Improved Performances and Motivation in Intelligent Tutoring Systems: Combining Machine Learning and Learner Choice

TL;DR

This study tackles the challenge of scaling personalized learning in large classrooms by coupling an intrinsic-motivation–driven learning-progress model with adaptive sequencing (ZPDES) and a contextual choice mechanism. In a large randomized trial with 265 primary-school children, LP-based personalization improved both math learning outcomes and the learning experience relative to a hand-designed linear curriculum, and adding learner choice amplified intrinsic motivation and further learning gains when paired with personalization. However, introducing choice within a fixed linear sequence reduced learning performance, underscoring that gamification effects depend on underlying instructional adaptivity. The results support integrating zone-of-proximal-development–oriented personalized curricula with learner agency to enhance both performance and motivation, and they provide practical guidance for designing ITS that combine adaptive sequencing with meaningful, non-difficulty–coupled choice.

Abstract

Large class sizes challenge personalized learning in schools, prompting the use of educational technologies such as intelligent tutoring systems. To address this, we present an AI-driven personalization system, called ZPDES, based on the Learning Progress Hypothesis - modeling curiosity-driven learning - and multi-armed bandit techniques. It sequences exercises that maximize learning progress for each student. While previous studies demonstrated its efficacy in enhancing learning compared to hand-made curricula, its impact on student motivation remained unexplored. Furthermore, ZPDES previously lacked features allowing student choice, a limitation in agency that conflicts with its foundation on models of curiosity-driven learning. This study investigates how integrating choice, as a gamification element unrelated to exercise difficulty, affects both learning outcomes and motivation. We conducted an extensive field study (265 7-8 years old children, RCT design), comparing ZPDES with and without choice against a hand-designed curriculum. Results show that ZPDES improves both learning performance and the learning experience. Moreover adding choice to ZPDES enhances intrinsic motivation and further strengthens its learning benefits. In contrast, incorporating choice into a fixed, linear curriculum negatively impacts learning outcomes. These findings highlight that the intrinsic motivation elicited by choice (gamification) is beneficial only when paired with an adaptive personalized learning system. This insight is critical as gamified features become increasingly prevalent in educational technologies.

Figures (13)

• Figure 1: The space of available activities always contain only one activity in the predefined linear sequence (Predef) while the space expends over time with ZPDES to allow a diversity of exploration and find the best activities for the learner. The type of activity is represented by a letter (A or B), and the difficulty by a number (1,2 or 3).
• Figure 2: Kidlearn software user interface
• Figure 3: After 15 steps, students working with ZCO and ZPDES are achieving and succeeding more difficult and diverse activities than student working with PCO and Predef conditions. The curves represent the average score over all students for one condition. The shaded area represent the standard error of the mean. Colored points indicate if the score differences are significant two by two for each time step through t-test procedure.
• Figure 4: Students working with ZPDES and ZCO go through the graph of learning activities faster than students working with Predef and PCO. This way, they reach and achieve a larger set of activities. There are $4$ types of activity M, MM, R and RM with their related levels (M: $6$, MM: $4$, R: $4$, RM: $4$). They are ordered here in a colored band displaying a relative difficulty hierarchy to be able to facilitate the visualisation of the students' evolution across activities. Each cells represent the state of an activity for a student at time "t"; white sells for not explored, purple for activity done at time "t"; green for explored activity
• Figure 5: Boxplots presenting the Pre-test scores and the Learning score, i.e the difference between Post-test score and Pre-test score for the four conditions. The Pre-test scores are homogeneous among the populations, assuring a fair comparison. The Learning scores are ordered as follow ZCO $>$ ZPDES $>$ Predef $>$ PCO, giving the order in term of learning efficiency between each condition.