Hugo Cossette-Lefebvre

The proliferation of increasingly undetectable large language models (LLMs) poses a novel challenge for social media users. As it becomes reasonable to suspect that one’s online interlocutor may not be human, participants are pushed toward what I call agency-directed distrust: a distinctive form of distrust directed not at another’s character or competence, but at the appearance of agency itself. Drawing on Strawson’s distinction between the participant and objective stances, I argue that agency-directed distrust pressures us to suspend the participant stance—the stance from which reactive attitudes such as blame, resentment, and gratitude are intelligible. To withdraw it is to treat the apparent interlocutor as an object of management rather than engagement. Uncertainty about which stance to adopt toward ambiguous interlocutors generates a moral dilemma for ordinary online participants. On one horn, suspending the participant stance risks disrespecting genuine persons by failing to recognize their standing as accountable agents. On the other, extending the participant stance to covert LLMs risks entering into asymmetric relationships that are at odds with the demands of self-respect and becoming complicit in eroding the norms of interpersonal accountability that sustain meaningful online discourse.