AI Adaptive Learning and Assessment System: A Pilot Evaluation of an AI-Powered Adaptive Quiz and Feedback Platform

This paper presents a pilot implementation of an Adaptive Learning and Assessment System, an online system where students upload their learning resources, and receive automatically generated and difficulty adjusted quiz with topic level feedback. As opposed to a fixed question bank, the system generates questions on the fly directly from a learner's own material, dynamically adjusts the difficulty level of the questions according to real-time performance data and creates a summary at the end of the quiz to show a learner what to focus on when studying the topics they're not mastering, and produces a downloadable report. A system evaluation questionnaire was given to 32 users after their last quiz session to assess their perceptions of the system through the use of an eight-item dimensional Likert-scale instrument that measured understanding, adaptive difficulty, feedback quality, weak-topic identification, trust, engagement, ease of use and preference over traditional self-study. Using response-frequency distributions, dispersion, correlation, and reliability statistics (Cronbach's alpha), the responses were analyzed. When it comes to ease of use and feedback quality, the system received the highest rating from the students, while it was rated the lowest, but still fairly positively, for their trust in the AI-generated content. The correlation analysis results indicated that adaptive difficulty and engagement were the most strongly correlated items in the instrument; trust and preference to traditional study were a second well-correlated group that indicated that usability and trust are evaluative dimensions that are distinct for this system. These results suggest that the platform is usable and well-received as is, and that the need for trust in content generated by AI is your apparent top priority for future development. Since it is a small-sample, single group pilot study with no control condition, these results are not causal, but are meant to encourage a larger controlled follow-on study.