How do children's perceptions of machine intelligence change when training and coding smart programs?

Abstract

Children are increasingly surrounded by AI technologies but can overestimate smart devices' abilities due to their lack of transparency. Drawing on the sense-making theory, this study explores how children come to see machine intelligence after training custom machine learning models and creating smart programs that use them. Through a 4-week observational study in after-school programs with 52 children (7 to 12 years old), we found that children engage in the scientific method while training, coding and testing their smart programs. We also found that children became more skeptical of certain abilities of smart devices as they shifted their attribution of agency from the devices to the people who program them. These changes in perception happened both through individual interactions with agents and prompted debates with peers. Based on these results, we conclude with discussions on strategies for promoting children's sense-making practices and sense of agency in the age of machine learning.