Large class sizes pose challenges to personalized learning in schools, which educational technologies, especially intelligent tutoring systems, aim to address. In this context, the ZPDES algorithm, based on the Learning Progress Hypothesis (LPH) and multi-armed bandit AI techniques, sequences exercises that maximize learning progress for each student. Previous field studies showed its learning efficacy compared to a hand-designed curriculum. However, its motivational impact was not assessed. Also, ZPDES did not allow students to express choices: this limitation in agency conflicts with LPH as a model of curiosity-driven learning. We study here how introducing choice (on dimensions orthogonal to exercise difficulty, acting as gamification) impacts both learning efficiency and motivation. We present an extensive field study (265 7-8 years old children, RCT design) showing that ZPDES indeed improves learning performance but also produces a positive learning experience. Combining choice with ZPDES triggers intrinsic motivation and reinforces the learning effectiveness of the LP-based personalization. Conversely, adding choice possibilities to a handdesigned linear pedagogical paths produces deleterious effects on learning. Thus, the intrinsic motivation elicited by choice (gamification) is beneficial only if the curriculum is personalized efficiently for the learner. This deserves attention due to increased use of playful features in educational technologies.