Daily Note: Efficient and Non-Obtrusive Deferral by Intelligent Systems

These notes are summary of concepts presented in “Intelligibility and Accountability: Human Considerations in Context Aware Systems.”

Victoria Bellotti and Keith Edwards. 2001. Intelligibility and accountability: human considerations in context-aware systems. Hum.-Comput. Interact. 16, 2 (December 2001), 193–212. https://doi.org/10.1207/S15327051HCI16234_05

  1. Component-Based Frameworks for Intelligent Systems
    • Separation of action-taking applications from sensing and inferencing architecture
    • Challenges in providing visibility and control in component-based approaches
    • Difficulty in representing human and social contexts deterministically
  2. Challenges in Handling Improvisation
    • Contingent and non-deterministic nature of human interactions (Suchman, 1987)
    • Poor user adherence to updating system representations of their state (Bellotti & Sellen, 1993)
    • Devices can handle only basic, well-defined contextual aspects autonomously
  3. Complexity of Engineering Context-Aware Systems
    • Greater unpredictability increases sensing and system complexity
    • Difficulty in predefining outcomes for highly variable conditions
      • Example: Air conditioning systems unable to autonomously determine user preferences
  4. User Involvement in Action Outcomes
    • Importance of user understanding of system actions and reasoning
    • Systems should involve users to ensure acceptability of outcomes
    • Intelligibility: Systems must explain their knowledge, methods, and actions
    • Accountability: Systems must enforce user accountability for actions impacting others
  5. Principles for Intelligibility and Accountability
    • Inform users of system capabilities and current context understandings
    • Feedback mechanisms
      • Feedforward: Indicate potential outcomes of user actions
      • Confirmation: Communicate ongoing or completed actions
    • Identity and action disclosure, especially for restricted information sharing
    • Provide user control over system and other user actions impacting them
  6. Contextual Awareness and Social Rules
    • Context-aware systems must consider human-salient details
      • Presence, identity, arrival, departure, status, and activity
      • Social rules and technical implementations governing interactions.
    • Need for explicit cues about technically enabled contexts (Goffman, 1959)
    • Systems must alert participants when sessions commence and provide adjustment options
  7. Feedback Types for User Interaction
    • Feedforward: Visual aids like flashing cursors, resizing indicators
    • In-process feedback: Progress indicators (e.g., hourglass icons)
    • Confirmation: Displays confirming actions or changes
  8. Information Capture, Construction, and Accessibility
    • Users need transparency about
      • Captured information and its use
      • System’s interpretation, storage, and sharing of data
      • Persistence and access to their information by others
    • Enforce mechanisms for identity and action disclosure
  9. Minimizing Human Effort Through Effective Control
    • Match control strategies to uncertainty levels
      • Slight doubt: Allow users to correct system actions
      • Significant doubt: Enable users to confirm intended actions
      • High uncertainty: Offer users explicit choices for actions
    • Effective control ensures users achieve desired outcomes with minimal effort
  10. Designing for Rich Representations of Context
    • Avoid over-reliance on simplistic representations and machine interpretations
    • Support fluid, multi-layered representations capturing the nuances of situations
    • Ensure design strategies accommodate the dynamic and layered nature of human contexts