Send message to Authors

Paper short abstract

Understanding AI policy documents by international organisations as discursive infrastructure, this paper interrogates how such policies bring into being a specific reading of this technology, direct future investment, and thereby constrain the possibility of tackling the environmental crises.

Paper long abstract

As the environmental crises continue, the AI hype unfolds in business and policy circles. Previous literature suggests that environmental concerns beyond material infrastructure are largely absent from AI policy, and the risk and harm typologies used therein. Because policy documents establish a discursive infrastructure that brings the inscribed objects and relationships into being, this paper interrogates how environmental concerns are (re)presented (or missing) in international AI policy documents.

Drawing on eleven reports or policy documents by international organisations, we apply "critical whataboutism" to reconstruct understandings of AI, associated risks, problematisations, outlined solutions, imaginaries, and the way the documents distribute power and agency.

We find that beyond tokenistic referral to carbon emissions, only few policy documents consider wider environmental impacts; however, these documents still often hedge their own agency in order to conform to discourses of international competitiveness and to reap supposed efficiency gains associated with AI; in these cases, generalised and strategic ignorance as well as priorisation of societal goals appear to contribute to lacking consideration for the environment. Overall, maintaining (or reaching) competitiveness appears to be the main aim of the screened policy reports even for international organisations.

As in society at large, the position of the environmental crisis is unfortunately lost in the context of geopolitics, which all screened documents navigate. Our analysis provides opportunities to rethink "AI" and its relationship to environmental crises. It urges policy makers to write environmental crises "into being" so that they become part of the discursive infrastructure, also on AI.