Convivial Conversational Agents -- shifting toward relationships

Basic Information

• Paper ID: 2510.09516
• Title: Convivial Conversational Agents -- shifting toward relationships
• Authors: Rafael A. Calvo (Imperial College London), Dorian Peters (University of Cambridge)
• Classification: cs.HC (Human-Computer Interaction)
• Publication Year: 2025
• Paper Link: https://arxiv.org/abs/2510.09516

Abstract

Conversational AI (CAI) systems present unprecedented opportunities for scaling service delivery, with governments and enterprises beginning to deploy these systems across various services. The economic rationale is consistent across domains: using CAI to automate time-consuming conversations required for customer, client, or patient support. Building on the authors' work in dementia care, this paper explores the challenges and opportunities of CAI, and how to ensure that essential elements of human flourishing are not lost in the automation process through new conceptualizations.

Research Background and Motivation

Problem Identification

The core problem this research addresses is: How can we maintain the quality of interpersonal relationships and human autonomy during large-scale deployment of conversational AI systems? This is manifested specifically in:

1. Service Supply-Demand Imbalance: Global healthcare service demand far exceeds available supply, particularly in dementia care
2. Social Impact of Automation: CAI as social agents may affect existing interpersonal relationship networks
3. System-Level Negative Effects: Individual-level technology optimization may cause damage at the relationship or group level

Importance of the Problem

The significance of this problem manifests at multiple levels:

• Scale Urgency: Nearly one million dementia patients in the UK, one million new cases annually in the US
• Economic Reality: Traditional care models require 25,000 case officers, which is economically unfeasible
• Social Responsibility: Technology automation may lead to interpersonal alienation and community dissolution

Limitations of Existing Approaches

• Most CAI designs focus on individual user experience while ignoring system-level relationship impacts
• Lack of deep understanding of how conversational agents alter existing interpersonal relationship dynamics
• Traditional HCI research lacks analytical frameworks at the group level

Core Contributions

1. Introduction of "Conviviality" Concept: Applying Ivan Illich's theory of convivial tools to modern conversational AI design
2. Proposing a Relationship-Centered Design Paradigm: Shifting from individual user experience to relationship and community-level impact assessment
3. Establishing a Theory-Practice Bridge: Using Self-Determination Theory (SDT) as an empirical foundation for operationalizing conviviality
4. Providing Concrete Design Recommendations: Four key recommendations guiding the design and evaluation of convivial conversational agents

Methodology Details

Theoretical Framework Construction

Three Core Dimensions of Conviviality

Based on Illich's theory, the authors define conviviality as three interconnected elements:

1. Tolerant Coexistence: The capacity to live together effectively
2. Autonomy: Maintaining individual self-determination rights, avoiding "enslavement"
3. Relatedness: Acknowledging the reality and importance of relationships and interdependence

Integration of Self-Determination Theory (SDT)

The authors align SDT's three fundamental psychological needs with the conviviality concept:

• Autonomy: Volition rather than control; the ability to act autonomously for collective benefit
• Relatedness: Meaningful connection with others
• Competence: Feeling capable and effective (a necessary condition for tool use)

Design Methodology

Systemic Approach

Adopting Voinea's systemic approach while considering both individual and group levels:

• Individual Level: Using SDT scales to measure autonomy and relatedness satisfaction
• Group Level: Requiring new qualitative frameworks to understand relationship dynamics

Participatory Design Principles

Emphasizing stakeholder involvement throughout:

• Patient and family participation in iterative design
• Integration of expert knowledge with user experience
• Cultural sensitivity and value alignment

Experimental Setup

Application Domain: Dementia Care

The authors selected dementia care as the primary application scenario for reasons including:

• Enormous and rapidly growing service needs
• Pronounced relational service characteristics
• Involvement of complex family and community dynamics

Evaluation Framework

While the paper is primarily a conceptual contribution, it proposes a framework for evaluating convivial CAI:

Individual-Level Measurement

• Using SDT's basic psychological needs satisfaction scales
• Self-report measures of autonomy, relatedness, and competence

