The study is particularly relevant as digital technology becomes increasingly integrated into higher education. AI is no longer used only to search for information. It is also being explored as a tool for providing feedback, helping students understand learning materials, adjusting task difficulty, and monitoring learning progress.
However, the use of AI in education does not automatically lead to better learning. Technology needs to be embedded in a learning design that keeps students actively thinking, discussing, making decisions, and evaluating solutions. PBL is considered a suitable approach because it encourages students to learn through authentic and complex problems.
In this study, AI did not replace lecturers or students’ thinking processes. Instead, the technology functioned as a learning support system by providing feedback, scaffolding, and progress monitoring. Lecturers remained responsible for facilitating learning and supervising the academic process.
Study Involved 120 Undergraduate Students
The study involved 120 undergraduate students, with 60 students from Al Khairiyah University and 60 from Serang Raya University. At each institution, 30 students were assigned to the experimental group and 30 to the control group.
The experimental groups participated in AI-supported PBL, while the control groups received conventional face-to-face instruction based on lectures, textbook exercises, and question-and-answer sessions with lecturers.
Before the intervention, students completed a pre-test to measure their initial problem-solving skills. After the intervention, they completed a post-test to measure changes in their abilities. Academic learning outcomes were also assessed through examinations, while students in the experimental groups completed a questionnaire about their learning experience.
The AI system used in the study was not generative AI or a large language model such as ChatGPT. Instead, the platform used a rule-based expert system and a natural language processing-based feedback engine. It included three main features: the Adaptive Problem Engine, Automated Feedback Module, and Instructor Dashboard.
The Adaptive Problem Engine adjusted problem difficulty according to student performance. The Automated Feedback Module provided feedback on different stages of the problem-solving process. Meanwhile, the Instructor Dashboard allowed lecturers to monitor student progress and provide additional assistance when needed.
Problem-Solving Skills Increased Sharply
The most notable difference appeared in students’ problem-solving skills.
At Al Khairiyah University, the average score of the AI-supported PBL group increased from 52.1 to 78.5, representing a gain of 26.4 points. By comparison, the control group increased from 51.8 to 61.3, a gain of only 9.5 points.
A similar pattern appeared at Serang Raya University. The experimental group increased from 50.9 to 76.2, a gain of 25.3 points. The control group increased from 51.2 to 60.8, representing a gain of 9.6 points.
Across both institutions, the average improvement in problem-solving skills reached 25.85 points in the experimental groups, approximately 2.7 times higher than the 9.55-point average gain recorded by the control groups.
Statistical analysis also showed that the difference was significant. The intervention produced a large effect on problem-solving skills, with a partial eta squared value of 0.418.
The findings indicate that combining authentic problems, group work, adaptive feedback, and lecturer supervision can help students develop their ability to identify problems, analyze information, formulate strategies, implement solutions, and evaluate results.
Academic Learning Outcomes Also Improved
The positive impact was also evident in academic learning outcomes.
At Al Khairiyah University, the experimental group’s average score increased from 63.2 to 82.4, while the control group increased from 62.8 to 71.6.
At Serang Raya University, the experimental group increased from 62.5 to 80.1, whereas the control group increased from 62.9 to 70.9.
Overall, the experimental groups recorded an average learning-outcome gain of 18.4 points, compared with only 8.4 points in the control groups.
Statistical analysis showed that instructional condition had a significant effect on learning outcomes, with F(1,116) = 42.15 and p < 0.001. The effect was also substantial, with a partial eta squared value of 0.267.
Importantly, no significant difference was found between the two universities. There was also no significant interaction between institution and instructional condition. This means that the benefits of AI-supported PBL were relatively consistent at both Al Khairiyah University and Serang Raya University.
Students Viewed AI as Helpful for Learning
The study also examined students’ perceptions of AI-supported learning.
A total of 60 students from the experimental groups completed a questionnaire using a five-point Likert scale. All measured dimensions received average scores above 4.
Feedback relevance received the highest score, at 4.48. Overall satisfaction reached 4.42, while students’ intention to continue using the learning model reached 4.35.
AI prompt clarity received a score of 4.32, while scaffolding effectiveness received 4.21.
These findings indicate that students generally viewed the AI-supported PBL model as clear, relevant, useful, and satisfying. However, the slightly lower score for scaffolding effectiveness suggests that some students may still require additional guidance in understanding or applying the system’s recommendations.
AI Still Requires Lecturer Supervision
Rusman Zaenal Abidin, Maman Fathurrohman, and Aan Hendrayana positioned AI as a supporting tool rather than a replacement for lecturers. The intervention was designed so that lecturers maintained control over the learning process and academic decisions.
This approach is important because the use of AI in higher education also raises concerns about student data protection, system transparency, academic responsibility, and students’ understanding of appropriate AI use.
The study further found that previous technology experience and participation in group activities were associated with problem-solving performance. In the regression analysis, instructional condition was the strongest predictor of post-test problem-solving scores, with a standardized beta of 0.52 and p < 0.001.
The findings reinforce the idea that the success of AI in education does not depend solely on technological sophistication. Learning design, student engagement, digital readiness, and lecturer supervision remain important factors.
Implications for Indonesian Higher Education
For universities, the findings provide a basis for gradually implementing AI-supported PBL, particularly in courses that emphasize analysis, problem solving, and the application of knowledge.
Lecturers need training not only in the use of AI technology but also in PBL design, scaffolding, interpretation of learning data, and ethical supervision of AI-supported learning.
Universities should also establish policies covering student data privacy, transparency in AI use, informed consent, and clear boundaries between AI assistance and students’ own academic work.
The researchers recommend that future studies involve more universities and regions. Long-term research is also needed to determine whether improvements in problem-solving skills remain after the intervention ends.
Future studies could compare rule-based AI with generative AI and hybrid AI models. Cost analysis, lecturer workload, infrastructure requirements, and long-term sustainability should also be examined before AI-supported learning is implemented on a wider scale.
Author Profiles
Rusman Zaenal Abidin, Maman Fathurrohman, and Aan Hendrayana are academics affiliated with the Doctoral Program in Education, Universitas Sultan Ageng Tirtayasa, Banten, Indonesia. Their study focuses on problem-based learning, the use of artificial intelligence in higher education, problem-solving skills, and student learning outcomes.
Research Source
Title: Problem-Based Learning Supported by Artificial Intelligence: An Empirical Study of Its Effects on Problem-Solving Skills and Student Learning Outcomes in Two Indonesian Higher Education Institutions
Authors: Rusman Zaenal Abidin, Maman Fathurrohman, and Aan Hendrayana
Affiliation: Doctoral Program in Education, Universitas Sultan Ageng Tirtayasa, Banten, Indonesia
Year: 2026
Status: Open access under the Creative Commons Attribution 4.0 International license.
0 Komentar