What Makes a Good Conversation? Challenges in Designing Truly Conversational Agents (1901.06525v1)

Abstract: Conversational agents promise conversational interaction but fail to deliver. Efforts often emulate functional rules from human speech, without considering key characteristics that conversation must encapsulate. Given its potential in supporting long-term human-agent relationships, it is paramount that HCI focuses efforts on delivering this promise. We aim to understand what people value in conversation and how this should manifest in agents. Findings from a series of semi-structured interviews show people make a clear dichotomy between social and functional roles of conversation, emphasising the long-term dynamics of bond and trust along with the importance of context and relationship stage in the types of conversations they have. People fundamentally questioned the need for bond and common ground in agent communication, shifting to more utilitarian definitions of conversational qualities. Drawing on these findings we discuss key challenges for conversational agent design, most notably the need to redefine the design parameters for conversational agent interaction.

• The paper employs semi-structured interviews and inductive thematic analysis to reveal the gap between human conversational traits and current agent capabilities.
• The study finds that users view agents primarily as transactional tools, highlighting a need to tailor design strategies to distinct social and functional demands.
• The research advocates for innovative approaches, including non-verbal communication features, to enhance the quality and naturalness of human-agent interactions.

An Analytical Overview of "What Makes a Good Conversation? Challenges in Designing Truly Conversational Agents"

The paper "What Makes a Good Conversation? Challenges in Designing Truly Conversational Agents" explores the critical disjunction between human conversational standards and the current capabilities of conversational agents like Siri, Amazon Alexa, and Google Assistant. Through a meticulous examination grounded in user studies, the authors aim to dissect the core attributes valued in human conversation and contrast them with user expectations and experiences of interactions with conversational agents.

Key Findings and Methodology

The paper employs semi-structured interviews to gather qualitative insights into user perspectives on conversations with both humans and agents. The demographic of participants, primarily consisting of technologically proficient individuals, provided a diverse set of responses that the authors analyzed rigorously through Inductive Thematic Analysis. This methodological approach allowed the exploration of characteristics such as mutual understanding, trustworthiness, active listening, and humor, commonly seen as integral to human-human conversation, and the redefinition these experience when interacting with agents.

Disparity in Conversational Objectives

A major insight from the research is the clear distinction that participants draw between social and transactional dialogues. Human-to-human conversation often encompasses both dimensions, serving to establish bonds through interactive exchanges. Conversely, interaction with conversational agents is predominantly perceived as transactional. Users approach these interactions with the expectation of efficient task completion rather than social bonding, viewing agents largely as functional tools rather than potential companions.

Implications on Agent Design and Human-Agent Interaction

The paper discusses the necessity for a paradigm shift in designing conversational agents, proposing that developers ought to recognize and cater to the fundamentally distinct nature of machine-user interactions. This indicates an exploration into a new genre of conversation, one that does not strive to replicate the human conversational experience, but rather fulfills and enhances user expectations and utilitarian needs.

Challenges and Future Directions

The work underscores fundamental challenges such as overcoming user perceptions anchored in existing interactions with IPAs that are inherently non-conversational in a traditional sense. The observed status asymmetry—where machines are seen as mere tools of efficiency—poses a significant barrier to any effort that aims to introduce more human-like social interactions into agent design. Yet, opportunities for adopting agent-based dialogue in specific contexts, such as health care or for users with limited social interaction, present compelling directions for future development.

Moreover, the paper recognizes the potential of exploring non-verbal communication features and other modalities beyond spoken interaction, especially in embodied agents, in overcoming these challenges and enhancing the user experience.

Conclusion

The research provides a comprehensive view into the core expectations and limitations of current conversational agents, underscoring a need for a reimagined framework that accepts the limitations yet explores the potentials of AI in conversational settings. The insights contribute valuable angles to the ongoing discourse in HCI, particularly concerning the integration of contextual, transactional, and emotional dynamics in agent interactions.

This work illustrates that while the path towards truly conversational agents is complex and nuanced, understanding user expectations and preferences is a critical step towards designing systems that not only meet user needs but also innovate conversational paradigms in human-computer interaction. Future research and applications would benefit from pivoting towards this nuanced understanding as they evolve the capabilities of conversational agents.