Making up with parrots: A critical praise of machine learning agency

This paper aims to contribute to the reconciliation between radical critiques and endorsements of machine learning's capacity to act, intervene, and produce social effects (i.e., its agency). Building upon recent ethnographic studies of the mundane shaping of algorithmic systems, the paper begins by underlining three positive attributes of machine learning agency: “experience” (the ability to embody insights derived from referential datasets), “consistency” (the ability to remain aligned to encoded principles), and “intelligence” (the ability to link new elements with prior knowledge). To balance this reverential perspective, the paper then highlights the limitations of these attributes through the lens of philosophy. Chuang Tzu's concept of emptiness reveals the struggle of machine learning agency with openness and confusion; Bruno Latour's reflections on psycho-bearing entities highlight machine learning agency's imperviousness to external disruptions; and Henri Bergson's discussion of intelligence exposes the inability of machine learning agency to engage with lifeful phenomena. This exploration results in a form of “critical praise” that acknowledges both the constraints and capacities of machine learning agency, fostering diplomatic rapprochement between its strongest critics and advocates.