Increasing Human-Likeness and Acceptance of Conversational Autonomy through Experience

Abstract

An increasing number of online platforms offer access to conversational artificial intelligence (AI), such as ChatGPT, allowing individuals to engage and talk with AI. Conversing using natural language may foster perceptions of AI as having more human-like qualities. Within the context of human-autonomy teaming (HAT), this may contribute to AI being viewed more as a human-like teammate than an instrumental tool. The present research investigated factors that influenced the perceptions of conversational AI as a human-like teammate. Using the autonomous agent teammate-likeness (AAT) framework, participants were presented with simulated conversations between an AI and user within a smartphone app across multiple contexts. Participants were more likely to rate the AI as a human-like teammate with increased exposure to conversations, when they had a less negative attitude towards robots, and with greater general propensity to trust. In addition, perceptions of human-likeness were associated with (stronger) perceptions of the AI as trustworthy, useful, and acceptable. This suggests that fostering human-likeness via experience with natural language conversations can contribute to AI being perceived more like a human teammate and improve (HAT) outcomes.