fairness in machine learning - tat's revolution · fairness in machine learning fernanda...
TRANSCRIPT
![Page 1: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/1.jpg)
Fairness in Machine LearningFernanda Viégas @viegasfMartin Wattenberg @wattenbergGoogle Brain
![Page 2: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/2.jpg)
As AI touches high-stakes aspects of everyday life, fairness becomes more important
![Page 3: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/3.jpg)
How can an algorithm even be unfair?Aren't algorithms beautiful neutral pieces of mathematics?
The Euclidean algorithm (first discovered in 300 BCE) as described in Geometry, plane, solid and spherical, Pierce Morton, 1847.
![Page 4: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/4.jpg)
"Classic" non-ML problem: implicit cultural assumptions
![Page 5: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/5.jpg)
Example: names are complex.
Brilliant, fun article.Read it! :)
Patrick McKenziehttp://www.kalzumeus.com/2010/06/17/falsehoods-programmers-believe-about-names/
...
![Page 6: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/6.jpg)
What's different with machine learning?
Algorithm, 300 BCE
Classical algorithms don’t rely on data
![Page 7: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/7.jpg)
What's different with machine learning?
ML systems rely on real-world data andcan pick up biases from data
Algorithm, 300 BCE
Algorithm, 2017 CE
Classical algorithms don’t rely on data
![Page 8: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/8.jpg)
Sometimes bias starts before an algorithm ever runs…It can start with the data
![Page 9: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/9.jpg)
Sometimes bias starts before an algorithm ever runs…It can start with the data
A real-world example
![Page 10: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/10.jpg)
![Page 11: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/11.jpg)
Can you spot the bias?
![Page 12: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/12.jpg)
Can you spot the bias?
![Page 13: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/13.jpg)
Model can’t recognize mugs with handle facing left
![Page 14: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/14.jpg)
How can this lead to unfairness?
![Page 15: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/15.jpg)
![Page 16: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/16.jpg)
word embeddings
![Page 17: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/17.jpg)
Word embeddings
Distributed Representations of Words and Phrases and their Compositionality
Mikolov et al. 2013
![Page 18: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/18.jpg)
Meaningful directions(word2vec)
![Page 19: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/19.jpg)
![Page 20: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/20.jpg)
![Page 21: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/21.jpg)
![Page 22: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/22.jpg)
Can we "de-bias" embeddings?
![Page 23: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/23.jpg)
Can we "de-bias" embeddings?
Bolukbasi et al.: this may be possible.
Idea: "collapse" dimensions corresponding to key attributes, such as gender.
![Page 24: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/24.jpg)
How can we build systems that are fair?
First, we need to decide what we mean by “fair”...
![Page 25: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/25.jpg)
Interesting fact: You can't always get what you want in terms of “fairness”!
![Page 26: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/26.jpg)
Fairness: you can't always get what you want!
https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
COMPAS (from company called Northpointe)● Estimates chances a defendant will be re-arrested
○ Issue: "rearrest" != "committed crime"● Meant to be used for bail decisions
○ Issue: also used for sentencing
![Page 27: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/27.jpg)
https://www.propublica.org/article/how-we-analyzed-the-compas-recidivism-algorithm
This conclusion came from applying COMPAS to historical arrest records.
![Page 28: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/28.jpg)
Enter the computer scientists...
![Page 29: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/29.jpg)
![Page 30: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/30.jpg)
![Page 31: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/31.jpg)
Low-riskFAIR
![Page 32: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/32.jpg)
Low-riskFAIR
Medium-highFAIR
![Page 33: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/33.jpg)
Low-riskFAIR
Medium-highFAIR
![Page 34: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/34.jpg)
Low-riskFAIR
Medium-highFAIR
Did not reoffendUNFAIR
Black defendants who did not reoffend were more often labeled "high risk"
![Page 35: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/35.jpg)
Low-riskFAIR
Medium-highFAIR
Did not reoffendUNFAIR
Unless classifier is perfect, can't all be fair due to different base rates
![Page 36: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/36.jpg)
![Page 37: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/37.jpg)
https://www.propublica.org/article/bias-in-criminal-risk-scores-is-mathematically-inevitable-researchers-say
![Page 38: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/38.jpg)
Lessons learned
● Fairness of an algorithm depends in part on how it's used ● In fairness, you can't always get (everything) you want
○ Must make a careful choice of quantitative metrics○ This involves case-by-case policy decisions○ These tradeoffs affect human decisions too!
