Generative artificial intelligence has swept into lecture halls and dorm rooms across Uganda faster than most universities have managed to write a single policy about it. A new study of 565 respondents, including 52 academic staff and 513 students drawn from five public and private universities, offers one of the clearest statistical portraits yet of how AI tools are reshaping learning within a competence-based education framework, and the results contain a paradox that should stop educators in their tracks. Students who used AI frequently reported significantly better academic performance, yet the same frequent use showed almost no measurable relationship with the deeper competencies that Uganda’s curriculum reforms are designed to cultivate.
The research, published in Discover Artificial Intelligence, arrives at a moment when Uganda’s higher education sector is undergoing a deliberate structural shift. The country’s adoption of Competence-Based Education, or CBE, represents an attempt to move away from content-heavy, teacher-centered instruction toward learner-centered approaches that emphasize critical thinking, creativity, problem-solving, communication, collaboration, and digital literacy. The model, philosophically anchored in constructivism, pragmatism, and realism, expects learners to construct their own meaning through hands-on engagement, inquiry, and the demonstration of performance in real-world contexts. The question the researchers set out to answer was whether AI, with its real-time tutoring and personalized feedback, strengthens this shift or quietly sabotages it.
The study employed a cross-sectional quantitative design across Kabale University, Mbarara University of Science and Technology, Bishop Stuart University, Uganda Martyrs University, and Busitema University, institutions purposively selected to capture variation in infrastructure and pedagogical practice. Participants were recruited through convenience sampling using online platforms, a choice the authors acknowledge may have introduced self-selection and technology-access biases. Data were collected with structured questionnaires measuring institutional readiness, stakeholder perceptions, AI usage patterns, and learner achievement, with internal consistency coefficients ranging from Cronbach’s alpha of 0.78 to 0.91. Analysis combined descriptive and inferential statistics in the JASP statistical tool and Python-based analysis software, testing relationships through Pearson correlation, chi-square tests of independence, one-way analysis of variance, and multiple linear regression.
The first striking finding concerns institutional readiness, which proved uneven in a way that mirrors a global pattern. Technological readiness recorded the highest mean score at 3.2 on a five-point scale, with 60 percent of respondents reporting moderate readiness and 20 percent high readiness, suggesting that foundational digital infrastructure, including internet connectivity, learning management systems, and access to AI-enabled tools, has largely taken root. Pedagogical readiness lagged slightly at a mean of 2.9, with 30 percent of respondents reporting low readiness, indicating that structured approaches aligning AI tools with competency development and formative assessment remain inconsistent. The most alarming gap was policy and governance readiness, which scored just 2.4, with 70 percent of respondents indicating low institutional support. In practice, this means Ugandan universities have the pipes but not the plumbing codes: AI flows freely through institutions that lack frameworks governing academic integrity, data privacy, ownership of AI-generated content, and the validity of competency assessment.
Stakeholder perceptions revealed a genuine dual consciousness about the technology. A substantial 85 percent of respondents agreed or strongly agreed that AI supports the achievement of learning competencies, with a mean of 4.1, while 80 percent agreed that AI enhances self-directed and personalized learning, with a mean of 4.0. Yet simultaneously, 65 percent agreed that AI may undermine independent problem-solving skills, and 70 percent believed it poses risks to authentic competency-based assessment. When educators were asked directly about the challenges they face, academic dishonesty and plagiarism topped the list at 94.4 percent, followed by over-reliance on AI by learners at 88.3 percent and difficulties verifying the originality of student work at 66.7 percent. These are not abstract philosophical worries; they are the daily friction points of lecturers trying to determine whether a submitted assignment reflects a student’s competence or the capabilities of a large language model.
The usage data show just how thoroughly AI has embedded itself in student life. ChatGPT was by far the dominant tool, used by 81.1 percent of students, followed by Google Gemini at 56.8 percent and Microsoft Copilot at 38.8 percent, while specialized research tools such as Elicit, Semantic Scholar, and Connected Papers attracted minimal uptake. More than half of respondents, 54.1 percent, reported using AI tools daily, with a further 18.9 percent using them weekly. The purposes were largely educational rather than fraudulent: 89.2 percent used AI to understand difficult concepts, 81.1 percent for studying and revision, and 70.3 percent for idea generation. Motivations told a similar story, with an overwhelming 91.9 percent citing improved understanding of course content as their reason for adoption, alongside improved academic performance and time savings. Students, in other words, are not adopting AI because it is fashionable; they see it as a practical resource for achieving educational goals.
