A more human-centered approach to AI
summary
Argues that AI must be designed around people not just metrics and shows concrete tools and fixes: People + AI Research (PAIR) publishes a six‑chapter guidebook and open source tools like the Language Interpretability Tool and Teachable Machine (built on tensorflow.js). Gives concrete examples — on‑device grocery label vision, flood‑forecasting models, QuickDraw bias that required data rebalancing, and Google Translate changed to return gender alternatives — and explains how labeling design, confidence displays and team diversity alter outcomes.
AI-written summary · about
chapters (7)
- 0:00Introduction to human-centered AI
- 4:15People + AI Research Group
- 7:26User needs and AI feasibility
- 12:40Data collection and evaluation
- 25:25Bias in training data and labeling
- 26:06Calibrating trust in AI systems
- 30:23Closing and call to action
related talks
38mDesigning Human-Centered AI Products
24mA Call for Human-Centered AI
39mThe Search for Transparency and Accountability in the Age of...
26mResponsible AI-What, Why, How and Future!
24mAI as a Teammate: Redefining Agile for Human–AI Collaboration
17mHuman-centered AI
54mEthical AI Development: Leveraging Human Centered Design for Inclusive and Trustworthy Solutions
49mThe Shape of AI