by Guillaume Gauthier (U. Basel)
EMCA at the intersection of technologies and disciplines
Hosted by the University of Basel and sponsored by the Hermann Paul School of Linguistics, the colloquium Conversational AI: Advances and Challenges (April 9-10, 2025), organized by Lorenza Mondada (U. Basel) and led by Saul Albert (U. Loughborough), brought together graduate students, PhD candidates, postdocs and professors for a collective exchange on how advances in conversational AI can be addressed, questioned, and complemented by EMCA. The event united an interdisciplinary group—from computer and communication sciences to linguistics and political science—to tackle two core issues: how the EMCA program fundamentally addresses the challenges of conversational AI, and how applied CA can not only inform the development of agents but also analyze how users locally engage with conversational AI, making sense of the practical problems they encounter.
An EMCA take on conversational AI
S. Albert began by outlining what an EMCA perspective on conversational AI can look like, insisting that the usage of these systems needs to be tackled as a members’ problem. This involves moving towards an analysis of how participants themselves orient to and make sense of such technologies in and through interaction. Discussions then turned to what conversational AI cannot do (yet), pinpointing fundamental issues tied to matters of interactional relevance, sequential organization and recipiency. The data-driven debates highlighted significant limitations in the agents’ responses to certain conversational phenomena, such as their limited ability to manage delicate courses of action like trouble-talk or to attend to the relevance of en passant remarks. Drawing on these observations, attendees remarked on systematic problems in the way conversational agents processed everyday sequential structures—including pre-sequences, retro-sequences, or complex instances of overlapping talk.
How do bots care?
With these shortcomings in view, the workshop turned to a practical exercise of situated integration: how might applied CA approach the implementation of conversational AI (with all its limitations) into the high-stakes environment of health/care, where talk is central to both organizing and legitimizing care? This hands-on engagement with the design of conversational agents led attendees to remark on how ‘end-user’ orientations to these agents were deeply consequential for the situated organization of participation. Discussion revealed how such orientations notably reflect in the positioning, emergence and temporal organization of actions, the multimodal composition of conduct, and the attribution of agency and responsibility within health/care activities. In turn, these insights prompted critical reflections on how AI-powered conversational technologies can both afford and constrain certain forms of sociality.
Observably consequential pieces of code
In all, the colloquium fostered a rich examination of how EMCA can advance the study of conversational AI, while also critically reflecting on how these technologies can transform, complement and even undermine social interaction. Key takeaways centered on the variety of orientations end-users can adopt toward conversational agents—ranging from anthropomorphization to recipient-designed talk for a machine—with each orientation responding to activity-specific, practical concerns. Analyses of machine-oriented recipient design, where users adapt their talk to trigger an adequate output from the system, revealed members’ complex, situated inferences about the capacities and limitations of conversational AI. Where anthropomorphization occurred (most notably among vulnerable populations such as children and the elderly), it became apparent to what extent making relevant one’s need for help or company was inextricably bound to moral expectations of engagement and mutual understanding; expectations such systems are programmed to simulate, not to meet.