Understanding Participation in Deliberative Polling: Insights from the Meta Community Forum on AI Chatbots

This paper presents findings from the second Meta Community Forum, a collaborative effort between Meta, the Stanford Deliberative Democracy Lab, and the Behavioral Insights Team in October 2023. The forum, modelled after Deliberative Polling, focused on defining principles to guide the engagement of generative AI with users, particularly in the context of increasingly powerful AI chatbots. Questions surrounding the human-like attributes of AI chatbots, user preferences in interactions, and delineating limits on emulating human traits were central themes. The discussion, attended by 1545 participants from Brazil, Germany, Spain, and the United States, spanned a weekend and addressed critical issues such as the level of human resemblance AI chatbots should exhibit, user preferences in interactions, and determining boundaries for emulating human traits. Participants also explored the balance between unpredictability and risk in AI chatbot responses, raising questions about prioritizing originality versus predictability to avoid offense. Additionally, a control group comprising 1108 individuals participated in the study, abstaining from deliberative discussions and completing only two surveys. The primary purpose of the control group was to establish a baseline and demonstrate that any observed changes post-deliberation were a direct result of the deliberative event. This paper contributes valuable insights into the considerations surrounding AI chatbot behavior and user expectations, offering a nuanced understanding of the principles guiding generative AI engagement with users. The findings hold implications for the design and implementation of AI systems, emphasizing the importance of user preferences and ethical considerations in shaping the future landscape of human-AI interactions.