Data simulation as centers of data power

Title: Data simulation as centers of data power

Autor: Victoria Kontrus, Roger von Laufenberg, Vera Gallistl

Published: 06.09.2024 at the Data Power Conference 2024

Full Text available: n.a.

Citation:

Kontrus, V., von Laufenberg, R. & Gallistl V. (2024, Sept 4-6). Data simulation as centers of data power [Presentation]. Data Power Conference 2024, Graz.

Abstract:

Training AI models requires copious amounts of data. As AI is often developed within profit-oriented structures, new ways of meeting this growing need for data ever-more efficiently are being explored constantly. The result is a shift from data gathering to data simulation, and from natural data to

synthetic data. Synthetic data is often cheaper, can be obtained faster and – to some extent – in greater variety than natural data. Simulation is especially useful for data on rare events or data from hardly accessible contexts. This economically driven shift in the mode and context of data production entails a shift in data power from the original data subjects towards those who develop AI. This entanglement of economic structures of AI, data power and data simulation shall be illustrated using the case of an AI-based fall detection and prevention system used in long-term care facilities. Instead of months of on-site data gathering of residents’ behavior in LTC facilities, developers simulate data of falls using a complex infrastructure of tools, including motion capture suits, 3D modelling programs as well as their own – often young, male, white and able – bodies. Especially in profit-oriented structures of production, these data simulation labs become new centers of data power where ground-truths of AI are fabricated and shaped. Reduced dependency on data subjects and suppliers of natural data further advances the power of those who develop AI, as the process towards market-readiness of their products is streamlined. Power is then further leveraged through deployed fall monitors which continually harvest new natural data, exploiting free labor of residents as data subjects without their knowledge, while these already pay for the service offered to them. As this case shows, data synthesis proves a valuable instance for the analysis of data power, illuminating the complex entanglement with economic power.