Synthetic inputs from different languages and contexts
Input, label, uncertainty, correction, and system response
Emotional experienceSocial experienceThinking experience
COLLEGE APPLICATION RESEARCH EVIDENCE · 고등
감정 자료를 나누는 AI의 편향과 한계 확인감정 자료를 종류별로 나누는 AI 편향·한계 감사
Subgroup checks, abstention, and user correction
The course was designed from open research published by Carnegie Mellon and Stanford. Students first explain how university researchers study whether technology is useful, easy, fair, and safe. They then distinguish their own question, design choice, A/B result, failure, revision, and limit for an authentic college application or interview explanation.
Research questionStudent A/B testEvidence and limitsOwn contribution
Average performance hides failures in specific contexts and disagreement between people.
Target experienceSubgroup checks, abstention, and user correction
Turn a need into a clear system promise
- Name the user and context, then map the full experience.
- Choose one design variable and a clear test rule.
- Let the user change settings or turn off an unwanted feature.
One variable, two versions
- Prototype A: show the information and action directly.
- Prototype B: test a different sense, order, or feedback style.
Check it with the same real task
Give both versions the same safe task. Compare success, time or errors, and a short user rating.
Use numbers and user comments
- At least one action measure
- Anonymous design or experience rating
- Accessibility check and an open comment
AB
The chart values are synthetic examples for this course page, not student results.
Use the evidence to improve once
Change one feature supported by both the numbers and user comments. Record the before-and-after result and one limit.
A good score is not the whole experience
Do not guess or manipulate a person's inner state. Use minimal data, clear choices, and an opt-out.
The user keeps final control.
clarifycorrectoverrideopt-outdelete
When the model lacks context or confidence, it should pause and ask the user.
Experience map, requirements, Prototype A/B, test chart, and before-and-after case study
Python/JupyterSynthetic data
Limit: A small class pilot is early evidence. Record how people and contexts differ, then name the next question to test.
Course consultation
Turn a student's interest
into an experience project.
We can match the grade, coding experience, and topic to paper, Figma, Scratch, web, Python, or robotics.