Democratising AI education: Teaching autoencoders to out-of-school children from low-income backgrounds

Abstract

Artificial Intelligence (AI) has become an integral part of personal computing devices and is gaining importance in school curriculums. Several studies and tools have been developed for teaching AI to school-aged children and it continues to be a growing area of research in computer science education. However, most of these interventions are targeted towards the more privileged and therefore cannot be directly used with children from socio-economically backward families. Children from such backgrounds are often unable to complete even basic elementary education due to several social, economic and systemic challenges. This work aims to help extend the reach of AI education to these less privileged children. A learning activity based on the five AI4K12 big ideas is presented, the activity follows a series of sessions aided by an easy to create tool for supporting tinkerability. The activity attempts to give a high-level overview of the functioning of autoencoders. The feasibility of this approach is verified with a case study involving out-of-school children from low-income families. The findings show that all of the AI4K12 five big ideas can be addressed and effectively taught to non-English speaking children with low numeric literacy. The proposed approach was also shown to increase children’s curiosity and sense of agency while learning about AI.