Philosophy & Technology Humanities and social sciences
On a Misleading Ethical Fluency in Language Models: Reply to Danagher
Abstract
Abstract In this brief comment, I discuss Luke Danagher’s recent warning that language models’ ethical fluency may inspire greater confidence than their demonstrated ethical reliability justifies. I consider three versions of this charge: that people would be making an ontological, an epistemic, or an ethical mistake. In the cases of the ontological and epistemic mistakes, I question whether such mistakes cause axiological bads that are serious enough to impugn the use of language models in ethically relevant decision-making. In the case of the ethical mistake, I argue that using a human in place of the language model to make the ethically relevant decision would not yet be clearly better from the perspective of expected ethical value.