This paper reports the findings of a scoping review on the environmental harms of AI. We reviewed 198 publications to consider the specific environmental harms explored in the literature, the life stages of AI development that are commonly discussed, and the potential solutions that are being researched in the field of AI development. The findings point to a dominant focus on the energy use and greenhouse gas emissions of AI in existing literature, at the expense of other environmental harms, such as water usage, mining, and waste. The literature also predominantly focused on the time during which AI is being used, with less focus on the production and aftermath of AI technology, and the environmental harms associated with these stages of use. The solutions proposed in the literature were wide-ranging, from reducing the carbon emissions of AI models, to producing alternatives to mainstream LLMs. Overall, there was little focus to date on the environmental harms of AI used in education settings in the literature, pointing to a significant gap in research. Drawing on the findings of this review, we suggest future directions for AI research in order to be more critically engaged with the environmental impacts in education policy and practice.
This paper discusses data governance models for educational platforms that could better protect students' and teachers' privacy and reduce data management burdens for teachers and administrators. Within an ethnographic study involving four Swedish secondary schools and municipal- and national-level administrators working with educational technology, we conduct qualitative interviews informed by educational data-tracing activities done with the help of computational methods from infrastructure studies. We highlight the extractive nature of the current data governance model, and demonstrate the power imbalances inherent to it, the uncertainties surrounding big tech companies' data management practices, the burdens placed on teachers and administrators, and the fragmentation of the data/platform ecosystem. This allows us to engage in conceptual analysis of the desirability and feasibility of three alternative data governance schemes as applied to education: personal information management systems (which enable individuals to make data-related decisions), data trusts (expert bodies handling data governance in a fiduciary way) and data cooperatives (which steward data in a decentralised fashion through participatory and democratic decision-making). We conclude that data trusts offer the best response to the concerns identified through our infrastructural and ethnographic study, while also being the most feasible to implement in a variety of school systems.
