Introducing children to machine learning concepts through hands-on experience

Abstract

Machine Learning (ML) processes are integrated into devices and services that affect many aspects of daily life. As a result, basic understanding of ML concepts becomes essential for people of all ages, including children. We studied if 10--12 years old children can understand basic ML concepts through direct experience with a digital stick-like device, in a WoZ-based experiment. To assess children's understanding we applied an experimental design including a pretest, a gesture recognition training activity, and a posttest. The tests included validating children's understanding of the gesture training activity, other gesture detection processes, and application to ML processes in daily scenarios. Our findings suggest that children are able to understand basic ML concepts, and can even apply them to a new context. We conclude that ML learning activities should allow children to sample their own examples and evaluate them in an iterative way, and proper feedback should be designed to gradually scaffold understanding.