Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mixed-Initiative Context: Structuring and Managing Context for Human-AI Collaboration

Haichang Li, Qinshi Zhang, Piaohong Wang, Zhicong Lu

Abstract

In the human-AI collaboration area, the context formed naturally through multi-turn interactions is typically flattened into a chronological sequence and treated as a fixed whole in subsequent reasoning, with no mechanism for dynamic organization and management along the collaboration workflow. Yet these contexts differ substantially in lifecycle, structural hierarchy, and relevance. For instance, temporary or abandoned exchanges and parallel topic threads persist in the limited context window, causing interference and even conflict. Meanwhile, users are largely limited to influencing context indirectly through input modifications (e.g., corrections, references, or ignoring), leaving their control neither explicit nor verifiable. To address this, we propose Mixed-Initiative Context, which reconceptualizes the context formed across multi-turn interactions as an explicit, structured, and manipulable interactive object. Under this concept, the structure, scope, and content of context can be dynamically organized and adjusted according to task needs, enabling both humans and AI to actively participate in context construction and regulation. To explore this concept, we implement Contextify as a probe system and conduct a user study examining users' context management behaviors, attitudes toward AI initiative, and overall collaboration experience. We conclude by discussing the implications of this concept for the HCI community.

Mixed-Initiative Context: Structuring and Managing Context for Human-AI Collaboration

Abstract

In the human-AI collaboration area, the context formed naturally through multi-turn interactions is typically flattened into a chronological sequence and treated as a fixed whole in subsequent reasoning, with no mechanism for dynamic organization and management along the collaboration workflow. Yet these contexts differ substantially in lifecycle, structural hierarchy, and relevance. For instance, temporary or abandoned exchanges and parallel topic threads persist in the limited context window, causing interference and even conflict. Meanwhile, users are largely limited to influencing context indirectly through input modifications (e.g., corrections, references, or ignoring), leaving their control neither explicit nor verifiable. To address this, we propose Mixed-Initiative Context, which reconceptualizes the context formed across multi-turn interactions as an explicit, structured, and manipulable interactive object. Under this concept, the structure, scope, and content of context can be dynamically organized and adjusted according to task needs, enabling both humans and AI to actively participate in context construction and regulation. To explore this concept, we implement Contextify as a probe system and conduct a user study examining users' context management behaviors, attitudes toward AI initiative, and overall collaboration experience. We conclude by discussing the implications of this concept for the HCI community.

Paper Structure

This paper contains 58 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Context in Human--AI Collaboration
  4. Structure of Collaboration
  5. Mixed-Initiative Interaction
  6. Mixed-Initiative Context
  7. The Concept of Mixed-Initiative Context
  8. Context as an Interactive Object
  9. Mixed-Initiative Interaction over Context
  10. Capabilities Enabled by Mixed-Initiative Context
  11. Probe System: Contextify
  12. Role and Design Rationale
  13. System Overview
  14. Interaction Design
  15. User Journey
  16. ...and 43 more sections

Figures (2)

  • Figure 1: The Mixed-Initiative Context framework. Left: Traditional vs. our proposed interaction model. Right: Context hierarchy (Unit, Structure, Pattern) and the interaction layer driving continuous structural updates and user mental adaptation.
  • Figure 2: Flat conversational context collapses heterogeneous task elements into a single linear transcript, making boundaries between accepted, rejected, temporary, and parallel lines of work difficult to maintain. This mismatch can lead to context pollution, fading instructions, and cross-thread interference, diverging from how users mentally organize complex work as structured, selective, and non-linear.