Reciprocal Self-disclosure and Active Listening in Human-agent Relationship Development = 인간-에이전트 관계 발달을 위한 에이전트의 상호적 자기노출과 적극적 경청

국문 초록 (Abstract)

대화형 에이전트는 단순한 작업지향적 보조 역할을 넘어, 이제 일상적으로 사람들에게 정서적 지지를 제공하고 동반자 역할을 수행하고 있다. 그런데 대화형 에이전트는 사람의 언어로 대화하지만, 실제로는 진정한 주체성을 가지지 않는다는 점에서 관계에서의 지위가 모호하다. 사람과 에이전트 간의 상호작용이 빠르게 늘고 있음에도 불구하고, 대인 관계의 형성 및 발달과 관련해 알려진 규범들이 사람과 에이전트 간의 관계에 어디까지 적용되는지, 그리고 에이전트의 구조적 제약이 어떤 지점에서 관계 형성 과정을 바꾸는지 충분히 밝혀지지 않은 지점들이 있다.
이 연구는 대인 커뮤니케이션 이론을 출발점으로 삼아, 관계 형성과 관련된 핵심 개념들과 규칙들이 대화형 에이전트에도 적용되는지, 적용된다면 어떤 방식으로 나타나는지를 살폈다. 이를 위해 이 연구는 선행연구를 바탕으로 개별 상호작용 경험이 초기 관계 평가를 형성하는 과정을 설명하는 모형을 제안하고, 상호작용성과 관련한 두가지 커뮤니케이션적 개념에 주목했다. 첫째는 상호적 자기노출이다. 자기노출의 맥락에서 상호성은 자신에 대한 정보를 서로 주고받는 특성을 의미한다. 상호성은 순서(턴) 단위 등가성, 깊이 단위 등가성, 순차적 조정의 3가지 특징을 가진다. 둘째는 적극적 경청이다. 적극적 경청은 상대의 말에 주의를 기울이면서 감정을 인정하는 과정을 통해 상대가 충분히 이해 받고 있다고 느끼게 만든다. 이 연구는 상호적 자기노출과 적극적 경청을 관계 형성에 핵심적인 에이전트 행동으로 보고, 이것이 사람의 상호작용 경험의 행동, 인지, 정서적 측면에 어떤 영향을 주는지, 그리고 그 영향이 친밀감과 애착과 같은 관계 평가를 거쳐 안녕감과 향후 이용 의도에 어떤 영향을 주는지 통합적으로 검증했다.
이를 위해 상호적 자기노출 관련 4조건과 적극적 경청 관련 2조건을 결합한 피험자 간 요인 설계의 실험을 진행했다. 참가자 146명은 프롤리픽을 통해 모집되었으며, 참가자들은 사전 설문 후 챗봇과 대화를 한 뒤 사후 설문에 응답했다. 상호성 조건은 고정 조건들, 즉 일관적으로 낮은 수준을 유지하는 조건(RC_F1), 일관적으로 높은 수준을 유지하는 조건(RC_F5), 고정적으로 순차 증가하는 조건(RC_FG), 및 사용자의 직전 자기노출에 맞추어 수준을 조정하는 적응 조건(RC_ADPT)으로 구성되었다. 추가적으로 이 적응 조건에서는 사람의 자기노출 깊이에 따라 챗봇의 자기노출 깊이도 상이하게 나타나기 때문에, 이 조건의 경우 탐색적으로 중앙값 기준 고깊이 집단(RC_ADPT_Hm)과 저깊이 집단(RC_ADPT_Lm)으로 추가 분류해 비교했다.
