External
Even as AI organizes personal documents and assists creativity, the sources and reasons behind its results stay hidden from users.
Research Statement · v0.2
I want to research how people understand,
correct, and confidently direct AI.
The questions found in Dear Me and TonePilot lead toward Human-Centered AI / HCI research. Current preparation, plans, and completed work are clearly separated.
Why This Matters
Why this research is needed
Internal
People use AI results for convenience while feeling anxious that they can't fix mistakes or understand the process.
Philosophical
Technology should go beyond replacing people — it should help them understand and decide better.
Research Themes
Three research axes
| Axis | Core question | Current link | Future artifacts |
|---|---|---|---|
| Personal Information & Everyday AI | How can AI organize personal documents across formats while preserving sources, privacy, and the right to correct? | Dear Me | Document schema, prototype, user studies, privacy principles |
| Human-AI Co-Creation for Music | How can AI propose creative starting points while strengthening the user's taste, learning, and agency? | TonePilot | Patch recommendation experiments, preference feedback, explanation UI, real-device evaluation |
| Explainable Applied ML | How can predictions from sports, music, and fandom data become explanations people can understand? | Planned studies on baseball player growth and K-pop patterns | Public data, notebooks, model comparison, visualization, explainability analysis |
Planned Studies
Planned studies, honestly labeled
Research that hasn't started is labeled as planned. Papers, degrees, or performance numbers that don't exist are never implied.
How should errors and sources be shown while structuring diverse statement formats?
Initial method: sample set → field schema → extraction comparison → review UI user testing
Study planned · Data design pending
How do reasons and adjustment rights affect a beginner's trust and learning?
Initial method: patch conditions → with/without explanations → real-device playing → qualitative/quantitative evaluation
Study planned · Prototype required
Short-term stats alone can't explain growth potential and context.
Initial method: public records → feature definition → model comparison → explainable factor visualization
Research planned · Dataset selection pending
Understanding patterns without reducing hits and engagement to a simple formula.
Initial method: audio/metadata/public behavioral data → ethics review → interpretable analysis
Research planned · Scope refinement
Current Preparation
What I'm preparing now
ML / DL · Statistics
Studying machine learning, deep learning, and statistics, practicing with public-data notebooks.
HCI · User research
Studying HCI concepts and user research methods, applying them to both projects' user flows.
Papers · Graduate prep
Reading Human-Centered AI/HCI papers and preparing English and graduate school applications.
Research Artifacts
Notebooks, data documents, model cards, research notes, and posters/reports will be published here and on Writing as they're ready.
Ethics & Agency
Ethics and user autonomy principles
Personal data
Research involving personal data includes anonymization, data boundaries, and deletion rights from the design stage.
Copyright
Music and content data are used only after reviewing copyright and permitted scope.
User autonomy
Automation must not remove choice — correction, undo, and final decisions stay with people.
Error disclosure
Failures, limits, and uncertainty of models and products are recorded, not hidden.