Understanding AI with IBM Developer - Digital Discrimination: Cognitive Bias in Machine Learning

Slide 1

Slide 1 text

IBM Developer UK: Understanding AI Mini Conf Digital Discrimination: Cognitive Bias in Machine Learning 21 April, 2020

Slide 2

Slide 2 text

Digital Discrimination: Cognitive Bias in Machine Learning Tweet at me! @Mo_Mack 2

Slide 3

Slide 3 text

A cognitive bias is a systematic pattern of deviation from norm or rationality in judgment. People make decisions given their limited resources. Wilke A. and Mata R. (2012) “Cognitive Bias”, Clarkson University 3 @Mo_Mack

Slide 4

Slide 4 text

Example of bias in machine learning. 4 @Mo_Mack

Slide 5

Slide 5 text

NorthPointe’s COMPAS Algorithm 5 Image Credit: #WOCinTech @Mo_Mack

Slide 6

Slide 6 text

Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing May 2016 - Northpointe’s COMPAS Algorithm equivant.com/solutions/inmate-classiﬁcation 6

Slide 7

Slide 7 text

May 2016 - Northpointe’s COMPAS Algorithm equivant.com/solutions/inmate-classiﬁcation Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing

Slide 8

Slide 8 text

May 2016 - Northpointe’s COMPAS Algorithm equivant.com/solutions/inmate-classiﬁcation Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing

Slide 9

Slide 9 text

May 2016 - Northpointe’s COMPAS Algorithm equivant.com/solutions/inmate-classiﬁcation Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing

Slide 10

Slide 10 text

Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing 10 Black Defendant’s Risk Scores

Slide 11

Slide 11 text

Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing 11 White Defendant’s Risk Scores

Slide 12

Slide 12 text

BLACK VS. WHITE DEFENDANTS ○ Falsely labeled black defendants as likely of future crime at twice the rate as white defendants. ○ White defendants mislabeled as low risk more than black defendants ○ Pegged Black defendants 77% more likely to be at risk of committing future violent crime 12 @Mo_Mack

Slide 13

Slide 13 text

Slide 14

Slide 14 text

Gender Shades Project February 2018 14 Image Credit: #WOCinTech @Mo_Mack

Slide 15

Slide 15 text

gendershades.org

Slide 16

Slide 16 text

“If we fail to make ethical and inclusive artiﬁcial intelligence we risk losing gains made in civil rights and gender equity under the guise of machine neutrality.” 16 - Joy Boulamwini @jovialjoy

Slide 17

Slide 17 text

17 http://www.aies-conference.com/wp-content/uploads/2019/01/AIES-19_paper_223.pdf

Slide 18

Slide 18 text

18 youtube.com/watch?v=Af2VmR-iGkY @Mo_Mack

Slide 19

Slide 19 text

Solutions? What can we do to combat bias in AI? 19 @Mo_Mack

Slide 20

Slide 20 text

20 vox.com/ezra-klein-show-podcast

Slide 21

Slide 21 text

“Coders are the most empowered laborers that have ever existed.” 21 - Anil Dash @anildash

Slide 22

Slide 22 text

FIX THE PIPELINE? 22 Image Credit: #WOCinTech @Mo_Mack

Slide 23

Slide 23 text

“Cognitive bias in machine learning is human bias on steroids.” 23 - Rediet Abebe @red_abebe @Mo_Mack

Slide 24

Slide 24 text

24 twitter.com/MatthewBParksSr/status/1133435312921874432

Slide 25

Slide 25 text

January 2019 - New Search Feature on... www.pinterest.com Source: /www.engadget.com/2019/01/24/pinterest-skin-tone-search-diversity/

Slide 26

Slide 26 text

“By combining the latest in machine learning and inclusive product development, we're able to directly respond to Pinner feedback and build a more useful product.” 26 - Candice Morgan @Candice_MMorgan @Mo_Mack

Slide 27

Slide 27 text

National and Industry Standards 27 Image Credit: #WOCinTech @Mo_Mack

Slide 28

Slide 28 text

EU Ethics Guidelines for Trustworthy Artiﬁcial Intelligence According to the Guidelines, trustworthy AI should be: (1) lawful - respecting all applicable laws and regulations (2) ethical - respecting ethical principles and values (3) robust - both from a technical perspective while taking into account its social environment Source: ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai

Slide 29

Slide 29 text

#1 - Human agency and oversight. #2 - Technical robustness and safety. #3 - Privacy and data governance. #4 - Transparency . 29 @Mo_Mack #5 - Diversity, non- discrimination and fairness. #6 - Societal and environmental well-being. #7 - Accountability

Slide 30

Slide 30 text

30 bit.ly/lfai-trust

Slide 31

Slide 31 text

OPEN SOURCE TOOLS 31 Image Credit: #WOCinTech @Mo_Mack

Slide 32

Slide 32 text

Tool #1: AI Fairness 360 Toolkit Open Source Library 32 @Mo_Mack

Slide 33

Slide 33 text

aif360.mybluemix.net @Mo_Mack

Slide 34

Slide 34 text

aif360.mybluemix.net @Mo_Mack

Slide 35

Slide 35 text

Machine Learning Pipeline In- Processing Pre- Processing Post- Processing 35 Modifying the training data. Modifying the learning algorithm. Modifying the predictions (or outcomes.) @Mo_Mack

Slide 36

Slide 36 text

aif360.mybluemix.net Demos @Mo_Mack

Slide 37

Slide 37 text

github.com/IBM/AIF360 AI Fairness 360 Toolkit Public Repo 37 @Mo_Mack

Slide 38

Slide 38 text

Tool #2: AI Explainability 360 Toolkit Open Source Library 38 @Mo_Mack

Slide 39

Slide 39 text

aix360.mybluemix.net @Mo_Mack

Slide 40

Slide 40 text

aix360.mybluemix.net Demos @Mo_Mack

Slide 41

Slide 41 text

github.com/IBM/AIX360 AI Explainability 360 Toolkit Public Repo 41 @Mo_Mack

Slide 42

Slide 42 text

Tool #3: Model Asset eXchange Open Source Pre-Trained Deep Learning Models 42 @Mo_Mack

Slide 43

Slide 43 text

ibm.biz/model-exchange 43 @Mo_Mack

Slide 44

Slide 44 text

ibm.biz/model-exchange Model Asset eXchange (MAX) 44 @Mo_Mack

Slide 45

Slide 45 text

Tool #4: Data Asset eXchange Open Source Data Sets 45 @Mo_Mack

Slide 46

Slide 46 text

/ibm.biz/data-exchange Data Asset eXchange (DAX) 46 @Mo_Mack

Slide 47

Slide 47 text

ibm.biz/codait-trusted-ai IBM CODAIT Trusted AI Work 47 @Mo_Mack

Slide 48

Slide 48 text

Take Control of the Machine Learning Pipeline

Slide 49

Slide 49 text

49 IBM’s AI Portfolio Everything you need for Enterprise AI, on any cloud Watson Studio Watson Machine Learning Watson AI OpenScale Build Deploy Manage Interact with Pre-built AI Services Watson Application Services Unify on a Multicloud Data Platform IBM Cloud Private for Data AI Open Source Frameworks

Slide 50

Slide 50 text

Slide 51

Slide 51 text

51 Photo by rawpixel on Unsplash No matter what it is our responsibility to build systems that are fair.

Slide 52

Slide 52 text

Thank you! Any questions for me? @Mo_Mack Find my team: @ibmcodait Trusted AI Projects: ibm.biz/codait-trusted-ai 52