing evidence, evaluating AI-generated responses, proposing solutions); and (4) reflection and presentation (peer feedback and reflection on AI-assisted collaboration). Instructors acted as facilitators, monitoring AI usage and encouraging critical assessment of AI content. Data sources included expert validation, pre- and post-intervention critical thinking and problem-solving assessments, structured questionnaires, classroom observations, reflective journals, focus-group discussions, and collaborative project reports. Qualitative data were analyzed thematically; quantitative data used descriptive statistics and paired-sample analysis (specific test statistics not reported in the paper), with findings triangulated across sources.

Key Results

• Model development: needs-analysis interviews found collaborative activities were often limited to information sharing and that generative AI use in instruction was largely unstructured and focused on content generation rather than collaborative knowledge construction — motivating the integrated AACL framework.
• Critical thinking gains: pre/post assessments covering analytical reasoning, evidence evaluation, argument construction, and reflective judgment showed substantial improvement across all dimensions; classroom observations tracked a shift from accepting AI outputs uncritically to questioning accuracy, comparing alternative viewpoints, and supporting arguments with empirical evidence.
• Problem-solving gains: students increasingly used AI to generate multiple perspectives rather than seek single answers, and group discussions became more analytical as learners debated AI-generated recommendations and selected solutions based on academic evidence and collaborative reasoning.
• Implementation challenges: students initially showed varying levels of AI literacy, requiring additional instructional support in prompt engineering and information verification; some groups relied excessively on AI-generated content before gradually developing critical evaluation strategies.

What this means for practice

• Instructional designers. Embed AI inside a structured collaborative cycle rather than adding a tool to an existing group task: this model's four stages — problem identification, AI-assisted collaborative inquiry, collaborative problem-solving, and reflection and presentation — are what produced the critical thinking and problem-solving gains.
• Instructional designers. Write prompt guidance and verification requirements into the activity design, because the needs analysis found AI use was largely unstructured and aimed at content generation rather than collaborative knowledge construction.
• Instructors. Require groups to verify AI-generated information against academic literature before deciding, and to justify decisions with evidence — the criteria on which final projects scored highest (team collaboration 90.02, decision justification 89.41).
• Faculty developers. Front-load explicit AI literacy and prompt engineering support: students began with varying AI literacy and some groups over-relied on AI-generated content until instruction pushed them toward critical evaluation.
• Instructors. Keep AI positioned as an idea-generation and analysis partner with students as the decision-makers, since the gains came from coached collaboration rather than from AI answering the task.

Limitations

• The study reports a single DBR implementation cycle within one course and institution; the participants section does not report a sample size, and the paired-sample statistical analysis is described without reporting test statistics.
• Students' initially varying AI literacy required extra instructional support, and some groups over-relied on AI-generated content before developing critical evaluation strategies.
• The authors call for future iterations with explicit AI literacy training, collaborative assessment rubrics, longitudinal implementation across multiple academic disciplines, and additional research with larger samples and comparative experimental designs to strengthen the empirical evidence.

Citation

Putra, A. D., Wijanarko, F., & Safitri, N. (2026). Design-based research for developing an AI-assisted collaborative learning model to enhance critical thinking and problem-solving skills in higher education.