Machine learning and the politics of synthetic data

Abstract

Machine-learning algorithms have become deeply embedded in contemporary society. As such, ample attention has been paid to the contents, biases, and underlying assumptions of the training datasets that many algorithmic models are trained on. Yet, what happens when algorithms are trained on data that are not real, but instead data that are ‘synthetic’, not referring to real persons, objects, or events? Increasingly, synthetic data are being incorporated into the training of machine-learning algorithms for use in various societal domains. There is currently little understanding, however, of the role played by and the ethicopolitical implications of synthetic training data for machine-learning algorithms. In this article, I explore the politics of synthetic data through two central aspects: first, synthetic data promise to emerge as a rich source of exposure to variability for the algorithm. Second, the paper explores how synthetic data promise to place algorithms beyond the realm of risk. I propose that an analysis of these two areas will help us better understand the ways in which machine-learning algorithms are envisioned in the light of synthetic data, but also how synthetic training data actively reconfigure the conditions of possibility for machine learning in contemporary society.

Actionability of Synthetic Data in a Heterogeneous and Rare Healthcare Demographic; Adolescents and Young Adults (AYAs) with Cancer

“…In part, the necessity of consistent scrutiny reiterates the need for established methods to assess SD’s actionability 2,23,24 . Regardless, SD can technically be made actionable, and exuberant innovation can not only simplify this process and appraisal thereof but can incorporate here unmentioned domains such as SD fairness 38 , yet for true SD actionability in healthcare established ethical and legal consensus is paramount as these should dictate what actionability actually implies 5,8 .…”

“…In this case, the goal of SD is to catalyse knowledge generation in healthcare, but without exposing real human subjects’ data. While use of SD within healthcare is still in development, ethicists, policy makers, and researchers are trying to establish how SD can be used appropriately 5,8 . Meanwhile, various techniques 9–12 are now openly available which researchers can use to train generative models and thus create SD for various purposes 13–22 .…”

Close to the metal: Towards a material political economy of the epistemology of computation

“…However, as machine learning algorithms gained traction, they required and, in turn, propelled a turn towards datafication, data extraction and accumulation and, increasingly data production ostensibly ex nihilo through generative models for synthetic data (Jacobsen, 2023;Steinhoff, 2022). Propelled by the uptake of social media and portable devices and by broader turns across industries towards data-driven business models that made data monetization an essential component of the revenue and profitability of small and large businesses alike, datafication, 'surveillance capitalism' (Zuboff, 2019) and 'surveillance advertising' (Crain, 2021) turned towards the production of data as a commodity and as capital (Sadowski, 2019), that is, both as an asset that can be bought and sold, and as a mix of raw material and productive factor for the 'platform political economy' (Langley & Leyshon, 2021).…”

“…Indeed, as Amoore and colleagues have shown, while connectionism was present since the 1950s, with Rosenblatt’s model of the perceptron (Rosenblatt, 1958), and while it started to gain momentum in the 1980s, it was only in the early 2000s and some of the ‘victories’ such as the ImageNet competition, thanks to GPUs, that this epistemological approach became mainstream (Amoore et al, 2023, p. 1). Convolutional neural networks for image recognition had been performing well, but they were rejected by the scientific community because, according to Hinton, the task of AI had been to define deductively the task that the machine was meant to solve, rather than programming a machine that inductively performed well at a task (Hinton, 2019, p. 24).…”

“…This article is based on close readings of computer science papers on hardware architectures, artificial intelligence, and cryptoassets, understood as invaluable sources of epistemological and ethico-political meaning making (Amoore et al, 2023). ‘Close to the metal’ is a term in computing science to denote a property of programming languages to directly access and influence the behaviour of hardware.…”