Would I lie to you? How interaction with chatbots induces dishonesty

Abstract

Is dishonesty more prevalent in interactions with chatbots compared to humans? Amidst the rise of artificial intelligence, this question holds significant economic implications. We conduct a novel experiment where participants report the outcome of a private, payout-relevant random draw to either a chatbot or a human counterpart, with varying degrees of signaled agency. We find that signaling agency increases honesty when interacting with humans but not with chatbots. Moreover, participants are consistently more honest with humans in the presence of agency cues. Our results suggest that social image concerns and perceived honesty norms play a more prominent role in human interactions. Surprisingly, standard online forms generate the same levels of honesty as human-to-human chat interactions. These findings offer valuable insights for designing effective communication and trust-building mechanisms in digital economies where human-chatbot interactions are increasingly prevalent.