Relational Dissonance in Human-AI Interactions: The Case of Knowledge Work

Paper Title

Relational Dissonance in Human-AI Interactions: The Case of Knowledge Work

Publication Info

• Topic area: Human-AI interaction in knowledge work
• Keywords: relational dissonance, human-AI interaction, anthropomorphic conversational agents, knowledge work, qualitative research, relational transparency, ontological ambiguity, relational ambiguity, human-computer interaction, AI governance

Background and Problem

• Problem / challenge: Existing human-computer interaction (HCI) frameworks inadequately describe the relational dynamics that emerge when humans engage with anthropomorphic conversational agents. These systems challenge established conceptual categories by blending tool-like functionality with human-like relational cues, leading to difficulties in maintaining consistent relational stances.
• Significance: Understanding these dynamics is critical for knowledge work, where AI systems are increasingly used for tasks requiring human agency, control, and interpretive skill. Misalignment in relational dynamics can have psychological, relational, and professional consequences.
• Motivation and related work: Prior research has explored AI's instrumental use, its relational effects, and its anthropomorphic qualities. However, these studies often fail to capture the dynamic, implicit relational negotiations that occur during human-AI interactions. This paper addresses the gap by introducing the concept of relational dissonance.

Solution

• Proposed approach: The authors propose relational dissonance as a novel analytical construct to describe the divergence between individuals’ explicit framing of AI systems and the relational dynamics that unfold during interactions.
• Novelty:
  1. Introduction of relational dissonance as a theoretical framework for understanding human-AI interactions.
  2. Development of a multilayered workshop methodology to study relational dynamics.
  3. Insights into how relational dissonance manifests in qualitative research, a domain with high expectations of human agency and control.
• Procedure and key techniques:
  • Conducted three workshops with 22 qualitative researchers, combining individual and group activities.
  • Used interpretative phenomenological analysis (IPA) and thematic analysis (TA) to examine individual and collective relational dynamics.
  • Analyzed pre-session surveys, chatlogs, image-selection exercises, group discussions, and worksheets to identify relational configurations and dynamics.

Results

• Concrete findings:
  • Identified five relational configurations: Director, Trainer, Partner, Student, and Consumer.
  • Observed frequent shifts between configurations during interactions, driven by both human and AI behaviors.
  • Documented relational dissonance as a persistent tension between users’ explicit framing of AI as a tool and the implicit relational dynamics that emerged.
• Advantage over baselines:
  • Provides a nuanced framework that captures dynamic relational shifts, unlike prior static role-based models.
  • Highlights the relational work required to navigate human-AI interactions, which is often overlooked in existing HCI research.
• Experiments / evaluation:
  • Workshops included pre-session surveys, hands-on AI interaction tasks, peer observation, metaphor elicitation, group discussions, and collaborative exercises.
  • Data sources included survey responses, chatlogs, selected visual metaphors, and group discussion transcripts.
  • Analysis revealed patterns of relational dissonance and collective sense-making through themes like depth, uncertainty, and othering.
• Limitations and future work:
  • Conceptual limitations: Need for quantitative measures of relational dissonance and further validation across diverse contexts.
  • Methodological limitations: Artificial workshop setting and short interaction periods may have influenced findings.
  • Empirical limitations: Results are specific to qualitative researchers and may not generalize to other populations or tasks.
  • Future work: Longitudinal studies, quantitative measures, and exploration of relational transparency in system design and governance.

Summary

This paper introduces relational dissonance as a framework to understand the implicit dynamics of human-AI interactions, particularly in knowledge work. Through workshops with qualitative researchers, the study reveals how relational dynamics emerge and shift during interactions with anthropomorphic conversational agents, often diverging from users’ explicit intentions. The findings highlight the need for relational transparency to address the tensions and relational work inherent in these interactions. This work provides a foundation for designing AI systems and policies that better account for the relational complexities of human-AI engagement.