Smart Experiments: AI-Driven Virtual Labs for Science Education Equity

Thesis Abstract

The persistent inequities in science education, particularly in resource-constrained schools, hinder proportional access to authentic inquiry experiences and modern scientific practices, undermining student engagement and achievement. This study investigates the potential of AI-driven virtual laboratories to promote science education equity by providing scalable, high-fidelity experimental experiences that are accessible irrespective of physical laboratory resources. The aim is to evaluate how AI-enhanced virtual labs influence learning outcomes, engagement, and conceptual understanding among learners from diverse socioeconomic backgrounds. Specifically, the study seeks to (1) determine the effect of AI-driven virtual labs on science achievement in comparison to traditional hands-on and standard virtual lab approaches; (2) examine changes in scientific reasoning, experimental literacy, and motivation; (3) explore teachers’ and students’ perceptions of usability, accessibility, and perceived equity; and (4) identify design features that maximize inclusivity and learning gains. A mixed-methods design was adopted, integrating a quasi-experimental component with a concurrent explanatory approach. The population comprised senior secondary school students (n = 480) and their science teachers (n = 40) from 20 schools across urban, peri-urban, and rural districts. The sample was stratified to ensure representation across gender, grade level, and prior achievement. In the experimental arms, students engaged with AI-driven virtual labs that incorporate adaptive scaffolding, natural language interaction, and intelligent feedback for biology, chemistry, and physics experiments, while control groups used either traditional hands-on laboratories or non-AI virtual labs. Data collection instruments included standardized science achievement tests, concept inventories (for each subject area), the Science Motivation Questionnaire II, an experimental design and reasoning assessment, user experience surveys, and teacher interview protocols. Instrument validity and reliability were established through expert review (content validity index > 0.85) and pilot testing (Cronbach’s alpha > 0.78 for all scales). Data collection occurred over a 16-week term, with pre-tests administered one week prior to exposure and post-tests immediately after the intervention. Quantitative analysis employed ANCOVA to compare post-test scores across groups while covarying pre-test performance, with effect sizes reported via partial eta squared. Multivariate regression models assessed the influence of learner characteristics (prior achievement, digital literacy, and socio-economic status) on learning gains, while repeated-measures ANOVA examined trajectory changes in motivation and experimental reasoning. Qualitative data from teacher and student interviews were analyzed thematically using a constant comparative method, with coding framed by constructivist theories of learning and Universal Design for Learning principles to identify factors affecting equity and engagement. The study also triangulated findings with usage analytics from the AI-driven platform, including time-on-task, number of experiments completed, and error patterns. Expected findings include statistically significant improvements in science achievement and concept mastery for the AI-augmented virtual lab group compared with traditional hands-on and non-AI virtual lab groups, particularly among students from historically under-resourced schools. Enhanced scientific reasoning and experimental literacy are anticipated, alongside higher motivation and perceived autonomy due to adaptive feedback and multilingual natural language support. Qualitative insights are expected to reveal that AI-driven features—personalized pacing, error-specific feedback, and equitable access to high-quality experiments—contribute to reduced achievement gaps. However, potential challenges such as initial technological anxiety, platform usability issues, and the need for teacher professional development are anticipated. The study contributes to knowledge by offering empirical evidence on the efficacy of AI-driven virtual laboratories as an equity-oriented intervention in science education, articulating a scalable model for integrating AI-enabled experimentation with standard curricula, and providing design guidelines aligned with Universal Design for Learning. The findings inform policymakers and educators about the feasibility of deploying AI-based virtual labs at scale to democratize access to authentic scientific inquiry, while highlighting critical implementation considerations for sustaining equitable outcomes. The principle conclusion is that well-designed AI-driven virtual labs can attenuate disparities in science education by delivering authentic, scaffolded, and accessible experimental experiences; recommendations include investing in teacher training, ensuring platform interoperability with existing LMS ecosystems, incorporating multilingual support, and conducting ongoing monitoring of equity indicators to guide iterative improvements.