Rapid Academic Levelling Through Artificial Intelligence Role Play Escape Room

Abstract

Gaps in previously taught probability and statistics concepts can present difficulties when engineering students enter courses that rely on this prerequisite knowledge. Educational escape rooms provide one possible mechanism for revisiting such concepts through contextualised problem-solving within a limited amount of classroom time. This exploratory study examines aggregate changes in assessment performance and response behaviour surrounding a single-session conversational artificial intelligence (AI) educational escape-room intervention with second-semester engineering students in Mexico. The complete cohort of 15 students enrolled in an Engineering in Interactive Software and Video Game Development programme participated voluntarily. Participants were aged 18–22 years; two identified as women, twelve as men, and one as non-binary. The one-hour classroom session consisted of a 10-minute pre-test, a 5-minute introduction, a 35-minute individually completed AI-mediated escape room, and a 10-minute post-test. The intervention, Probabilidades Sangrientas: El Castillo de Strahd, was implemented as a customised GPT-based conversational experience containing 20 open-response probability and statistics challenges embedded within a gothic escape-room narrative. Fifteen pre-test records and fourteen post-test records were available for analysis; individual responses could not be matched. Consequently, analysis was restricted to descriptive aggregate comparisons. The proportion of attempted items increased from 40.0% to 68.6%, while correct responses as a proportion of all possible items increased from 19.3% to 55.7%. Among attempted items, correctness increased from 48.3% to 81.3%. These differences indicate improved immediate aggregate performance and greater willingness to attempt assessment items after the session. However, because the study lacked a control group, matched individual observations, and a delayed assessment, the findings cannot establish individual learning gains or attribute the observed changes specifically to the escape-room or AI components. The study instead provides an exploratory account of how a conversational AI escape room can be incorporated into a time-constrained engineering classroom for prerequisite-knowledge reinforcement.