Progressive disclosure: empirically motivated approaches to designing effective transparency
summary
Two user studies test how transparency actually affects a simple emotion-prediction tool: a unigram linear regression model (trained on EmotiCal text) with MAE 0.95 on a −3 to +3 mood scale. Seventy‑four MTurk participants watch videos then use both interfaces and end up evenly split (37 vs 37 split) as transparency proves distracting, violates simple user heuristics, and exposes low‑level errors; a lab followup (n=12) adds a “Why am I seeing this rating?” button and motivates staged progressive disclosure (natural language, top feature highlights, then explorable tweaks).
AI-written summary · about
chapters (4)
- 0:00Start
- 1:16Intelligibility Explanation Transparency
- 11:10Fortuitous Interaction
- 11:59Progressive Disclosure
related talks
24mDesigning for Transparency in Machine Learning
18mBelieve it or not: Designing a Human-AI Partnership for Mixed-Initiative Fact-Checking
41mIf We Want AI to be Interpretable, We Need to Measure Interpretability
39mThe Search for Transparency and Accountability in the Age of...
14mSensing Affect to Empower Students: Learner Perspectives on Affect-Sensitive Technology
10mI Think I Get Your Point, AI! The Illusion of Explanatory Depth in Explainable AI
39mKeynote 3: How did you know? Explaining Black Box Model Predictions in Python
37mMind-Reading with adaptive UIs