Suspicious Minds: the Problem of Trust and Conversational Agents

In recent years, the field of natural language processing has seen substantial developments, resulting in powerful voice-based interactive services. The quality of the voice and interactivity are sometimes so good that the artificial can no longer be differentiated from real persons. Thus, discerning whether an interactional partner is a human or an artificial agent is no longer merely a theoretical question but a practical problem society faces. Consequently, the ‘Turing test’ has moved from the laboratory into the wild. The passage from the theoretical to the practical domain also accentuates understanding as a topic of continued inquiry. When interactions are successful but the artificial agent has not been identified as such, can it also be said that the interlocutors have understood each other? In what ways does understanding figure in real-world human–computer interactions? Based on empirical observations, this study shows how we need two parallel conceptions of understanding to address these questions. By departing from ethnomethodology and conversation analysis, we illustrate how parties in a conversation regularly deploy two forms of analysis (categorial and sequential) to understand their interactional partners. The interplay between these forms of analysis shapes the developing sense of interactional exchanges and is crucial for established relations. Furthermore, outside of experimental settings, any problems in identifying and categorizing an interactional partner raise concerns regarding trust and suspicion. When suspicion is roused, shared understanding is disrupted. Therefore, this study concludes that the proliferation of conversational systems, fueled by artificial intelligence, may have unintended consequences, including impacts on human–human interactions.