Machine (network) learning in K-12 classrooms

Exploring the state of the actual with Actor-Network-Theory

Abstract

In times when machine learning (ML) and other artificial intelligence (AI) technologies are expanding the role and definition of network learning in schools, this short paper reports from a practice-centred research project that explores how K-12 teachers affect and are affected by educational technologies with AI. Accelerated by the COVID-19 pandemic, data-driven and decision-making systems with ML are already entering various educational policy and practice realms, often underpinned by promises of automation and personalization. A growing number of research, drawing from the theoretical orientations and empirical approaches from Science & Technology Studies is increasingly unpacking such promises as well as addressing controversies directly related to the constitutions of ML AI in education. Still, little research explores the adoption of data-driven AI technologies in classrooms from a socio-material, networked learning stance. This short paper introduces such work (in progress) drawing on ethnographic fieldwork conducted in Sweden. Guided by the ontological and methodological approaches of Actor-Network-Theory (ANT), the study focuses on the interactions in K-12 classrooms between commercial ML technologies and teachers. Methodologically this means engaging with both human and non-human actors through ethnographic approaches striving for very specific descriptions of interactions within the actor-network and its enacted realities. Preliminary findings from the first of two envisaged case studies in which a ML-based teaching aid in mathematics was tried out in 22 classrooms indicate how compensatory and contradictory actions and accounts emerge within the network of heterogeneous actors. Human actors seem to compensate for the algorithmic actions of the specific educational technology with ML. This is however not a fait accompli but a continuous and unsettled process in the making between humans and the (non-human) technology. Preliminary results also suggest how controversies of ML algorithms in teaching aids, such as their lack of transparency and algorithmic “governance” play out in authentic learning contexts. In conclusion, the paper argues that theoretical and methodological principles of ANT grant for non-deterministic narrative of the heterogeneous nature of educational practice and have the potential to open the black-box of machine learning in the emerging networked learning settings of K-12 classrooms.