Red Hat DevNation - Digital Discrimination: Cognitive Bias in Machine Learning

Tools/Communities:
Enter IBM's Call for Code Competition: https://ibm.biz/BdzPJn
AI Fairness 360 Toolkit: http://aif360.mybluemix.net/
Model Asset Exchange: http://ibm.biz/model-exchange
IBM's Data Science and AI Elite Team: http://community.ibm.com/DSE
Watson Studio: https://www.ibm.com/cloud/watson-studio
Watson Machine Learning: https://www.ibm.com/cloud/machine-learning
Watson OpenScale: https://www.ibm.com/cloud/watson-openscale/

Talk Sources:

Cognitive Bias Definition:
https://adweb.clarkson.edu/~awilke/Research_files/EoHB_Wilke_12.pdf

Podcasts/Tweets
https://leanin.org/podcast-episodes/siri-is-artificial-intelligence-biased
https://art19.com/shows/the-ezra-klein-show/episodes/663fd0b7-ee60-4e3e-b2cb-4fcb4040eef1
https://twitter.com/alexisohanian/status/1087973027055316994
https://twitter.com/MatthewBParksSr/status/1133435312921874432

Northpointe/Equivant's COMPAS - Recidivism Score Algorithm
https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
https://www.technologyreview.com/s/612775/algorithms-criminal-justice-ai/

Data for Black Lives
http://d4bl.org/about.html
2019 Conference Notes: https://docs.google.com/document/d/1E1mfgTp73QFRmNBunl8cIpyUmDos28rekidux0voTsg/edit?ts=5c39f92e

Gender Shades Project
http://news.mit.edu/2018/study-finds-gender-skin-type-bias-artificial-intelligence-systems-0212
MIT Media Lab Overview for the project: https://www.youtube.com/watch?time_continue=1&v=TWWsW1w-BVo
FAT* 2018 Talk about outcomes: https://www.youtube.com/watch?v=Af2VmR-iGkY
https://www.ajlunited.org/fight

Other articles referenced in this talk:
https://www.nytimes.com/2018/02/12/business/computer-science-ethics-courses.html
https://www.vox.com/science-and-health/2017/4/17/15322378/how-artificial-intelligence-learns-how-to-be-racist
https://www.engadget.com/2019/01/24/pinterest-skin-tone-search-diversity/

Transcript

1. Red Hat DevNation
   Digital Discrimination:
   Cognitive Bias in Machine
   Learning
   June 27, 2019
   

   View Slide

2. Digital Discrimination:
   Cognitive Bias in Machine
   Learning
   My name is:
   Maureen McElaney
   Tweet at me! @Mo_Mack
   2
   @Mo_Mack
   

   View Slide

3. 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
   

   View Slide

4. 4
   @Mo_Mack
   

   View Slide

5. Example of bias in machine
   learning.
   5
   @Mo_Mack
   

   View Slide

6. NorthPointe’s
   COMPAS
   Algorithm
   6 Image Credit: #WOCinTech
   

   View Slide

7. Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
   May 2016 - Northpointe’s COMPAS Algorithm
   http://www.equivant.com/solutions/inmate-
   classification
   7
   @Mo_Mack
   

   View Slide

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

   View Slide

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

   View Slide

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

    View Slide

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

    View Slide

12. Source: https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
    12
    White Defendant’s Risk Scores
    @Mo_Mack
    

    View Slide

13. 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
    13
    @Mo_Mack
    

    View Slide

14. 14
    @Mo_Mack
    

    View Slide

15. Gender
    Shades Project
    15 Image Credit: #WOCinTech
    

    View Slide

16. Joy Buolamwini,
    Algorithmic Justice League
    Gender Shades Project
    Released February 2018
    16
    @Mo_Mack
    

    View Slide

17. View Slide

18. “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.”
    18
    - Joy Boulamwini
    @jovialjoy
    

    View Slide

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

    View Slide

20. 20
    https://www.vox.com/ezra-klein-show-podcast
    @Mo_Mack
    

    View Slide

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

    View Slide

22. EDUCATION IS
    KEY
    22 Image Credit: #WOCinTech
    

    View Slide

23. https://www.nytimes.com/2018/02/12/business/computer-science-
    ethics-courses.html
    @Mo_Mack
    

    View Slide

24. Questions
    posed to
    students
    in these
    courses...
    Is the
    technology
    fair?
    How do you
    make sure
    that the
    data is not
    biased?
    Should
    machines
    be judging
    humans?
    24
    @Mo_Mack
    

    View Slide

25. 25
    https://twitter.com/Neurosarda/status/1084198368526680064
    

    View Slide

26. FIX THE
    PIPELINE?
    26 Image Credit: #WOCinTech
    

    View Slide

27. “Cognitive bias in
    machine learning is
    human bias on
    steroids.”
    27
    - Rediet Abebe
    @red_abebe
    

    View Slide

28. 28
    https://twitter.com/MatthewBParksSr/status/1133435312921874432
    @Mo_Mack
    

