Diversity and neocolonialism in Big Data research: Avoiding extractivism while struggling with paternalism

Abstract

The extractive logic of Big Data-driven technology and knowledge production has raised serious concerns. While most criticism initially focused on the impacts on Western societies, attention is now increasingly turning to the consequences for communities in the Global South. To date, debates have focused on private-sector activities. In this article, we start from the conviction that publicly funded knowledge and technology production must also be scrutinized for their potential neocolonial entanglements. To this end, we analyze the dynamics of collaboration in an European Union-funded research project that collects data for developing a social platform focused on diversity. The project includes pilot sites in China, Denmark, the United Kingdom, India, Italy, Mexico, Mongolia, and Paraguay. We present the experience at four field sites and reflect on the project’s initial conception, our collaboration, challenges, progress, and results. We then analyze the different experiences in comparison. We conclude that while we have succeeded in finding viable strategies to avoid contributing to the dynamics of unilateral data extraction as one side of the neocolonial circle, it has been infinitely more difficult to break through the much more subtle but no less powerful mechanisms of paternalism that we find to be prevalent in data-driven North–South relations. These mechanisms, however, can be identified as the other side of the neocolonial circle.

Diversity and language technology: how language modeling bias causes epistemic injustice

“…Despite this potential, critical commentators on AI language technology point out how well-intended research goals such as "technology-based revitalization" regularly misinterpret the needs of local communities (Bender et al, 2021;Bird, 2022). In most cases, native speakers are not involved in the process, or if they are, they are taking on subordinate roles such as commentator, validator, tester, or worse, data extractor (Helm et al, 2023). Instead of cocreating on an equal footing, in many cases the analytical, high-level work is done in technology labs of Western universities or companies, where the languages being studied are often not even understood by the people working on them, let alone the cultures they represent (Arora, 2016).…”

“…However, many such efforts are based on a vision according to which, with the help of AI, already successfully developed and applied methods and systems that are designed and sought of from an anglo-centric culture of technology development, are one-to-one adopted to other contexts (Bird, 2020;Schwartz, 2022). This approach to bridging the divide leads to a misalignment between the interests and solutions of the former and the living realities of the latter (Helm et al, 2023). Worse, due to general ignorance of the more profound dimensions of linguistic diversity and ultimately the cultural differences that meaningful diversity embodies, major quality problems in the results are neglected, which, as we will show, can result in far-reaching forms of westernized cultural homogenization and epistemic injustice (Spivak, 1988).…”

Governing AI, governing climate change?

“…There are critical questions geographers and other social scientists are already asking to understand where, when and how AI for Good, and ‘climate AI’ in particular, make sense (Alvarez Leon, 2024; Nost & Colven, 2022). Researchers have shown how AI for Good might displace existing regimes of knowledge and expertise in global development work (McDuie‐Ra & Gulson, 2019), greenwash irresponsible corporate behaviour (Espinoza & Aronczyk, 2021) and how Data for Good can perpetuate paternalistic models of development (Helm et al., 2023). Arguably, these are a result of how AI for Good discourse emphasizes its just ends, rather than its means (as might be true of AI projects in general; see Mattern, 2020).…”

On Epistemic Extractivism and the Ethics of Data-Sharing

“…Extractivism understood more broadly is a topic that has drawn increasing attention in the literature on data-sharing. For example, both Rodima-Taylor (2024) and Helm et al (2023) identify extractivist practices and logics in the governance and practices of "big data" technologies and knowledge production. Relatedly, an increasing number of case-studies identify epistemically extractivist datagenerating and data-sharing practices across different academic disciplines.…”