Beyond artificial intelligence controversies: What are algorithms doing in the scientific literature?

Abstract

Mounting critique of the way AI is framed in mainstream media calls for less sensationalist coverage, be it jubilant or apocalyptic, and more attention to the concrete situations in which AI becomes controversial in different ways. This is supposedly achieved by making coverage more expert-informed. We therefore explore how experts contribute to the issuefication of AI through the scientific literature. We provide a semantic, visual network analysis of a corpus of 1M scientific abstracts about machine learning algorithms and artificial intelligence. Through a systematic quali-quantitative exploration of 235 co-word clusters and a subsequent structured search for 18 issue-specific queries, for which we devise a novel method with a custom-built datascape, we explore how algorithms have agency. We find that scientific discourse is highly situated and rarely about AI in general. It overwhelmingly charges algorithms with the capacity to solve problems and these problems are rarely about algorithms in their origin. Conversely, it rarely charges algorithms with the capacity to cause problems and when it does, other algorithms are typically charged with the capacity to solve them. Based on these findings, we argue that while a more expert-informed coverage of AI is likely to be less sensationalist and show greater attention to the specific situations where algorithms make a difference, it is unlikely to stage AI as particularly controversial. Consequently, we suggest conceptualising AI as a political situation rather than something inherently controversial.