The inferential statistics, however, expose the paradox at the heart of the study. Pearson correlation analysis revealed a moderate positive relationship between AI use frequency and academic performance, with r equal to 0.544, and a chi-square test confirmed a highly significant association between AI use and academic performance, with chi-square equal to 241.898 across nine degrees of freedom and p less than 0.001. Yet the correlation between AI use frequency and competency attainment was nearly negligible at r equal to 0.045, even though competency attainment itself correlated positively with academic performance at r equal to 0.234. A multiple linear regression model with academic performance as the dependent variable proved statistically significant, F(3, 509) equal to 129.5, p less than 0.001, explaining 43.3 percent of the variance. Within that model, AI use frequency carried the largest standardized coefficient at beta equal to 0.486, followed by responsible AI use at 0.256 and competency attainment at 0.217. Responsible AI use also correlated positively with academic performance at r equal to 0.313, suggesting that ethical and purposeful engagement with the technology matters as much as frequency.
The authors interpret this divergence between performance and competency through two complementary theoretical lenses. The Unified Theory of Acceptance and Use of Technology, or UTAUT, explains adoption through performance expectancy, effort expectancy, social influence, and facilitating conditions, and the study’s findings suggest that environmental factors, including peer trends, encourage learners to adopt AI in ways that respond to fashion rather than pedagogy. The Task-Technology Fit theory, developed by Goodhue and Thompson, holds that technology only improves performance when its capabilities match the tasks users must perform. Applied here, AI can fit some cognitive tasks, such as concept clarification and analytical processing, but may be a misfit for affective and psychomotor competencies, including values, attitudes, teamwork, and real-world practical skills, which cannot be developed or assessed through AI outputs alone. A one-way ANOVA also revealed significant differences in academic performance across levels of study, F equal to 6.194, p equal to 0.002, suggesting that more academically mature students are better positioned to integrate AI tools critically into their work.
From these findings, the study distills a set of pedagogical guidelines that respondents endorsed with striking consensus. AI literacy training received the strongest support, at 80 percent agreement with a mean of 4.2, reflecting the need for learners to critically evaluate AI-generated outputs, formulate effective prompts, and apply tools purposefully. Lecturer capacity building followed at 78 percent, acknowledging that instructors must redesign learning activities and align AI use with competency-based outcomes. Institutional AI policies drew 75 percent support, and process-based assessment approaches, including project-based assessment, portfolios, practical demonstrations, and oral presentations that evaluate reasoning and application rather than final outputs, received 70 percent. The authors emphasize that these strategies are mutually reinforcing: access to technology alone does not produce competency development, and the educational value of AI is determined less by the technology itself than by the pedagogical approaches that guide its use.
The study’s conclusion is ultimately one of conditional optimism. AI has significant potential to enhance competence-based education by supporting self-directed, personalized learning, and the strong association between responsible AI use and academic performance suggests that ethical engagement with the technology pays dividends. But the near-zero relationship between AI use frequency and competency attainment delivers a sobering message: grades can rise while deeper skills stagnate. Within a framework where academic performance is not supposed to equate to the acquisition and demonstration of competencies, universities that simply celebrate improved performance metrics may be mistaking efficiency for mastery. The authors call for longitudinal research tracking AI’s impact on higher-order competencies such as critical thinking, problem-solving, and creativity, alongside experimental studies evaluating structured AI literacy programs and process-based assessment. They also acknowledge the limitations of convenience sampling and self-reported achievement measures, which were not verified against official academic records. For now, the evidence from Uganda suggests that the relationship between AI and competence-based education is complementary rather than contradictory, provided institutions build the governance, capacity, and assessment reforms that transform AI from a potential crutch into a cognitive partner supporting deeper learning.
Subject of Research: AI integration and learner achievement in competence-based higher education in Uganda
Article Title: Artificial intelligence integration and learner achievement in competence-based education in higher education institutions in Uganda
Article References: Mabirizi, V., Nampijja, F. K., Katushabe, C., Kasozi, J. B. M., Muhoza, G. B., Omali, D., & Kawuma, S. (2026). Artificial intelligence integration and learner achievement in competence-based education in higher education institutions in Uganda. Discover Artificial Intelligence, 6(1), Article 1227. https://doi.org/10.1007/s44163-026-02315-8
Image Credits: AI Generated
DOI: 10.1007/s44163-026-02315-8
Keywords: artificial intelligence, competence-based education, higher education, Uganda, academic performance, learner achievement, AI literacy, assessment integrity, ChatGPT, institutional readiness, Artificial, intelligence
News Source: Denise Maddox. (October 19, 2026). AI use linked to better grades but not deeper skills in Ugandan universities. Scienmag.
