This article identifies a novel challenge to standard understandings of responsibility practices, animated by experimental studies of biases and heuristics. It goes on to argue that this challenge illustrates a general methodological challenge for theorizing about responsibility. That is, it is difficult for a theory to give us both guidance in real world contexts and an account of the metaphysical and normative foundations of responsibility without treating wide swaths of ordinary practice as defective. The general upshot is that theories must (...) either hew more closely to actual practice than they appear to, or they must provide some normative foundation for responsibility that does not go through actual practice. (shrink)

Addressing biases in natural-language processing (NLP) systems presents an underappreciated ethical dilemma, which we think underlies recent debates about bias in NLP models. In brief, even if we could eliminate bias from language models or their outputs, we would thereby often withhold descriptively or ethically useful information, despite avoiding perpetuating or amplifying bias. Yet if we do not debias, we can perpetuate or amplify bias, even if we retain relevant descriptively or ethically useful information. Understanding this dilemma provides for a (...) useful way of rethinking the ethics of algorithmic bias in NLP. (shrink)

Determinism is frequently understood as implying the possibility of perfect prediction. This possibility then functions as an assumption in the Manipulation Argument for the incompatibility of free will and determinism. Yet this assumption is mistaken. As a result, arguments that rely on it fail to show that determinism would rule out human free will. We explain why determinism does not imply the possibility of perfect prediction in any world with laws of nature like ours, since it would be impossible for (...) an agent to predict with certainty any future event that is causally influenced by events outside her own backward light cone yet inside the backward light cone of the future event. This is the _light-cone limit_ and it undermines the Manipulation Argument or limits what this argument can tell us about the relevance of determinism to free will. We also respond to objections that the light-cone limit is irrelevant to the Manipulation Argument. (shrink)