Relationship-Level Assessment

• Dyadic or group interviews
• Co-design workshops
• Cultural probes methodology
• Ethnographic observation

Case Study Methodology

The paper provides several concrete cases illustrating the application of convivial design:

1. Positive Case: Mindfulness conversational agent designed by Seah et al., supporting relationships between dementia patients and caregivers
2. Negative Case: Unauthorized LLM role-playing research in Reddit communities

Experimental Results

Theoretical Validation

The paper primarily provides conceptual contributions, validating the theoretical framework through case analysis:

Success Factors in the Seah Case

• Relationship-Centered: Specifically designed to improve relationships between dementia patients and caregivers
• Participatory Approach: Joint participation of experts, caregivers, and patients in design
• Iterative Evaluation: Continuous optimization to support conviviality goals

Lessons from the Reddit Case Failure

• Lack of Consent: Creating fake accounts without community consent
• Trust Damage: Harming trust relationships within human communities
• Value Conflict: Commercial models conflicting with care and coexistence values

Design Impact Analysis

Through comparative analysis, the paper demonstrates how a conviviality perspective changes CAI design:

Related Work

Conversational AI Applications in Healthcare

• Technical support for intelligent home care systems
• Automated periodic check-ins
• Immediate response to urgent needs
• Home wellness interventions

Autonomy Research in Human-Computer Interaction

• Bennett et al.'s understanding of human agency and autonomy in HCI
• Evidence-based approaches to supporting wellbeing in technology design

Relational Service Theory

• Cipolla's concept of relational services
• Interpersonal relationships as core components of service production

Conviviality in Multi-Agent Systems

While conviviality concepts have been applied in multi-agent systems research, measurement methods for conviviality in Illich's sense remain lacking.

Conclusions and Discussion

Main Conclusions

1. Necessity of Paradigm Shift: From "how to scale through automating human tasks" to "how automation can support humans in expanding their impact"
2. Value of Relationship-Centered Design: Maintaining interpersonal relationship quality is more important than simply improving interaction efficiency
3. Criticality of Participatory Methods: True convivial AI tools cannot be realized without stakeholder participation

Four Core Recommendations

Recommendation 1: Understanding Individual and Group Perspectives and Dynamics

• Use SDT methods to measure individual-level autonomy and relatedness
• Develop new methods to gain insights into collective community perspectives
• Draw from sociology, anthropology, and indigenous methodologies

Recommendation 2: Assess Automation's Impact on Interpersonal Relationships

• Develop deep understanding of relationship nuances, dynamics, and key benefits
• Collaborate with dyads or groups for interviews and co-design
• Use ethnographic methods to observe relationship dynamics in natural settings

Recommendation 3: Include Stakeholders Throughout the Design Process

• Overcome challenges of participatory design in healthcare settings
• Create partnerships with stakeholders creatively
• Address relationships in ways that support coexistence, autonomy, and relatedness

Recommendation 4: Reframe the Problem

• Acknowledge that technology may not be the optimal solution
• Consider increasing human resources as an alternative
• Learn from the success of community health worker movements

Limitations

1. Lack of Measurement Tools: Currently no direct measurement method for conviviality in Illich's sense
2. Difficulty in Group-Level Analysis: Existing SDT measurements rely on individual self-report, making group-level analysis challenging
3. Implementation Challenges: Participatory design in healthcare settings faces time, resource, and institutional constraints
4. Business Model Conflict: Profit-driven business models often conflict with conviviality principles

Future Directions

1. Development of New Measurement Methods: Develop measurement tools that capture conviviality at relationship and community levels
2. Interdisciplinary Collaboration: Collaborate with anthropology, sociology, and STS fields to develop assessment frameworks
3. Relational-Computing Interaction: May require developing new RCI or CCI research fields
4. Long-Term Impact Research: Evaluate long-term social and relationship impacts of CAI deployment

