Rapid Academic Levelling Through Artificial Intelligence Role Play Escape Room

This paper presents an exploratory demonstration of a single-session, GPT-based AI escape room designed to level up probability and statistics prerequisites for second-semester engineering students. Its central purpose is to show how a conversational AI role-play escape room can be integrated into one class period, rather than to establish that the intervention improves learning. The authors are explicit that the work does not demonstrate effectiveness, retention, or any distinctive contribution of conversational AI.

The study involved 15 engineering students who completed a 35-minute AI escape room, with an anonymous pre-test and post-test used for comparison. Descriptive aggregate results showed substantial increases: the attempt rate rose from 40.0% to 68.6%, correct responses across all possible items rose from 19.3% to 55.7%, and accuracy among attempted items rose from 48.3% to 81.3%. These figures are large on their face, but the design cannot support causal or individual-level conclusions.

The limitations are substantial and clearly stated. There was no control group, individual records could not be matched between administrations, the post-test contained one fewer record than the pre-test, and no delayed assessment was conducted. As a result, the observed changes cannot be attributed to the escape room or the AI component, nor can they be interpreted as individual learning gains.

The paper closes by calling for controlled studies using validated instruments and delayed outcome measures. It also documents ethical procedures, including voluntary participation, adult consent, anonymised data, and the absence of any effect on course grades, and it discloses the use of ChatGPT and Scite in manuscript preparation alongside a review of supporting literature on escape rooms and game-based learning.