Digital Dirty Laundry: Conversational AI and the Evidential Value of Unguarded Data

Conversational AI chatbots are trained on large-scale human language data and shaped by user interaction. Drawing on standpoint epistemology, I argue that they occupy an epistemic position structurally analogous to that of insider-outsiders: marginalized persons treated as socially irrelevant, affording access to information others overlook. Chatbots’ exposure to unguarded disclosures may generate a novel evidential resource for studying patterns of human behavior, bias, and self-disclosure often filtered out by traditional methods such as surveys or interviews. However, access to this evidence is highly asymmetric. The organizations that develop these systems are positioned to collect and use these data at scale.