Social Implications of Bias in Machine Learning

on YouTube ↗38m

" summary

Shows concrete failures and fixes: city pothole app undercounts poor neighborhoods until buses and garbage trucks add data; Google Translate learns doctors = men / nurses = women from historic corpora and later adds gender-neutral outputs; facial recognition from major vendors flunks women of color (Joy Buolamwini / Algorithmic Justice League) and ProPublica finds the COMPAS bail score harms Black defendants; recommends tools like LIME, Google Facets / What‑If and IBM AI Fairness 360 and notes OpenAI withheld a powerful model to avoid misuse.

AI-written summary · about

" chapters (46)

1. 0:00Introduction
2. 0:36Social Bias
3. 0:58Prejudice
4. 1:17Statistical Bias
5. 1:51The Human Condition
6. 2:18What is Machine Learning
7. 6:23Overview
8. 9:51Garbage in Garbage out
9. 10:06Open Data Institute Canvas
10. 11:09Summary
11. 11:52Orchestras
12. 13:25Protected Attributes
13. 14:59Reducing Data
14. 15:17Adding Missing Data
15. 15:44Raising Machines
16. 16:38Google Translate
17. 18:45algorithmic justice league
18. 19:26the stem
19. 19:47diversity
20. 20:19The Forgotten
21. 20:58People Plus AI Research
22. 21:43Google
23. 22:06The Black Box
24. 22:22The Right to an Explanation
25. 22:36Open the Black Box
26. 23:19Simplifying Algorithms
27. 23:31Fixed Bugs
28. 24:10Lime
29. 24:58Tree Frog
30. 25:55What If
31. 27:02Feedback loops
32. 27:27Facebook
33. 28:56Compass
34. 30:07Silver Lining
35. 30:44Governance Usage
36. 31:00Amazon
37. 31:40Whos responsible
38. 32:39Pharmaceutical example
39. 33:44OpenAI example
40. 34:56We can quickly patch
41. 35:41We can demand better products
42. 36:06We can create updates very quickly
43. 36:30We have the power
44. 37:12We want to go
45. 37:34We have an immense power
46. 37:55Change and shape society

" related talks

• 38mSocial Implications of Bias in Machine LearningFiona Coath · Øredev 2019
• 37mKeynote: Ethical Machine Learning: Creating Fair Models in an Unjust WorldKatharine Jarmul · PyData Amsterdam 2017
• 47mEthical Machine LearningKatharine Jarmul · QCon São Paulo 2018
• 27mCrouching Harm, Hidden Bias—Addressing issues of fairness....Ranjan Bhattacharya · Global Artificial Intelligence Virtual Conference 2022
• 47mEthics & AI: Identifying & preventing bias in predictive modelsFederica Pelzel · Code Mesh LDN 2018
• 51mDigital Discrimination: Cognitive Bias in Machine LearningSaishruthi Swaminathan · All Things Open 2020
• 31mAvoiding Digital BiasAdam L Smith · Full Stack Fest 2018
• 21mFairness and bias in AIStephen Thomas · Big Data & AI Toronto 2020