Secondhand AI

A person could go their whole life without using an AI chatbot, and yet still feel its effects: reading long texts at work that sound like “AI slop,” suspecting that a heartfelt message was written by AI, or watching news stories in their feed portraying people who never existed. These are more than mere annoyances; a growing body of research suggests that the effects of AI use could spillover from the primary user to downstream recipients. How can this happen? To explain such effects, we introduce secondhand AI: situations in which one person’s use of AI affects others in their social network. In this talk, we’ll present empirical evidence about how AI can make some feel they know more than what they actually do, and how this feeling can be transmitted through ordinary social interactions. We call for urgent efforts to measure these effects in domains where AI systems are reshaping the fabric of our social interactions.