Abstract
In this symposium, we use the history of philosophy to illuminate specific, grounded aspects of contemporary practice in machine learning, raising new problems and proposing new frameworks for the philosophy of machine learning. Our symposium draws on 200 years of empiricism and its critics, ranging from eighteenth century’s Hume, Smith, de Grouchy, Locke, and Leibniz to the twentieth century’s Carnap, Putnam, and Goodman. Our methodology is shared with contemporaneous work in the philosophy of machine learning such as Buckner (2018, 2020), Chirimuuta (2020), Haas (2022), Nefdt (2020), Sterkenburg & Grunwald (2020). Our five topics bear a family resemblance to one another. Three (Buckner, Creel, and Clatterbuck) draw on historical themes of empiricism and rationalism. Three (Buckner, Creel, and Romeijn) address the conditions necessary for machine systems to succeed at learning from data produced by humans. Two (Buckner and Clarke) provide concrete proposals for creating systems or agents that can model, understand, and intervene in the social world. Two (Creel and Romeijn) address the preconditions and limitations of automated science and inference. And two (Clatterbuck and Clarke) characterize existing debates in machine learning within a broader historical frame, allowing the central questions to be productively re-oriented.
