Challenges Applying Traditional UI/UX Principles to M.L.
summary
A real-time transaction classifier flags a supposedly impossible column value (50 where schema allows 1–10), which turns out to be an upstream data change and becomes a valuable signal; a price estimator using multivariate regression shows coefficients flip when new features are added, confusing non-analyst users; a sales-lead prioritizer demotes an entire sector so reps ignore the system and disobey recommendations. Practical rules: design for audit, don't prescribe persuade, limit surprises to content not behavior.
AI-written summary · about
chapters (9)
- 0:00Introduction to talk
- 1:13Traditional UI/UX design principles
- 8:18Mental model alignment challenges
- 11:05Surprise and delegation in ML
- 23:43User mental models in practice
- 26:12Psychology of uncertainty and risk
- 30:09Concrete design principles for ML
- 35:22Future of ML design practices
- 36:46Q&A session
related talks
37mBehind the AI Curtain - Designing for Machine Learning Products
56mDehumanizing Agents: Why Explainability is Crucial in the LLM Era
32mReliable Machine Learning
35mImproving Credibility of ML Solutions
42mMasterclass: The right path to ML
38mDo's and don'ts of big data analytics
30mBetter data quality with ML & visualization
40mDesign for Data