séminaire général

Nous aurons le plaisir d'accueillir : Carina Prunkl (University of Utrecht, philosophy)

Are algorithms more objective than humans? Objectivity, trust, and the epistemic authority of AI

Abstract : Calling a decision ‘objective’ typically implies praise. Objectivity, in this case, not only describes a property of the decision-making process but also serves as a marker for epistemic authority or fairness. Machine Learning (ML) systems are sometimes hailed for being more objective than humans, other times criticised for displaying the same (or worse) kind of biases that mar human decision-making. In this talk, I argue that proponents and opponents of algorithmic objectivity (here the relative claim that algorithms are more objective than human decision-makers) focus on and conflate different aspects of algorithmic judgment: execution (the decision-making process itself) and model building (the design choices shaping the algorithm’s decisions). While proponents of algorithmic objectivity emphasize execution-related qualities such as consistency or data-driven reasoning, critics focus on biases embedded in model-building and data collection. In this talk I clarify these distinctions and examine how objectivity serves as an indicator of epistemic authority. Drawing on Fricker’s (2011) framework of epistemic trust, I explore alternative markers of epistemic reliability—such as robustness and intersubjective agreement—that may offer more productive ways to evaluate algorithmic decision-making.