● Improvements to fairness may come with their own costs to other values we want to retain (privacy, performance, etc.)
![Page 39: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/39.jpg)
Can computer scientists do anything besides depress us?
Hardt, Price, Srebro (2016)On forcing a threshold classifier to be "fair" by various definitions:
● Group-unawareSame threshold for each group
● Demographic ParitySame proportion of positive classifications
● "Equal opportunity"Same proportion of true positives
![Page 40: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/40.jpg)
Can computer scientists do anything besides depress us?
A visual explanation...
![Page 41: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/41.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
![Page 42: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/42.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
![Page 43: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/43.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
Score: 23 Score: 79
![Page 44: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/44.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
![Page 45: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/45.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
![Page 46: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/46.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
![Page 47: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/47.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
Lower score than threshold but would pay back loan
![Page 48: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/48.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
Would pay back but isn’t given a loan
![Page 49: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/49.jpg)
Attacking discrimination in ML mathematically
Would default on loan Would pay back loan
Credit Score 0 10 20 30 40 50 60 70 80 90 100
Would pay back but isn’t given a loan
Would default and is given a loan
![Page 50: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/50.jpg)
Attacking discrimination in ML mathematically
![Page 51: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/51.jpg)
Attacking discrimination in ML mathematically
Profit: 1.2800
![Page 52: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/52.jpg)
Attacking discrimination in ML mathematically
Profit: 1.2800
![Page 53: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/53.jpg)
Attacking discrimination in ML mathematically
Profit: 1.2800
![Page 54: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/54.jpg)
Multiple groups and multiple distributions
![Page 55: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/55.jpg)
Multiple groups and multiple distributions
![Page 57: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/57.jpg)
Case Study
![Page 58: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/58.jpg)
Conversation AI /Perspective API (Jigsaw / CAT / others)
Jigsaw / CAT / others
![Page 59: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/59.jpg)
Conversation AI /Perspective API
![Page 60: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/60.jpg)
Conversation AI /Perspective API
![Page 61: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/61.jpg)
False "toxic" positives
Comment Toxicity scoreThe Gay and Lesbian Film Festival starts today. 82%Being transgender is independent of sexual orientation. 52%A Muslim is someone who follows or practices Islam. 46%
![Page 62: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/62.jpg)
How did this happen?
![Page 63: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/63.jpg)
How did this happen?
![Page 64: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/64.jpg)
One possible fix
![Page 65: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/65.jpg)
False positives - some improvement
Comment Old NewThe Gay and Lesbian Film Festival starts today. 82% 1%Being transgender is independent of sexual orientation. 52% 5%A Muslim is someone who follows or practices Islam. 46% 13%
Overall AUC for old and new classifiers was very close.
![Page 66: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/66.jpg)
A common objection...
● Our algorithms are just mirrors of the world. Not our fault if they reflect bias!
![Page 67: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/67.jpg)
A common objection...
● Our algorithms are just mirrors of the world. Not our fault if they reflect bias!
Some replies:
● If the effect is unjust, why shouldn't we fix it?● Would you apply this same standard to raising a child?
![Page 68: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/68.jpg)
Another objection
● Objection: People are biased and opaque. ● Why should ML systems be any different?
○ True: this won't be easy○ We have a chance to do better with ML
![Page 69: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/69.jpg)
Another objection
● Objection: People are biased and opaque. ● Why should ML systems be any different?
○ True: this won't be easy○ We have a chance to do better with ML
![Page 70: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/70.jpg)
What can you do?
1. Include diverse perspectives in design and development
2. Train ML models on comprehensive data sets
3. Test products with diverse users
4. Periodically re-evaluate and be alert to errors
![Page 71: Fairness in Machine Learning - Tat's Revolution · Fairness in Machine Learning Fernanda Viégas @viegasf Martin Wattenberg @wattenberg Google Brain. As AI touches high-stakes aspects](https://reader033.vdocuments.mx/reader033/viewer/2022043019/5f3b9f4eee723e1b3d4f3da7/html5/thumbnails/71.jpg)
Fairness in Machine LearningFernanda Viégas @viegasfMartin Wattenberg @wattenbergGoogle Brain