Designing for Transparency in Machine Learning
summary
Compares real failures and fixes: biased Google image results and a secret Palantir–New Orleans deployment trained on biased arrest data; an art test that produces 30 algorithmic portraits of Chelsea Manning that don’t resemble her to show model error; an open dataset product Common Voice: 1,400+ hours, 18 languages and a project that builds a feminist dataset with a living‑wage task calculator; proposes three practical design principles — legibility, auditability and spaces for user interaction — and sketches an editable Spotify Discover Weekly UI.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:52About Caroline
- 2:00What is AI
- 3:14Salt AI
- 4:26Data
- 5:17Library of Missing Data Sets
- 6:35Probably Chelsea
- 8:00Google Search
- 13:16Common Voice
- 14:06Feminist Data Set
- 17:00Human Rights Centered Design
- 19:42Privacy on Design
related talks
47mCreating solutions for everyone
26mBuilding ethical AI: Transparency, bias mitigation, and trust
41mDesign for Cognitive Bias
26mConsequences of an Insightful Algorithm
12mHow we infuse our biases in the products...
33mBehind the AI Curtain - Designing for Trust in Machine Learning Products
24mHey you, do you even design ?
11mMachines of Love & Grace – Building Truly Inclusive AI