    View Slide

29. TOOLS TO
    COMBAT BIAS
    29 Image Credit: #WOCinTech
    

    View Slide

30. Tool #1:
    AI Fairness
    360 Toolkit
    Open Source Library
    30
    @Mo_Mack
    

    View Slide

31. http://aif360.mybluemix.net/
    @Mo_Mack
    

    View Slide

32. http://aif360.mybluemix.net/
    @Mo_Mack
    

    View Slide

33. TYPES OF METRICS
    ○ Individual vs. Group Fairness, or Both
    ○ Group Fairness: Data vs Model
    ○ Group Fairness: We’re All Equal vs What
    You See is What You Get
    ○ Group Fairness: Ratios vs Differences
    33
    @Mo_Mack
    

    View Slide

34. @Mo_Mack
    

    View Slide

35. @Mo_Mack
    

    View Slide

36. Machine Learning
    Pipeline
    In-
    Processing
    Pre-
    Processing
    Post-
    Processing
    36
    Modifying the
    training data.
    Modifying the
    learning
    algorithm.
    Modifying the
    predictions (or
    outcomes.)
    @Mo_Mack
    

    View Slide

37. @Mo_Mack
    

    View Slide

38. http://aif360.mybluemix.net/
    Demos
    @Mo_Mack
    

    View Slide

39. https://github.com/IBM/AIF360
    AI Fairness 360 Toolkit Public Repo
    39
    @Mo_Mack
    

    View Slide

40. http://aif360.mybluemix.net/community
    AI Fairness 360 Toolkit Slack
    40
    @Mo_Mack
    

    View Slide

41. Tool #2:
    Model Asset
    Exchange
    Open Source Pre-Trained
    Deep Learning Models
    41
    @Mo_Mack
    

    View Slide

42. Step 1: Find a model
    ...that does what you need
    ...that is free to use
    ...that is performant enough
    42
    @Mo_Mack
    

    View Slide

43. Step 2: Get the code
    Is there a good implementation available?
    ...that does what you need
    ...that is free to use
    ...that is performant enough
    43
    @Mo_Mack
    

    View Slide

44. Step 3: Verify the
    model
    ○ Does it do what you need?
    ○ Is it free to use (license)?
    ○ Is it performant enough?
    ○ Accuracy?
    44
    @Mo_Mack
    

    View Slide

45. Step 4: Train the model
    45
    @Mo_Mack
    

    View Slide

46. Step 4: Train the model
    46
    @Mo_Mack
    

    View Slide

47. Step 5: Deploy your
    model
    ○ Adjust inference code (or write from
    scratch)
    ○ Package inference code, model code, and
    pre-trained weights together
    ○ Deploy your package
    47
    @Mo_Mack
    

    View Slide

48. Step 6: Consume your
    model
    48
    @Mo_Mack
    

    View Slide

49. Model Asset
    Exchange
    The Model Asset Exchange (MAX) is a one
    stop shop for developers/data scientists to
    find and use free and open source deep
    learning models
    ibm.biz/model-exchange
    49
    @Mo_Mack
    

    View Slide

50. ○ Wide variety of domains (text, audio,
    images, etc)
    ○ Multiple deep learning frameworks
    ○ Vetted and tested code/IP
    ○ Build and deploy a model web service in
    seconds
    50
    Model Asset
    Exchange
    @Mo_Mack
    

    View Slide

51. ibm.biz/model-exchange
    51
    @Mo_Mack
    

    View Slide

52. http://ibm.biz/model-exchange
    http://ibm.biz/max-slack
    Model Asset eXchange (MAX)
    52
    @Mo_Mack
    

    View Slide

53. Take Control
    of the Machine
    Learning
    Pipeline
    53
    @Mo_Mack
    

    View Slide

54. 54
    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
    @Mo_Mack
    

    View Slide

55. © 2018 IBM Corporation 30 April 2019 IBM Data Science Elite
    55
    Need help?
    IBM Data Science & AI Elite team.
    Find them: community.ibm.com/DSE
    @Mo_Mack
    

    View Slide

56. UPDATE TO
    THE GENDER
    SHADES
    PROJECT
    56 Image Credit: #WOCinTech
    

    View Slide

57. 57
    http://www.aies-conference.com/wp-content/uploads/2019/01/AIES-19_paper_223.pdf
    @Mo_Mack
    

    View Slide

58. 58
    FAT* 2018: Joy Buolamwini - Intersectional Accuracy Disparities in Commercial
    Gender Classification
    https://www.youtube.com/watch?v=Af2VmR-iGkY
    @Mo_Mack
    

    View Slide

59. 59
    Photo by rawpixel on Unsplash
    No matter what it is our
    responsibility to build
    systems that are fair.
    @Mo_Mack
    

    View Slide

60. 60
    @Mo_Mack
    

    View Slide

61. Enter the Call for Code
    https://ibm.biz/BdzPJn
    Slides
    http://bit.ly/redhat-biasinai
    Any questions?
    @Mo_Mack
    61
    

    View Slide