Key Idea
What if advisor could see your screen?
Motivations
We want to live according to our Intentions, but too many distractions pull our attention away.
Problem
Same app, completely different context.
Existing tools can't tell the difference. Just block YouTube entirely.
We need context-aware support
AI now understands what you're doing
The user is watching "What is HCI", an academic lecture video on YouTube.
The user is browsing YouTube Shorts, viewing cute animal videos.
What if we built an assistant with this context-aware understanding?
INtent Assistant
An AI Assistant for Intentional Digital Living
1
The user declares their intention.
System Design
Demo
Intention Bench
Evaluation of Distraction Detection
How accurately can INA detect off-task behavior?
To find out, we built IntentionBench , a benchmark dataset that combines focused sessions into realistic mixed sessions containing both on-task and off-task activity. We then evaluate how well INA detects transitions between the two.
INA achieved high accuracy and F1 score.
Based on these results, we moved on to a real-world user study.
Clarification and feedback jointly improve distraction detection.
| Clarification | Feedback | Accuracy | Precision | Recall | F1 |
|---|---|---|---|---|---|
| – | – | 0.805 | 0.897 | 0.629 | 0.739 |
| ✓ | – | 0.871 | 0.949 | 0.748 | 0.836 |
| – | ✓ | 0.845 | 0.959 | 0.677 | 0.794 |
| ✓ | ✓ | 0.878 | 0.959 | 0.755 | 0.845 |
The check mark (✓) indicates whether clarification or feedback is included; the best performance is in bold. Employing both features shows the best balance. The bottom row is our deployed configuration.
User Study
We conducted a three-week user study with 22 users, comparing three systems during everyday computer use.
Baseline Systems and INA
* To minimize bias, all systems were presented to participants using color-based names: Orange, Blue, and Purple.
Study Procedure
Week 1
Week 2
Week 3
Experiment
Orientation
Post-survey
Post-interview
Post-survey
Post-interview
Post-survey
Post-interview
Participants completed the same post-survey after each weekly condition. Closed-ended items used a 1–5 Likert scale.
01
User experience
- All applications
The program understood my intentions.
- All applications
The program helped me act on the activities I intended to do.
- All applications
I felt supported by the program.
- All applications
The program made me feel capable of controlling my digital activities.
- All applications
The program used my data in a meaningful way.
- All applications
I worried my personal information might leak.
- All applications
The program gave me new insights into my computer use.
- All applications
Using this program interfered with my workflow. Reverse-scored
- INA + Simple Reminder
Notifications helped me act according to my intentions.
- INA + Simple Reminder
Notifications motivated intended activities.
- INA + Simple Reminder
Notifications arrived at appropriate times.
- INA + Simple Reminder
The program tailored its messages to my intentions and activities.
- INA + Simple Reminder
The program provided clear explanations for its message.
- INA + Simple Reminder
Notifications felt burdensome. Reverse-scored
- INA only
The program adapted its behaviour based on my feedback.
02
Focused Immersion
- All applications
While using the computer I am able to block out most other distractions.
- All applications
While using the computer, I am absorbed in what I am doing.
- All applications
While on the computer, I am immersed in the task I am performing.
- All applications
When on the computer, I get distracted by other attentions very easily. Reverse-scored
- All applications
While on the computer, my attention does not get diverted very easily.
03
Open-ended survey
- All applications
Was there a moment when the program helped you focus or carry out tasks as intended? If so, please explain.
- All applications
Has the program helped you focus or carry out tasks as intended? If so, please describe with examples, including the context and how the program helped.
- All applications
What did you like or dislike about the program?
- All applications
What insights, if any, have you gained about your digital habits or computer use?
Results
Q1
Do users carry out their intended tasks with greater focus when using INA?
Study
How well did your activity align with your intention?
(self-reported score collected at the end of each session)
Q2
What is the overall user experience with INA?
Three-week within-subjects study with 22 participants. INA reduced LLM-estimated off-task ratio from 0.166 to 0.104 compared with simple reminder (p < .001), and increased intention alignment from 4.23 to 4.44 (p < .001).
Focused immersion was 3.74 for INA, 3.34 for simple reminder, and 2.90 for logging only. The study logged 1,786 sessions, 1,360 hours of usage, and 2,449,690 context records.
The weekly survey reports category-level scores. INA scored 3.75 on Support, compared with 3.56 for simple reminder and 2.75 for logging only. Message Effectiveness was 3.74 for INA and 3.24 for simple reminder.
Workflow Disruption was lower for INA (3.09) than both baselines (3.91), indicating that the more active assistant introduced interruptions that future designs should tune.
Discussion and Future Work
Ego-INA: Toward Ambient Support
INA can evolve into an ambient assistant that supports intentional living implicitly across devices, including smart glasses, without requiring explicit intent input.
Risks of unsafe intentions
INA may reinforce , as the LLM predicts intent rather than executing it, bypassing standard refusal mechanisms. Integrating external models can mitigate this risk.
Resources
Citation
@inproceedings{choi2026state,
title = {State Your Intention to Steer Your Attention: An AI Assistant for Intentional Digital Living},
author = {Choi, Juheon and Lee, Juyong and Kim, Jian and Kim, ChanYoung and Min, Taywon and Knox, W. Bradley and Lee, Min Kyung and Lee, Kimin},
booktitle = {Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems},
pages = {1--29},
year = {2026}
}