AI-assisted Virtual Labs for Conceptual Chemistry Learning and Assessment

Thesis Abstract

The rapid digitization of chemistry education has amplified the need for scalable, interactive, and assessment-rich learning environments that support conceptual understanding beyond traditional laboratory experiences. Despite widespread adoption of virtual labs, existing platforms often emphasize procedural replication over deep conceptual reasoning, leaving gaps in students’ ability to visualize molecular processes, predict outcomes, and transfer understanding to novel contexts. This study investigates AI-assisted virtual labs designed to enhance conceptual chemistry learning and formative assessment by integrating intelligent tutoring, adaptive feedback, and multimodal data analytics to diagnose and scaffold learner difficulties. The aim of the study is to evaluate the effectiveness of AI-enhanced virtual laboratories in improving conceptual understanding, procedural fluency, and assessment accuracy among undergraduate chemistry students. The specific objectives are (1) to determine the impact of AI-assisted virtual labs on conceptual gains measured by standardized concept inventories and course performance; (2) to examine the role of adaptive feedback and scaffolded hints in reducing cognitive load and time-to-solution in problem-solving tasks; (3) to analyze student engagement and affective responses associated with immersive simulations using actionable analytics; (4) to identify predictive indicators of successful learning trajectories via machine learning models trained on interaction logs; and (5) to assess instructors’ perceptions of the tool’s usability, reliability, and instructional value. A mixed-methods research design will be employed. The quantitative strand will adopt a quasi-experimental design with a treatment group (n=120) using AI-assisted virtual labs integrated into the second-year organic and physical chemistry modules and a comparison group (n=120) following conventional virtual labs. Data sources will include pre- and post-tests comprising a validated Conceptual Chemistry Inventory, course exam scores, and time-on-task metrics gathered automatically from the platform. Multivariate analyses will utilize ANCOVA to compare post-test gains while controlling for baseline performance, and hierarchical linear modeling will assess engagement effects across module units. Regression analyses will identify predictors of achievement, and item response theory will examine the validity of assessment items embedded in the platform. The qualitative strand will involve purposeful sampling of 24 students across groups for semi-structured interviews and four focus groups with eight instructors total. Thematic analysis will be conducted on transcribed interviews to extract perceptions of AI guidance, perceived cognitive load, and of the transferability of learned concepts to novel problems. Interaction logs will be analysed using sequence analysis to map learning pathways, with unsupervised clustering (k-means, silhouette scoring) revealing common trajectories. The theoretical framework integrates constructivist learning theory and cognitive load theory, complemented by the Self-Regulated Learning (SRL) model, guided by the Theory of Multimedia Learning to interpret multimodal instructional design. Expected findings include (i) statistically significant improvements in conceptual understanding and problem-solving efficiency for the AI-assisted group relative to the control group; (ii) reduced cognitive load as evidenced by faster correct solution times and more efficient troubleshooting patterns; (iii) higher engagement and favorable affective responses to simulations, with sustained use over the module duration; (iv) robust predictive models identifying interaction features (e.g., frequency of reflective prompts, utilization of hints, pause-to-think intervals) that forecast achievement; and (v) positive instructor evaluations regarding usability, pedagogical value, and ease of integration into existing curricula. The study contributes to knowledge by (a) demonstrating how AI-driven adaptive feedback within virtual labs can bridge conceptual gaps in chemistry education; (b) providing a validated framework for measuring learning analytics and predictive indicators in AI-supported laboratories; (c) offering design principles for scalable, evidence-based virtual lab enhancements that balance guidance with exploratory inquiry; and (d) informing policy and practice on integrating intelligent tutoring elements into science curricula to support equitable conceptual mastery. Conclusions are anticipated to support adopting AI-assisted virtual labs as a core component of chemistry education innovations, with recommendations including iterative refinement of personalization algorithms, alignment of assessment items with core conceptual frameworks, professional development for instructors on interpreting analytics, and further research into long-term retention and transfer of conceptual reasoning beyond the laboratory context.