Developing an AI-Enhanced Virtual Lab for Biology Education Assessment

Thesis Abstract

The study addresses the persistent gap between theoretical biology instruction and practical skill development in authentic assessment by developing an AI-enhanced virtual laboratory (AIVL) to augment biology education assessment. This work situates itself within the context of high-environmental fidelity digital simulations and intelligent tutoring systems, aiming to provide scalable, evidence-based assessment of both procedural mastery and conceptual understanding in biology laboratories. The aim is to evaluate the effectiveness of an AI-enhanced virtual lab in improving student learning outcomes, lab skills, and assessment validity compared with traditional hands-on laboratories. The specific objectives are (1) to design and implement an AI-driven virtual lab that incorporates real-time feedback, adaptive difficulty, and multimodal data capture (including action logs, sensor-like virtual measurements, and student annotations); (2) to examine the impact of the AIVL on learning gains in procedural competencies (pipetting accuracy, experimental planning, data interpretation) and conceptual understanding (hypothesis formulation, control of variables, data justification) among undergraduate biology students; (3) to validate an automated assessment framework combining rubric-based scoring, performance analytics, and machine learning-derived proficiency models; (4) to investigate student engagement, cognitive load, and perceived self-efficacy when interacting with AI-driven simulations; and (5) to explore the transferability of skills from the virtual lab to physical wet-lab performance. The methodology adopts a mixed-methods research design guided by constructivist and socio-constructivist learning theories, with Vygotsky’s Zone of Proximal Development informing the scaffolding strategy and Kolb’s experiential learning cycle shaping assessment hooks. A quasi-experimental design will be conducted at a mid-sized university with two intact biology cohorts (N = 240 total; 120 per group). The treatment group (n ? 120) will engage with the AIVL across core modules (e.g., DNA extraction, enzyme kinetics, cell culture simulations) over a 12-week term, while the control group (n ? 120) completes equivalent learning activities through conventional virtual labs and paper-based assessments. Data collection instruments include (i) standardized pre- and post-tests measuring procedural fluency and conceptual knowledge, (ii) automated analytics from the AIVL capturing action sequences, timing, and decision points, (iii) validated laboratory skills rubrics (aspects such as accuracy, precision, and justification), (iv) adapted NASA-TLX scales to assess cognitive load and engagement, and (v) semi-structured interviews and focus groups with a purposive sample of 20 students and 6 instructors. Validity and reliability of instruments will be established through expert review, pilot testing (n = 30), and Cronbach’s alpha checks (target ? ? 0.80 for the knowledge and skills scales). Data analysis will employ a multi-layer approach descriptive statistics and normality checks will characterize the data; inferential statistics will include ANCOVA controlling for prior achievement to compare post-test outcomes, and multivariate regression to model the relationship between engagement metrics, cognitive load, and learning gains. For the qualitative component, thematic analysis will be conducted on interview transcripts, guided by Braun and Clarke’s framework, triangulated with log data to extract corroborating patterns. The study anticipates that students using the AIVL will exhibit statistically significant improvements in procedural fluency and conceptual understanding (p < 0.05), with higher engagement and lower cognitive load during complex tasks compared to the control group. It is expected that automated proficiency models built on performance logs will closely align with instructor-rated outcomes, thereby enhancing assessment validity. The contribution to knowledge includes demonstrating a scalable, AI-assisted assessment model that integrates objective performance data with expert-driven rubrics to provide real-time, formative feedback, and establishing evidence for the equivalence or superiority of AI-enhanced simulations over traditional virtual labs in biology education. The research will offer practical recommendations for curriculum designers, instructional technologists, and policymakers on adopting AI-enabled virtual laboratories to augment laboratory education, reduce resource constraints, and improve equity of access to high-quality practical assessment. The study concludes with implications for future work on cross-disciplinary applicability, longitudinal retention of skills, and the refinement of governance frameworks for AI in educational assessment.