분석 결과, 상호적 자기노출과 적극적 경청은 상호작용 경험의 행동적 측면(자기노출의 깊이와 양)을 변화시키지 않았다. 적극적 경청은 상호작용 경험의 인지적 경험과 정서적 측면에 긍정적인 영향을 미쳐, 챗봇에 대한 지각된 반응성과 상호작용의 즐거움을 유의하게 높이는 것으로 나타났다. 상호적 자기노출의 경우 상호성의 효과는 적응 조건에서 자기노출의 깊이가 깊을 때 상호작용의 즐거움을 유의하게 높이는 것으로 나타났다. 즉 고깊이 적응 조건(RC_ADPT_Hm)의 참가자는 저깊이 적응 조건 및 모든 고정 조건들의 참가자보다 더 높은 상호작용의 즐거움을 보고했다. 친밀감과 애착을 유의하게 설명한 변수는 상호작용 경험의 인지적 측면과 정서적 측면, 즉 지각된 반응성과 상호작용의 즐거움이었고, 사람의 행동적 측면, 즉 자기노출 깊이와 양은 어떤 결과에서도 유의한 영향을 보이지 않았다. 친밀감은 긍정 정서와 부정정서를 예측했고, 애착은 긍ㆍ부정 정서 및 삶의 만족을 예측했으며, 두 변수 모두 향후 이용 의도를 예측했다. 경로분석 결과 적극적 경청이 지각된 반응성을 높이고, 이 지각된 반응성이 애착을 거쳐 삶의 만족과 향후 이용의도로 이어지는 인지적 경로와, 고깊이 상호성이 상호작용의 즐거움을 높이고, 상호작용의 즐거움이 친밀감을 거쳐 정서적 안녕감과 향후 이용의도로 이어지는 정서적 경로를 밝힐 수 있었다. 반면 외로움은 어떤 경로에서도 유의한 영향을 받지 않았는데, 이는 외로움이 상대적으로 안정적인 성향 특성이며 실제 사회적 연결망의 접근성과 더 밀접할 수 있음을 시사한다.
이러한 결과는 대화형 에이전트와의 초기 관계 형성이 사람의 행동보다는 지각된 반응성이나 상호작용의 즐거움과 같은 주관적 경험에 더 크게 좌우됨을 의미한다. 또한 단순히 에이전트의 자기노출의 깊이가 순차적으로 증가하는 것만으로는 충분하지 않았고, 적절한 깊이 이상에서 사람의 자기노출 수준에 맞추어 조정하는 상호적 행동이 상호작용의 즐거움을 높였다. 적극적 경청은 반응성을 높이는 기본 조건으로서 안정적인 효과를 보였으며, 주로 인지적 경로를 통해 관계 평가에 기여했다. 종합하면, 인간과 에이전트의 관계 형성은 행동 규범의 등가성보다는, 사용자가 주관적으로 느끼는 조율감과 상호작용의 질에 더 영향을 받는다. 이는 대화형 에이전트를 관계적 파트너로 설계할 때, 적극적 경청을 기본적으로 수행하는 한편 적응형 자기노출을 구현하되 그 깊이 수준을 정교하게 조절함으로써 정서적 몰입과 인지적 안정감을 균형 있게 제공해야 함을 시사한다.

다국어 초록 (Multilingual Abstract)

Conversational agents have become increasingly embedded in everyday life, moving beyond task-oriented assistance to roles that involve emotional support and companionship. As these systems now function as interlocutors that speak in human language yet lack genuine agency, their relational status remains theoretically ambiguous. Despite rapid growth in human–agent engagement, it remains unclear which elements of interpersonal relationship formation extend to human–agent interaction, and where agent-specific constraints alter these processes. This study uses interpersonal communication theory as a starting point to examine whether and how relational principles transfer to conversational agents, while specifying how structurally constrained partners may produce divergent mechanisms.
To address this question, the study builds on prior work to propose a process account of initial relational evaluations emerge from episode-level interactional experiences. Within this framework, interactivity is examined through two dimensions particularly consequential for relationship formation: reciprocity and responsiveness. Reciprocal self-disclosure enacts reciprocity through coordinated mutual sharing, with reciprocity decomposed into turn-based equivalence, depth-based equivalence, and sequential adaptivity. Active listening realizes responsiveness by demonstrating attention, understanding, and validation, shaping perceived partner responsiveness and the felt quality of being heard. Together, these mechanisms are expected to structure behavioral, cognitive, and affective layers of interactional experience, which then inform relational evaluations and ultimately influence downstream outcomes.
Empirically, the present study focuses on reciprocal self-disclosure and active listening as two designable agent behaviors that instantiate interactivity in relationship formation. It first examines whether these behaviors influence three categories of interactional experience: behavioral aspect (users’ self-disclosure depth and amount), cognitive aspect (perceived responsiveness), and emotional aspect (interactional enjoyment). These interactional experiences are then analyzed as predictors of relational perceptions, specifically intimacy and attachment to the agent. Finally, the study investigates whether such relational perceptions have downstream implications for users’ well-being and future engagement intentions. This framework enables an integrated examination of how specific agent communication behaviors may cascade from momentary interactional processes to relational and psychological consequences.
