The Relational Risks of Generative Artificial Intelligence

Abstract

The increasingly widespread use of generative artificial intelligence has begun to reshape the landscape of interpersonal communication. It is no longer unusual, for example, for co‐workers to reply to each other's emails, or even romantic partners and family members to respond to each other with AI‐generated messages. Yet the suspicion that one has received AI‐generated, rather than human‐authored content is often accompanied by strong negative reactions ranging from disgust to disappointment. Focusing on the case of AI‐generated apologies, my paper systematically examines the moral reasons we typically appeal to for resenting someone who uses AI to convey an apology. Interestingly, these moral reasons do not decisively prohibit the use of generative AI in interpersonal communication. Rather, the emergent social practice of apologizing through AI poses a distinctive risk to interpersonal relations themselves. Chiefly, this practice threatens to degrade our ability to engage in careful “translation” of each other's linguistic performances, or conversely, enact “performance” in a way that can be taken up by the other (Brandom 1979, 190). This is because AI‐generated messages aim to “optimize” communication and minimize conflict and friction that, while not always pleasant, is vital to our understanding of, and responsiveness in relation to each other as authentic subjects. Consequently, closer attention must be paid to these relational risks of generative AI.