ETD

• causal inference
• econometrics
• microeconometrics
• OLS
• propensity score matching
• statistics
• survey

This thesis examines how Artificial Intelligence (AI) tools, especially generative chatbots like ChatGPT, influence students’ performance in Italian universities. While AI can improve productivity, self-efficacy, and reduce stress, critics warn of over-reliance, superficial learning, and weaker critical thinking. Using survey data from 113 students, the study explores whether frequent AI use correlates with Grade Point Average (GPA). Nearly all respondents use AI, but patterns differ: “Frequent” users show some of the highest GPAs, while “Very Frequent” users record lower medians and greater variability.
OLS regressions reveal that very frequent users score about 1–1.5 GPA points lower than frequent users, even after controlling for demographics, degree type, exams, and enrolment year. Master’s students, on average, perform significantly better than Bachelor’s students, but interaction models show no clear evidence that AI’s impact varies by degree type.
To approach causality, propensity score matching compares very frequent users with similar peers. ATT estimates indicate roughly a 1-point GPA penalty for heavy users, though statistical significance is limited by sample size.
Overall, findings suggest moderate AI use may support learning, while overuse can harm performance, consistent with cognitive load theory. Despite limitations—small sample, self-reported data, possible confounding—the thesis contributes early evidence from Italy, underscoring the need for balanced, guided AI integration in higher education.