Impact of Educational Convergence with AI in Higher Education (2015–2026): A Systematic Review and Meta‑Analysis
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This project preregisters a systematic review and meta analysis on the impact of educational convergence with Artificial Intelligence (AI)—explicitly including generative AI—in Higher Education. We define convergence as the integration of AI with other areas/disciplines (e.g., STEAM, design, humanities, engineering) through active pedagogies, with a special emphasis on project/challenge based learning (PBL/CBL). The aim is to estimate effects on achievement, computational thinking (CT), creativity, and motivation/interest; to identify moderators of pedagogical and technological design (e.g., the type of AI and, in particular, generative AI); and to derive practical recommendations for e learning and hybrid implementation in universities. • Research question 1 (effectiveness): What is the effect of AI enabled educational convergence—including generative AI—on achievement, CT, creativity, and motivation/interest in Higher Education? • Research questions 2 (moderators): How do effects vary by (a) AI component (tutoring/adaptive/generative/learning analytics), (b) PBL/CBL pedagogy, (c) delivery mode (face to face/hybrid/online), (d) region/country, and (e) dose (duration/intensity)? • Research questions 3 (narrative): What equity/ethics implications are reported (narrative synthesis only; not treated as quantitative outcomes)? Scope and eligibility: • Population: Higher Education students (undergraduate/postgraduate). • Interventions: Convergent learning experiences that integrate AI (including generative AI) with other areas; PBL/CBL is a priority. • Comparators: Conventional/non AI teaching or non convergent AI; where appropriate, pre post with control. • Outcomes: Achievement, CT, creativity, motivation/interest (at least one required). • Eligible designs: RCTs, quasi experimental, pre post with control, and observational studies with comparison groups. • Languages & timeframe: English and Spanish; 2015–2026. • Exclusions: Editorials/opinion pieces without data; K 12 and non tertiary training (except as theoretical context). Information sources and search strategy. We will search Scopus, Web of Science Core Collection, ERIC, IEEE Xplore, and the ACM Digital Library, plus backward/forward citation chasing. Faceted strings will combine AI/convergence/STEAM/PBL terms with Higher Education and outcomes (achievement, computational thinking, creativity, motivation/interest), including an explicit generative AI facet (e.g., “ChatGPT,” “generative AI,” “large language model,” LLM). We will document all database specific queries and the last search date for full reproducibility. Study selection and data extraction. We will conduct dual independent screening (title/abstract and full text) with third reviewer adjudication and record reasons for exclusion. Data will be extracted with a structured template capturing: study metadata, sample, discipline, intervention description using TIDieR (what AI, who, materials, dose, tailoring, fidelity), AI component (including generative AI), PBL/CBL pedagogy, delivery mode, outcome instruments and statistics, design, and funding/conflicts. Methodology We will apply RoB 2 for randomized trials and ROBINS I for non randomized studies, report stratified syntheses by risk level, and conduct sensitivity analyses excluding high risk studies. Quantitative synthesis. For continuous outcomes, we will compute standardized mean differences (Hedges g) with small sample correction; for dichotomous outcomes, OR/RR. We will use random effects models (REML or DerSimonian–Laird), quantify heterogeneity (I², τ²), produce forest plots by outcome, and assess publication bias (funnel plots, Egger’s test, trim and fill). If a study reports multiple effects per outcome, we will apply predefined prioritization rules and, in sensitivity analyses, use robust variance/three level meta analytic approaches where data permit. Pre specified moderators. 1. AI component (tutoring, adaptive, generative, learning analytics/early warning). 2. Pedagogy (PBL/CBL as primary). 3. Delivery mode (online/hybrid/face to face). 4. Region/country, e.g., Organisation for Economic Co‑operation and Development (OECD) versus non OECD. 5. Dose (weeks, study hours, intensity of AI use). 6. Discipline (engineering, education, art design, computer science, etc.; exploratory). Expected outcomes and contributions. • Pooled effect sizes by outcome (achievement, CT, creativity, motivation/interest). • Evidence on when/how generative AI and PBL/CBL maximize effects. • An evidence map (AI component × outcome) and design recommendations (course/program levels). • Open materials on OSF: protocol, search strings, extraction/TIDieR templates, RoB 2/ROBINS I forms, effect size dataset, and analysis scripts for full transparency and reproducibility. • Inputs for institutional guidance on AI supported e learning and for the target Special Issue.
Fields of Study & Subjects
Artificial Intelligence in Education
Artificial Intelligence in Healthcare and Education
Problem and Project Based Learning
Publication details
- Journal / source
- OSF Preprints (OSF Preprints)
- Publication date
- 24 March 2026
- Language
- EN
- Document type
- Preprint