Are You Even a Person? LLMs and Agentic Distrust
Work-in-Progress anim茅 par Alexis Morin-Martel, doctorant 脿 l鈥橴niversit茅 McGill.
Leacock 927, de 13h30 脿 15h30, le 18 d茅cembre.
Ouvert 脿 toutes et 脿 tous, inscription obligatoire.
Pour vous inscrire, veuillez envoyer un courriel 脿 alessandra.destison [at] mail.mcgill.ca
R茅sum茅 de l鈥檃rticle :
Meaningful online conversation is increasingly difficult, not only because of bad-faith arguments or disinformation, but because we can no longer be sure our interlocutors are even agents. What looks like ordinary human communication can now just as easily be generated by a large language model (LLM), and empirical studies show that humans are poor at distinguishing LLM-generated from human-generated content. I call this increasingly reasonable doubt about whether others are agents聽agentic distrust聽and argue that it constitutes the primary moral wrong of LLM proliferation. I first consider and reject three familiar explanations of what is troubling about LLMs: harmful content, deception, and their inability to satisfy familiar norms of assertion. None captures the distinctive interpersonal harm introduced by agentic distrust. I then argue that agentic distrust generates a potentially intractable moral dilemma. If we treat suspicious interlocutors as bots, we risk improperly withholding the recognition respect owed to persons. If we treat them as human, we risk misallocating our moral attention and becoming complicit in destabilizing the norms of interpersonal accountability that make discourse meaningful.