Algorithmic reparation

Abstract

Machine learning algorithms pervade contemporary society. They are integral to social institutions, inform processes of governance, and animate the mundane technologies of daily life. Consistently, the outcomes of machine learning reflect, reproduce, and amplify structural inequalities. The field of fair machine learning has emerged in response, developing mathematical techniques that increase fairness based on anti-classification, classification parity, and calibration standards. In practice, these computational correctives invariably fall short, operating from an algorithmic idealism that does not, and cannot, address systemic, Intersectional stratifications. Taking present fair machine learning methods as our point of departure, we suggest instead the notion and practice of algorithmic reparation. Rooted in theories of Intersectionality, reparative algorithms name, unmask, and undo allocative and representational harms as they materialize (American English sp) in sociotechnical form. We propose algorithmic reparation as a foundation for building, evaluating, adjusting, and when necessary, omitting and eradicating machine learning systems.

Ethics for the majority world: AI and the question of violence at scale

“…Drawing on Barad's (2007) notion of response-ability, an decolonial intersectional feminist agenda for artificial intelligence emphasises shared response-ability (Cortés et al, 2020). Algorithmic repair (Davis et al, 2021) is inextricably linked to ethics of co-responsibility that takes into account AI's broader and deeper effects.…”

Artificial intelligence, algorithms, and social inequality: Sociological contributions to contemporary debates

“…Given that work on fairness, ethics, and bias in AI is now part of the ‘normative construction of the world’ (Green & Hu, 2018, p. 5), sociologists can contribute to this world‐building (Joyce et al., 2021) and help articulate possible directions for change that go beyond nebulous notions of an unbiased society or undesirable tendencies. Where vague gestures towards a ‘social good’ have predominated in AI ethics, there is an opportunity to specify desirable futures based on substantive equality and anti‐oppression (Green, 2019; also; Davis et al., 2021). Unfortunately, while sociologists are on comfortable ground when critiquing representational inequalities and algorithmic harms, we are less comfortable in making normative proposals for what to do with algorithmic systems, beyond rejecting their use in problematic cases.…”

“…Sociology's strengths include our theoretical and empirical analyses of inequality – an understanding of systems of stratification and forms of discrimination. Systemic problems such as racism and poverty are structurally reproduced and require systemic approaches; sociologists can help to articulate what these approaches might look like (e.g., Davis et al.’s [2021] proposal for ‘algorithmic reparation’), or where to focus our energies to achieve social change. Therefore, the greatest contributions that sociologists can make are at earlier stages in the development of policies and sociotechnical systems, where goals and techniques are less clearly defined.…”

“…Over the past several years, the idea that society is being radically transformed through artificial intelligence (AI), machine learning, and automation has been widely discussed in popular venues. This transformation has been addressed through a growing body of sociological scholarship (see Joyce et al., 2021; also; Airoldi, 2022; Burrell & Fourcade, 2021; Davis et al., 2021; Elliott, 2022; Issar & Aneesh, 2022; Jaton, 2021), although some of the most prominent sociological analyses of these technologies have come from academics who are not (formally) sociologists (Crawford, 2021; Eubanks, 2017; Zuboff, 2018). Societal questions are now recognized as crucial ones for the developers of new algorithmic technologies, but ‘AI scientists continue to demonstrate a limited understanding of the social’ (Joyce et al., 2021, p. 5).…”

“…These inequalities are sometimes categorized as ‘societal bias’ (Friedman & Nissenbaum, 1996) or ‘historical bias’ (Suresh & Guttag, 2020) by AI practitioners, which has the effect of black‐boxing any specifics about where the bias comes from or is structured. But sociologists can name the intersecting structures (Davis et al., 2021; Hoffmann, 2019) of inequality, oppression, or exploitation that produced the biases AI and data science practitioners are concerned about: colonialism, heteropatriarchy, ableism, white supremacy, and various articulations of capitalism or political economy. While AI developers have been repeatedly criticized for being unwilling to look beyond their own field to ‘see the social’ (Irani & Chowdhury, 2019), this has slowly been changing, and social scientists have been identified as a source of relevant expertise by AI researchers (CIFAR, 2020; Kusner & Loftus, 2020), tech companies (Maris, 2022), academic bodies and governments (G7 Science Academies, 2019).…”