Explainable AI as a Rhetorical Technology: Promoting Civic Virtue in the Age of AI

<jats:title>Abstract</jats:title> <jats:p>Is explainable AI feasible and desirable, and are explanations of AI decisions always good? Many answer this question in the affirmative, but there is a growing discourse that is suspicious of the promises of explainable AI used to explain decisions towards users, considering it unfeasible, undesirable, and even potentially misleading. Could explainable AI be just a rhetorical foil? This paper proposes that, yes, explainable AI is a rhetorical technology but no, this does not necessarily make it undesirable. It starts by revisiting this debate according to the initial charge against rhetorics by Plato and Aristotle’s response, considering rhetoric pharmacologically, as not only a poison but also a cure for political life. It argues that, just as rhetoric was necessary to take care of the temporalities of public life, in court, the public event, and the assembly, so may explainable AI contribute to a rhetorical context. Yet, whether it does so is conditioned on the extent to which it cultivates the civic virtues relative to a respective context. The paper considers the examples of predictive policing, credit scoring, and prediction markets to argue about ideal states – civic virtues that may be cultivated in each appropriate context – and deviations that point at the risks of explainable AI to lead to domination, conformism, and political recalcitrance.</jats:p>