The experiment employed a 4 × 2 between-subjects factorial design manipulating (1) the reciprocity of the agent’s self-disclosure and (2) the chatbot’s listening style (active vs. non-active). Participants (N = 146), recruited from Prolific, completed a pre-survey, engaged in a five-round conversation with the chatbot, and then responded to a post-survey. The four reciprocity conditions included consistently low disclosure (RC_F1), consistently high disclosure (RC_F5), a fixed-gradual increase (RC_FG), and an adaptive condition (RC_ADPT) in which the chatbot’s disclosure level dynamically adjusted to the participant’s prior disclosure. Because the adaptive condition produced a continuous range of disclosure depths, it was further stratified into high-median (RC_ADPT_Hm) and low-median (RC_ADPT_Lm) adaptive subgroups for exploratory comparisons. Analyses included planned-contrast tests and regression models to evaluate the effects of reciprocity and active listening on interactional, relational, and downstream well-being outcomes, and supported paths were integrated into a path analysis to examine the overall process through which agent behaviors shaped users’ experiences and evaluations.
The results showed that neither reciprocal self-disclosure nor active listening influenced users’ behavioral disclosure. Active listening reliably increased perceived responsiveness and interactional enjoyment. Reciprocity effects emerged only when the adaptive strategy resulted in higher disclosure depth; participants in the high-depth adaptive condition (RC_ADPT_Hm) reported significantly greater enjoyment than those in the low-depth adaptive and all fixed conditions. Across outcomes, perceived responsiveness and enjoyment were the strongest and most consistent predictors of relational evaluations, while users’ own disclosure depth and amount showed no effects. In turn, intimacy predicted positive and negative affect, and attachment predicted life satisfaction; both relational constructs strongly predicted future engagement intention. Path analysis supported an integrated process model in which active listening and high-depth reciprocity shaped interactional experiences, which then predicted intimacy and attachment, leading to downstream well-being and usage intention. These findings indicate that initial relationship formation with conversational agents is driven primarily by episode-level subjective experience, particularly perceived responsiveness and interactional enjoyment, rather than by users’ disclosure behavior.
The findings indicate that reciprocal self-disclosure enhances interactional enjoyment only when adaptive strategy is coupled with appropriately deep disclosure; mere adaptiveness was insufficient, and enjoyment increased only when the adaptive strategy aligned with users’ deeper disclosures. In contrast, active listening exerted reliable effects on perceived responsiveness and played a primarily cognitive role. Despite the salience of these communication behaviors, neither reciprocity nor active listening altered users’ self-disclosure behavior, suggesting that behavioral reciprocity norms weaken under temporally segmented or procedurally constrained interaction. Instead, relational evaluations were driven by users’ subjective experiences of responsiveness and interactional enjoyment, which differentially predicted intimacy and attachment and, in turn, shaped affective well-being, life satisfaction, and future engagement intention. Loneliness remained unaffected, reflecting its more trait-like stability and dependence on access to human or agent-based social channels. Collectively, these results suggest that human–agent relationship formation relies less on users’ expressive behaviors and more on perceived attunement and interactional quality, highlighting both the potentials and the boundaries of conversational agents as relational partners.
Building on these results, several implications can be discussed. In human–agent interaction, sequential adaptivity enhances interactional enjoyment only when it results in appropriately deep reciprocity, indicating that adaptivity alone is insufficient without effective depth calibration. Relationship formation is driven more by users’ subjective experience of responsiveness and enjoyment than by mutual behavioral exchange. These results refine existing frameworks by distinguishing cognitive and affective pathways, with responsiveness fostering attachment and life satisfaction, and reciprocal self-disclosure fostering intimacy and affective well-being. Practically, the study highlights the need for conversational agents to implement sequentially adaptive self-disclosure, treat active listening as a baseline requirement, and balance emotional engagement with cognitive assurance. Taken together, these insights broaden the scope of interpersonal communication theories within the human–agent context and articulate design principles for developing relationally intelligent agents that foster meaningful and psychologically supportive user experiences.

