Drawing with Le Magique

Abstract

Creativity is central to how children explore and make sense of the world, yet the divergent, playful thinking that underpins it is often marginalised by education systems that favour concrete reasoning, and further eroded by a digital landscape of passive consumption and shortening attention spans. Compounding this, most creative technologies and creative social robots are designed for a narrow age range and short-term use, and cannot evolve alongside a growing child. This thesis asks how an embodied AI system might facilitate children’s artistic expression across childhood, and how its behaviour must adapt across three developmental brackets — 2–4, 5–7, and 8–11 years.

The project followed an exploratory design approach with a concentration on Human Centred Design & Participatory Design. A literature review, expert interviews with teachers, and participatory co-design workshops with children first established how creative and collaborative behaviours differ by age, and translated these into design requirements for an age-adaptive AI system that co-creates with the child. These requirements were realised as a proof-of-concept, Le Magique Drawing: an embodied AI system that reads and analyses a child’s drawing with a vision-language model and projects age-calibrated suggestions from three distinct personalities (additions, realism, and imagination)directly onto the child’s paper, sustaining a continuous “back-and-forth” drawing loop while keeping the child the author of the work.

The prototype was evaluated with children across the three age groups in Wizard-of-Oz sessions. It engaged them strongly and supported their creativity within the act of drawing: they explored beyond the familiar, reworked the system’s suggestions into ideas of their own, and remained the authors of their work. Tellingly, the developmental gradient anticipated in the workshops reappeared in testing, and in places the co-creation visibly matured the drawings themselves. Its value did, however, prove conditional on how well the system adapted to the individual child — a finding that reframes the central design challenge from generating good suggestions to delivering them with personalization, to each child’s intent, and pace. This work contributes an evaluated proof-of-concept and a set of design guidelines for age-adaptive, embodied AI that supports children’s artistic expression, and charts the path to a fully adaptive, embodied next iteration.