Enhancing Science Inquiry through AI-Driven Virtual Labs for Classrooms

Thesis Abstract

This study investigates the potential of AI-driven virtual labs to enhance science inquiry in classroom settings, addressing persistent gaps in student engagement, inquiry-based skill development, and equitable access to high-quality experimental experiences. The aim is to determine how adaptive AI-augmented virtual laboratories influence students’ inquiry processes, conceptual understanding, and experimental reasoning compared with traditional hands-on labs. Specific objectives include (i) evaluating changes in inquiry-based performance using a validated Inquiries in Science Assessment (ISA) instrument, (ii) examining shifts in science epistemic beliefs as measured by the Schommer Epistemic Style Questionnaire, (iii) identifying the effect of AI feedback granularity on students’ metacognitive regulation during inquiry tasks, (iv) assessing teacher perceptions of usability and instructional fit for AI-driven virtual labs, and (v) analyzing potential differential effects across gender and prior achievement. The study adopts a quasi-experimental design with mixed methods, enrolling two matched groups of 240 secondary science students (120 in AI-Driven Virtual Lab classrooms and 120 in conventional labs) across six urban middle schools over a 12-week term. The AI-driven condition employs an adaptive virtual lab platform that provides real-time scaffolding, hypothesis generation prompts, and data-driven feedback using machine learning to tailor task difficulty. Data collection instruments include the ISA, a validated science reasoning assessment, a pre- and post-test on content knowledge, classroom observation rubrics, think-aloud protocols from a subsample of 30 students, and teacher interview guides. Validity and reliability are established through pilot testing (n=60) andCronbach’s alpha coefficients above 0.80 for primary instruments; instrument alignment is confirmed via expert review. Quantitative analyses use ANCOVA to compare post-test outcomes while controlling for pre-test scores, multiple regression to explore predictors of inquiry performance, and item response theory (IRT) to validate assessments. Mediation analysis tests whether metacognitive regulation mediates the relationship between AI-driven feedback and inquiry performance. Descriptive statistics summarize engagement and time-on-task metrics captured by the platform. Qualitative data from think-aloud protocols and teacher interviews undergo thematic analysis to triangulate findings and illuminate mechanisms of effect, with coding guided by the dual-process theory of inquiry and the growth mindset framework. A cross-case synthesis examines contextual factors such as classroom culture and teacher professional development experiences. Expected findings indicate that students in AI-driven virtual labs exhibit significantly higher post-test inquiry scores (p < .05, ?2 = .08) and greater improvement in hypothesis formation, experimental planning, and data interpretation than those in traditional labs. The AI condition is anticipated to enhance metacognitive regulation, reflected in increased planning and monitoring actions during inquiry tasks, and to shift epistemic beliefs toward more sophisticated, tentative scientific reasoning. Differential analyses may reveal larger gains for students with lower initial achievement, suggesting a compensatory effect of AI scaffolding. Classroom observations are expected to show higher engagement, more equitable participation, and smoother management of complex experiments through AI-guided pacing. Teachers’ perceived usability and instructional alignment are anticipated to be favorable, with identified needs for professional development in data interpretation and ethical use of AI. The study contributes to knowledge by delineating the instructional affordances and boundaries of AI-driven virtual laboratories in supporting inquiry-based science education, articulating a theoretical model that integrates the dual-process perspective with AI-mediated scaffolding, and providing empirical evidence on equity-enhancing outcomes. It offers practical guidelines for designing scalable AI-enhanced laboratory experiences, informs policy on technology integration in science curricula, and highlights implications for teacher professional development. The main conclusion posits that AI-driven virtual labs can substantially bolster science inquiry skills and conceptual understanding when embedded within thoughtfully designed instruction and targeted professional support; recommendations include iterative platform refinement, ongoing teacher training, and longitudinal studies to assess long-term retention and transfer to higher-order scientific reasoning.