Lessons Learned from Modeling the Interaction with Conversational Agents

Abstract

Intelligent conversational agents have become widespread. Inspired by conversations in natural language, they present different degrees of intelligence and autonomy, bringing challenges for Human-Computer Interaction (HCI). One such challenge concerns design languages for modeling user-agent interaction. We focus here on MoLIC, a design-phase dialogue model based on Semiotic Engineering theory devised to represent user-system interactions as conversations. We performed two case studies with MoLIC interaction diagrams representing two conversational agents – the ANA chatbot and Samsung Bixby. We examined how the interactive aspects of these agents could be expressed in MoLIC. Although it was possible to express the general interaction, our results showed limitations related to the language expressiveness or its inadequacy to represent these systems. We identified limitations in the applicability of MoLIC in modeling and pondered on how to extend or adapt it; directing the HCI community to issues and initiatives that can help design and model these technologies.