Abstract:

The importance of Machine Learning (ML) is evident through everyday life activities and it affects the lives of most people. At the same time, ML systems do not learn effortlessly. First, they need data, and lots of it, and today's world is commonly viewed as increasingly run by data. Secondly, ML systems work in a new way, they can evolve. Nevertheless, today’s computing education continues to be dominated by teaching rule-based programming, while the much-needed shift from rule-driven thinking to ML-based data-driven thinking is on the agenda.

The critical problem of integrating ML in education is the significant difference between the basic principles of ML-based and traditional, rule-based, programming. While in traditional coding, the computer was perceived as a means of executing user-created software, in the ML-based case, the computer becomes a learning device, and creating a software product involves a data-driven learning process. It means that the very principle of the learning environment in which the software is created changes. Traditional methods for studying the effectiveness of new technologies in education are usually based on the assessment of habitual programming skills. The present paper suggests new methods suitable for the ML era. It proposes to analyze particular types of relations of students with digital technology and, through it, with the world around them. These relations are considered postphenomenological relations.

Our study focuses on the various ways in which ML technologies mediate between human beings and the world, a vital and often neglected aspect of educational practice. We use postphenomenology, a branch of philosophy of technology that highlights the importance of technological mediation of experience that helps researchers explore technological mediation.

This paper describes findings of interviews conducted with 5th and 6th-grade students at the end of two yearlong courses: "Rule-driven construction" and "Data-driven construction." Two main questions initiated the present study:
1) What are the differences between data-driven learning and traditional, rule-based programming, regarding students' computational perspectives?
2) How to assess students' computational perspectives in the realm of ML-based data-driven thinking?

ML learning in elementary school is in its infancy. A reliable assessment tool for students' computational perspectives in the realm of ML-based data-driven thinking is absent from the current literature. Therefore, introducing postphenomenological methods will be valuable for researchers and practitioners alike.

Keywords:

Elementary school, Computational thinking, Machine learning, Postphenomenology, Constructionism.