Responsible AI In Practice

on YouTube ↗38m

" summary

Open-source tooling (Fairlearn, InterpretML, ErrorAnalysis) plus a web Responsible AI Dashboard and a new RI mitigations library are usable end-to-end and integrated into Azure ML. Demo inspects an employee churn model with 21% overall error, a cohort at 21% versus 8% elsewhere and another cohort spiking to 58%. Error analysis, feature importance and counterfactuals expose causes; targeted cohort rebalancing raises overall performance by ~6% while avoiding precision drop that a blanket rebalance introduces.

AI-written summary · about

" chapters (6)

1. 0:00Introduction
2. 1:06Why Responsible AI?
3. 3:03Introduction to our Responsible AI Principles
4. 6:14Introduction to Open Source Tools
5. 12:29Demo
6. 36:06Q&A

" related talks

• 34mImproving the Fairness and Reliability of AI SolutionsRuth Yakubu · Global AI Developer Day 2022
• 48mResponsible AIValerie Hyde · Global Summer 2020 Azure Data Fest 2020
• 39mResponsible AISuresh Paulraj · Data Con LA 2020
• 31mOperationalizing Responsible AIRachel Thomas, Guillaume Saint-Jacques, Parinaz Sobhani, Khari Johnson · TWIMLcon: AI Platforms 2019
• 32mOperationalizing Responsible AISarah Bird, Kathy Baxter, Romer Rosales · TWIMLcon: AI Platforms 2021
• 19mA developer's guide to responsible AI review processesJen Gennai, Madeleine Elish · Google I/O 2022
• 24mDeveloping Innovative Systems Through Responsible AI In HealthcarePablo Casaubon, Dr. Jutsin Green, Philip Jarratt, Caio Moreno · Big Data LDN 2022
• 34mResponsible AI: The Key to Ethical, Safe and Inclusive Software DevelopmentLisa Falco · Kubernetes Community Days Zürich 2023