Queering the Center of Machine Learning: Lifting the Veil of Objectivity and Conceptualizing Research Beyond the Deep Learning Paradigm
Type: article
Abstract: The well-documented issues around social bias and the negative environmental impact of ubiquitous machine learning-based artificial intelligence (AI) systems now more than ever motivate critical engagement with the potential dangers of this technology, as well as the conceptualization of research with the goal of reducing harm to marginalized groups and the environment. Previous scholarship has demonstrated how the design of ML models to be used in AI technology is not a neutral endeavor and that ML practitioners must make value judgments at various stages in this process. This thesis asks how it is that ML research can retain a veil of objectivity and how we can conceptualize ML research to facilitate the reduction of harm to marginalized groups and the environment by taking this dynamic into account. From a hybrid queer theory and feminist philosophy of science perspective, I will defend the thesis that due to the current commitment to deep learning (DL) models, a center/margin dichotomy has developed on the terrain of values in ML research, making paradigmatic values that guide decisions in the ML pipeline seem particularly neutral, objective, and apolitical. By problematizing this center/margin dichotomy both through critical analysis and conceptual engineering, we can start imagining ML research beyond the potentially harmful DL paradigm.
Author: Somcutean, Cristina
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Reduced inequalities
Open Access status: green
OpenAlex ID: W7115980517
