AI-Driven Virtual Labs for Agricultural Science Education Assessment

Thesis Abstract

The rapid digitization of agricultural education in higher institutions has created a persistent gap between theoretical instruction and practical competency, particularly in resource-constrained settings where access to physical laboratories is limited. This study addresses the problem by developing and evaluating AI-driven virtual labs as a scalable solution to enhance agricultural science education assessment, with a focus on evaluating procedural mastery, decision-making, and analytical skills in plant pathology, soil science, and irrigation management. The aim is to determine whether AI-enabled virtual laboratories improve students' competencies in experimental design, data interpretation, and performance reliability compared with traditional laboratory instruction. Specific objectives include (1) to design and validate an AI-driven virtual laboratory ecosystem that simulates authentic agricultural experiments with adaptive feedback; (2) to assess learning gains in procedural knowledge, conceptual understanding, and problem-solving abilities using a quasi-experimental design; (3) to examine the accuracy and reliability of AI-generated assessments relative to conventional instructor-based evaluations; (4) to investigate student engagement, motivation, and perceived self-efficacy in virtual lab activities; and (5) to model the relationship between AI-supported assessment outcomes and course performance across multiple modules. A mixed-methods research design is employed, combining a quasi-experimental approach with qualitative inquiry. The study population comprises 320 undergraduate and postgraduate students enrolled in agricultural science programs at three comparable universities. A stratified random sampling procedure yields 160 participants for the treatment group (AI-driven virtual labs) and 160 for the control group (traditional labs). Data collection instruments include (a) a validated Practical Competency Assessment Battery (PCAB) administered as pretest and posttest, (b) a set of AI-generated analytics and scoring rubrics capturing procedural accuracy, data interpretation, and hypothesis testing, (c) standardized concept inventories for plant pathology, soil science, and irrigation, (d) the Technology Acceptance Model (TAM) questionnaire adapted for virtual labs to gauge perceived ease of use and usefulness, and (e) semi-structured interviews with a subsample of 40 participants and 10 instructors for thematic insights. Validity and reliability are established through content validity by a panel of five agricultural education experts, pilot testing with 40 students, Cronbach’s alpha assessments above 0.78 for all scales, and inter-rater reliability for performance scoring (kappa > 0.70). Quantitative data are analyzed using ANCOVA to compare post-intervention achievements between groups while controlling for baseline scores, MANOVA to assess multiple outcome domains (procedural skill, data interpretation, and experimental design), and multiple regression to examine predictors of learning gains, including TAM dimensions and prior achievement. AI-logged interaction data are analyzed with sequence analysis to explore learning trajectories, complemented by hierarchical linear modeling to account for nested data across modules. Qualitative data undergo thematic analysis using an inductive approach to identify emergent themes related to perceived authenticity, cognitive load, and assessment credibility. The study also triangulates results with instructor observations and content analysis of assessment artifacts. Expected findings indicate that students using AI-driven virtual labs will demonstrate statistically significant improvements in procedural mastery (effect size partial ?² ? 0.08–0.15), data interpretation (?² ? 0.06–0.12), and experimental design competencies compared with the control group, with higher engagement and self-efficacy scores. AI-based assessments are anticipated to exhibit strong concurrent validity with human rubric scores (correlation r ? 0.75–0.85) and enhanced reliability (Cronbach’s alpha > 0.80). The qualitative inquiry is expected to reveal enhanced perceptions of realism and immediate feedback as critical enablers, alongside manageable cognitive load when the system offers adaptive scaffolding aligned with Vygotsky’s zone of proximal development. The study contributes to knowledge by (1) operationalizing AI-driven virtual labs as a rigorous assessment modality in agricultural science education, (2) providing empirical evidence on learning gains and assessment validity for AI-enabled practicals, (3) detailing an integrative framework linking adaptive feedback, student engagement, and achievement, and (4) offering design recommendations for scalable deployment in higher education. The main conclusion posits that AI-driven virtual labs can close the gap between theory and practice by delivering reliable, valid, and actionable assessments while enhancing student motivation. Practical recommendations include adopting modular AI lab components across core agricultural domains, investing in instructor training for interpretation of AI analytics, and ensuring alignment with established accreditation standards. Further research should explore longitudinal retention effects and cross-cultural applicability across diverse institutional contexts.