It's incomprehensible: On machine learning and decoloniality

As technologies begin to permeate society, people, and society itself, reorganize around the technology. It is therefore imperative that academic communities, civil society, and regulatory bodies address the impacts of technology on society and human life. With machine learning, scholars, activists, and regulators have dedicated significant efforts toward quantifying and mitigating the expressions of social biases in machine learning models. Recently, researchers have argued that machine learning reproduces colonial logics and proposed decolonization as an avenue for future machine learning efforts. Here we examine the aims of machine learning and decolonization, and argue that their goals, i.e., to abstract away and attend to detail and histories, respectively, are inherently in tension. Their origins, statistics, and phrenology on one hand, and the liberation from marginalization and abstraction on the other, point in different directions. However, this tension can be resolved by situating machine learning within communities that can fill in detail that the technologies abstract away.