목차 (Table of Contents)

• 1. Introduction 1
• 1.1. Research background and overview 1
• 1.2. Definition of an agent 7
• 1.3. Approaches to relationships with conversational agents 8
• 2. Review of Prior Research 11
• 2.1. Why relationships with conversational agents matter 11
• 2.2. What it means for humans to form relationships 17
• 2.3. Relationships with conversational agents: Potentials and limitations 20
• 2.4. How interactional experiences shape relational evaluations 29
• 2.5. Interactivity as reciprocal and responsive communication 33
• 3. Theoretical Background 40
• 3.1. Reciprocity in self-disclosure 40
• 3.1.1. Conceptual foundations on reciprocity 43
• 3.1.2. Dimensions of reciprocity in interpersonal theories 48
• 3.1.3. Underexamined dimension of reciprocity in human-agent interaction: sequential adaptivity 50
• 3.1.4. Empirical evidence on sequential adaptivity 53
• 3.2. Active listening 58
• 3.2.1. Listening as a relational process 59
• 3.2.2. Active listening: conceptualization and behavioral manifestations 60
• 3.2.3. Active listening in conversational agents 62
• 3.3. Interactional experience 65
• 3.3.1. Behavioral aspect: Self-disclosure patterns 66
• 3.3.2. Cognitive aspect: Perceived responsiveness 68
• 3.3.3. Affective aspect: Interactional enjoyment 70
• 3.4. Indicators of relationship development 72
• 3.4.1. Intimacy 74
• 3.4.2. Attachment 75
• 3.4.3. Distinguishing intimacy and attachment 77
• 3.5. Well-being in relationships 78
• 3.6. Future interaction expectations / Future intention to use 84
• 4. Hypothesis and Research Question 88
• 4.1. Hypothesis 89
• 4.2. Research questions 93
• 5. Method 95
• 5.1. Power analysis and participants 95
• 5.2. Interaction interface and implementation 96
• 5.3. Study design 96
• 5.4. Procedure 100
• 5.4.1. Pre-survey 100
• 5.4.2. Chat with VoxBot 100
• 5.4.3. Post-survey 101
• 5.5. Chatbot design 102
• 5.5.1. Introduction phase 104
• 5.5.2. Small talk session 105
• 5.5.3. Main conversation session 105
• 5.5.4. Evaluation session 107
• 5.5.5. Reflection session 108
• 5.6. Message bank 109
• 5.6.1. Messages for small-talk 109
• 5.6.2. Messages for the main conversation 109
• 5.6.3. Topics for the reflection session 111
• 5.7. Measures 112
• 5.7.1. Pre-survey (Control variables) 112
• 5.7.2. Manipulation checks 114
• 5.7.3. Depth and amount of self-disclosure 115
• 5.7.4. Post-survey (Dependent Variables) 119
• 5.8. Analysis procedure 122
• 6. Result 125
• 6.1. Manipulation check 125
• 6.1.1. Reciprocity perception 125
• 6.1.2. Active listening perception 127
• 6.1.3. Overlap between reciprocity and active listening manipulations 127
• 6.2. Preliminary analyses of key variables 128
• 6.2.1. Factor structure and reliability 128
• 6.2.2. Descriptive statistics of measured variables 134
• 6.2.3. Correlations among main variables 136
• 6.3. Hypothesis testing results 139
• 6.3.1. Main hypothesis testing 140
• 6.3.2. Path analysis 162
• 6.4. Additional analysis 165
• 6.4.1. Interaction between reciprocity and active listening 166
• 6.4.2. Exploratory analysis: Decomposed analysis of disclosure depth dimensions 168
• 6.4.3. Robustness checks for the reciprocity effect 170
• 6.5. Summary of the study and key findings 173
• 6.5.1. Effects of reciprocity and active listening on interactional experience 175
• 6.5.2. Relationships between interactional experience and relational evaluation 176
• 6.5.3. Downstream outcomes: Well-being outcomes and future intention 177
• 6.5.4. Psychological processes 178
• 6.5.5. Path analysis 178
• 7. Discussion 180
• 7.1. Discussion of the findings 180
• 7.1.1. Reciprocity Effect: The importance of reciprocity 180
• 7.1.2. Active listening effect - Strong cognitive, weak affective 186
• 7.1.3. Limits of reciprocity and active listening 190
• 7.1.4. Behavior-experience dissociation 192
• 7.1.5. Relational evaluations to well-being outcomes and future intention to use 197
• 7.2. Theoretical and practical implications 208
• 7.2.1. Theoretical implications 208
• 7.2.2. Practical implications 226
• 7.3. Limitations of the study 237
• 7.3.1. Design and methodological constraints 238
• 7.3.2. Measurement, operationalization, and conceptual boundaries 239
• 7.3.3. Sample and generalizability limitations 241
• 7.4. Suggestions for future research 243
• 7.4.1. Advanced experimental designs 243
• 7.4.2. Refined theoretical mechanisms 247
• 7.4.3. Diverse populations and contexts 251
• 8. Conclusion 254
• Reference 258
• Supplementary Materials 300
• Supplementary A. Additional method details 300
• Supplementary B. Additional statistical tables 325
• Supplementary C. Full survey instrument 372
• 초 록 391