I’m Only Human: The Effects of Trust Dampening by Anthropomorphic Agents

"It's like a rubber duck that talks back": Understanding Generative AI-Assisted Data Analysis Workflows through a Participatory Prompting Study

“…Some participants noted this as a benefit as it improved the collegiality of the interaction, while others felt that it introduced an unwarranted expectation of politeness, verbosity, and conversationality on the part of the user (Section 4.3). This is also reflected in other studies of anthropomorphism in AI, which have found that introduction of human-like features can help users be more forgiving of a system that makes errors [24], and improves its perceived likeability, but can be counterproductive for a system with high performance and focus on task completion [11]. It is unclear from our findings whether there is a single correct approach for data sensemaking, which includes a blend of activities, some of which the system may be able to perform with high accuracy, and some not.…”