Co-Disclosing the Computer: LLM-Mediated Computing through Reflective Conversation

Paper Title

Co-Disclosing the Computer: LLM-Mediated Computing through Reflective Conversation

Publication Info

• Topic area: Interaction paradigms in human-computer interaction with large language models.
• Keywords: LLM-mediated computing, reflective conversation, postphenomenology, co-disclosure, human-computer interaction, interpretive systems, emergent computing, interaction design, mediation, computational environments.

Background and Problem

• Problem / challenge: Traditional application-centered computing relies on fixed interfaces and predefined capabilities, limiting how users engage with and imagine computing. Current paradigms do not fully address the dynamic, interpretive potential of LLMs in reshaping computing environments.
• Significance: Understanding and designing for LLM-mediated computing can redefine the role of computers, enabling more fluid, relational, and context-sensitive interactions that align with human intentionality.
• Motivation and related work: Prior work in HCI critiqued static, representational systems and proposed more malleable, situated, and embodied approaches. However, these approaches often rely on explicit user manipulation and do not account for the interpretive and generative capabilities of LLMs. This paper builds on postphenomenology, reflective design, and conversational interaction to address this gap.

Solution

• Proposed approach: LLM-mediated computing, a paradigm where large language models dynamically generate code, interfaces, and computational environments in response to user intent, enabling emergent and interpretive interaction.
• Novelty:
  1. A new interaction metaphor: reflective conversation, emphasizing negotiation, responsiveness, and co-disclosure.
  2. A postphenomenological analysis of the human–LLM–computer relation, introducing the concept of co-disclosure.
  3. A redefinition of computing as a relational and emergent activity, moving beyond static application-centered paradigms.
• Procedure and key techniques:
  • Introduce the concept of LLM-mediated computing, where functionality is dynamically assembled through user input and LLM interpretation.
  • Illustrate the paradigm with examples (e.g., dynamic email composition, garden design vignette).
  • Analyze the triadic relationship between human, LLM, and computer, emphasizing sequential reconfiguration and interpretive mediation.
  • Develop the metaphor of reflective conversation, grounded in conversation analysis and Schön’s reflective practice.
  • Extend postphenomenological analysis to include co-disclosure, where the computer’s identity and functionality are constituted in use.

Results

• Concrete findings:
  • LLM-mediated computing enables dynamic reconfiguration of computational environments, where functionality emerges through interaction (e.g., email composition, garden design).
  • Interaction unfolds as a generative feedback loop: user intent → LLM interpretation → computer reconfiguration → user engagement.
  • The computer is no longer a stable substrate but an emergent, co-constructed medium.
• Advantage over baselines:
  • Moves beyond static, application-centered paradigms by enabling interpretive, relational, and emergent computing.
  • Integrates multimodal inputs and supports repair, ambiguity, and reflection as natural parts of interaction.
• Experiments / evaluation:
  • Conceptual and illustrative examples (email composition, garden vignette) demonstrate the paradigm’s potential.
  • Proposed evaluation methods include trajectory analysis (tracking interactional paths) and repair audits (assessing support for clarification and revision).
• Limitations and future work:
  • Empirical studies are needed to validate the conceptual framework and explore real-world user experiences.
  • Future research could focus on cultivating literacy for reflective conversation and designing systems that balance openness with accountability.

Summary

This paper introduces LLM-mediated computing, a paradigm where large language models dynamically generate computational environments in response to user intent, redefining the computer as an emergent and relational medium. The authors propose the metaphor of reflective conversation to describe this interaction, emphasizing negotiation, responsiveness, and co-disclosure. Using postphenomenology, they analyze how the computer’s identity and functionality are constituted in use. The paper outlines design principles (e.g., legibility, repair, multimodality) and evaluation methods (e.g., trajectory analysis, repair audits) to support this paradigm. Future work is needed to empirically explore how users engage with reflective conversation and how systems can balance interpretive openness with ethical accountability.