In-Depth Evaluation

Strengths

Innovation in Theoretical Contribution

1. Concept Integration: Successfully connecting Illich's 50-year-old theory with modern AI technology
2. Interdisciplinary Perspective: Dual perspective combining design engineering and social psychology
3. Systemic Thinking: Providing multi-level analytical framework spanning individual-relationship-community

Value of Practical Guidance

1. Operationally Concrete: Four recommendations provide clear guidance for actual design
2. Case-Driven: Effectively illustrating theoretical application through positive and negative cases
3. Contemporary Relevance: Selecting dementia care as an urgent social problem

Clarity of Writing

1. Logical Structure: Clear logic from problem identification through theory construction to practical recommendations
2. Concept Explanation: Sufficient background and explanation of key concepts like conviviality
3. Balanced Expression: Acknowledging technology potential while pointing out risks

Weaknesses

Methodological Limitations

1. Lack of Empirical Validation: Primarily conceptual contribution lacking large-scale empirical research
2. Insufficient Measurement Tools: Acknowledging lack of direct conviviality measurement methods
3. Incomplete Assessment Framework: Group-level assessment methods still require further development

Implementation Challenges

1. Insufficient Cost Consideration: Participatory design and relationship-centered approaches may increase development costs
2. Scaling Difficulties: Unclear how to maintain conviviality principles in large-scale deployment
3. Commercial Viability: Conflicts with existing business models may hinder adoption

Theoretical Depth

1. Cultural Specificity: Conviviality concept may have cultural variations requiring more discussion
2. Insufficient Trade-off Analysis: Lacking deep analysis of how to balance efficiency against relationship quality
3. Fuzzy Boundary Conditions: Unclear standards for when to choose technology solutions vs. human resource solutions

Impact

Contribution to the Field

1. Paradigm Shift: Potentially driving HCI field shift from individual-centered to relationship-centered approaches
2. Interdisciplinary Bridge: Providing new pathways for combining technology design with social sciences
3. Policy Implications: Offering new thinking frameworks for AI governance and health policy

Practical Value

1. Design Guidance: Providing practical design principles for CAI developers
2. Assessment Framework: Offering new tools for evaluating social impacts of AI systems
3. Risk Identification: Helping identify and prevent potential negative impacts of AI deployment

Reproducibility

1. Clear Theoretical Framework: Conviviality concept and SDT integration framework easy to understand and apply
2. Transferable Methodology: Four recommendations applicable to CAI design in other domains
3. Reference Cases: Provided cases offer concrete references for other researchers

Applicable Scenarios

Most Suitable Application Domains

1. Healthcare: Particularly chronic disease management requiring long-term relationship building
2. Education: Maintaining teacher-student relationships and peer learning environments
3. Community Services: Public services requiring maintenance of community cohesion
4. Family Support: Services involving family dynamics and intergenerational relationships

Application Conditions

1. Strong Relationality: Service itself has strong interpersonal relationship characteristics
2. Value Sensitivity: Domains involving cultural values and ethical considerations
3. Long-Term Impact: Application scenarios requiring consideration of long-term social impacts
4. Multiple Stakeholders: Involving multiple interconnected stakeholder groups

References

The paper cites 29 important references covering the following key areas:

1. Theoretical Foundations: Illich (1973) on convivial tool theory, Ryan & Deci (2017) on self-determination theory
2. HCI Research: Bennett et al. (2023) on human autonomy research, Peters & Calvo (2023) on technology design research
3. Healthcare AI Applications: Guo et al. (2024) on mental health applications review, Thirunavukarasu et al. (2023) on medical LLMs
4. Design Methods: Sadek et al. (2023) on conversational agent co-design review
5. Social Impact: Pariser (2011) on filter bubbles, Burton et al. (2024) on collective intelligence research

Overall Assessment: This is a high-quality paper with important theoretical contributions and practical guidance value. The authors successfully combine classical social theory with modern AI technology, proposing a relationship-centered CAI design paradigm. While further development is needed in empirical validation and implementation details, it opens new research directions for the HCI field and holds significant importance for responsible AI development.