State Your Intention to Steer Your Attention: An AI Assistant for Intentional Digital Living

Paper Title

State Your Intention to Steer Your Attention: An AI Assistant for Intentional Digital Living

Publication Info

• Topic area: AI-driven tools for managing digital distractions and enhancing intentional digital activity.
• Keywords: AI assistant, digital well-being, distraction management, large language models, intentional activity, productivity tools, context-aware systems, user focus, digital self-control, human-computer interaction.

Background and Problem

• Problem / challenge: Existing productivity tools rely on static, rule-based methods that fail to adapt to user-specific contexts, leading to frustration and inefficiency. These tools cannot differentiate between task-related and distracting uses of the same application.
• Significance: Digital distractions reduce productivity, increase stress, and widen the intention-behavior gap. A more context-aware and adaptive solution could significantly improve digital well-being.
• Motivation and related work: Prior work on Digital Self-Control Tools (DSCTs) and AI-based systems has explored blocking apps, time tracking, and chat-based interfaces. However, these approaches lack real-time context awareness and fail to provide personalized, adaptive interventions. This paper addresses these gaps by introducing a system that dynamically aligns user activities with their stated intentions.

Solution

• Proposed approach: Intent Assistant (INA), an AI system that uses large language models (LLMs) to clarify user intentions, monitor on-screen activities, detect distractions, and provide timely, context-aware nudges.
• Novelty:
  1. Development of a context-aware assistant that adapts to user-specific digital behaviors.
  2. Introduction of IntentionBench, a dataset for evaluating distraction detection in realistic user workflows.
  3. Deployment and evaluation of INA in a three-week field study, demonstrating its effectiveness in reducing distractions and improving focus.
• Procedure and key techniques:
  1. Users input their intentions, which are clarified through an LLM-driven Q&A process.
  2. INA monitors on-screen activities, analyzing screenshots, application metadata, and URLs to detect distractions using semantic alignment scores.
  3. When distractions are detected, INA delivers polite, dismissible notifications to nudge users back on task.
  4. User feedback on notifications is incorporated to refine distraction detection accuracy over time.

Results

• Concrete findings:
  • INA achieved a distraction detection accuracy of 0.878 and an F1-score of 0.845 on the IntentionBench dataset.
  • In a real-world dataset, INA achieved an accuracy of 0.899 and a balanced accuracy of 0.815.
  • During a three-week field study with 22 participants, INA reduced the off-task ratio (0.104 vs. 0.166 for a simple reminder, p <.001) and increased intention alignment ratings (4.44 vs. 4.23, p <.001).
• Advantage over baselines:
  • INA outperformed a simple reminder system and a logging-only system in reducing distractions and improving user focus.
  • Participants reported higher focused immersion scores with INA (3.74 vs. 3.34 for simple reminder, p =.045; vs. 2.90 for logging only, p =.0003).
• Experiments / evaluation:
  • IntentionBench was used to evaluate distraction detection under controlled conditions.
  • A three-week, within-subjects field study compared INA to two baseline systems, collecting both quantitative metrics (e.g., off-task ratio, intention alignment) and qualitative feedback through surveys and interviews.
• Limitations and future work:
  • INA's notifications were sometimes perceived as intrusive, and the Q&A process was seen as repetitive.
  • Privacy concerns were raised due to the use of screenshots and metadata.
  • Future work could focus on adaptive notification policies, on-device LLMs for enhanced privacy, and long-term evaluations of INA's impact on digital habits.

Summary

This paper introduces the Intent Assistant (INA), an AI-powered system designed to help users align their digital activities with their stated intentions. By leveraging LLMs for real-time context analysis and distraction detection, INA provides timely, personalized nudges to reduce off-task behavior. Evaluations on the IntentionBench dataset and a three-week field study demonstrate INA's effectiveness in improving focus and intentional activity compared to baseline systems. While INA shows promise in fostering intentional digital living, addressing concerns about notification intrusiveness and data privacy will be critical for its broader adoption.