The Ethics of Algorithms: When Your Code Can Harm Others


The Presentation inside:

Slide 0

CONSEQUENCES OF AN INSIGHTFUL ALGORITHM CARINA C. ZONA @CCZONA


Slide 1

grieving surveillance PTSD racial profiling
 depression
 
 
 
 
 ENT W C
 O N T 
 G ARNIN 
 
 
 sexual history miscarriage consent infertility
 assault


Slide 2

ALGORITHMS IMPOSE CONSEQUENCES ON PEOPLE ALL THE TIME


Slide 3

ALGORITHM S T E P - B Y- S T E P S E T O F 
 O P E R AT I O N S F O R 
 P R E D I C TA B LY A R R I V I N G AT AN OUTCOME


Slide 4

ALGORITHMS OF COMPUTER SCIENCE & MATHEMATICS PAT T E R N S O F I N S T R U C T I O N S , A R T I C U L AT E D I N C O D E O R F O R M U L A S


Slide 5

ALGORITHMS IN ORDINARY LIFE PAT T E R N S O F I N S T R U C T I O N S , A R T I C U L AT E D I N W AY S S U C H A S …


Slide 6


Slide 7


Slide 8

Ch 12, sl st to join. Rnd 1: ch 3, 17 dc. Rnd 2: ch 3, 1 dc, ch 4, skip 1 dc, *2 dc, ch 4, skip 1 dc*, repeat 5 times, sl st. Rnd 3: ch 3, 2 dc, ch 5, skip 2 ch and 1 dc, *3 dc, ch 5, skip 2 ch and 1 dc*, repeat 5 times, sl st. Rnd 4: ch 3, 3 dc, ch 5, skip 3 ch and 1 dc, *4 dc, ch 5, skip 3 ch and 1 dc*, repeat 5 times, sl st. Rnd 5: ch 3, 4 dc, ch 6, skip 3 ch and 1 dc, *5 dc, ch 6, skip 3 ch and 1 dc*, repeat 5 times, sl st. Rnd 6: ch 3, 6 dc, ch 6, skip 3 ch and 1 dc, *7 dc, ch 6, skip 3 ch and 1 dc*, repeat 5 times, sl st. Rnd 7: ch 3, 8 dc, ch 6, skip 3 ch and 1 dc, *9 dc, ch 6, skip 3 ch and 1 dc*, repeat 5 times, sl st. Rnd 8: ch 3, 10 dc, ch 6, skip 3 ch and 1 dc, *11 dc, ch 6, skip 3 ch and 1 dc*, repeat 5 times, sl st .Rnd 9: ch 3, 12 dc, ch 6, skip 3 ch and 1 dc, *13 dc, ch 6, skip 3 ch and 1 dc*, repeat 5 times, sl st. Rnd 10: ch 3, 14 dc, ch 7, skip 3 ch and 1 dc, *15 dc, ch 7, skip 3 ch and 1 dc*, repeat 5 times, sl st. Rnd 11: ch 3, 16 dc, ch 7, skip 4 ch and 1 dc, *17 dc, ch 7, skip 4 ch and 1 dc*, repeat 5 times, sl st. Rnd 12: ch 3, 18 dc, ch 8, skip 4 ch and 1 dc, *19 dc, ch 8, skip 4 ch and 1 dc*, repeat 5 times, sl st. Rnd 13: ch 3, 20 dc, ch 8, skip 5 ch and 1 dc, *21 dc, ch 8, skip 5 ch and 1 dc*, repeat 5 times, sl st. Rnd 14: ch 3, 16 dc, ch 8, skip 3 dc and 2 ch, 1 dc, ch 8, skip 1 dc, *17 dc, ch 8, skip 3 dc and 2 ch, 1 dc, ch 8, skip 1 dc,* repeat 5 times, sl st.


Slide 9

DEEP LEARNING A L G O R I T H M S F O R F A S T, TRAINABLE, ARTIFICIAL NEURAL NETWORKS


Slide 10

"DEEP LEARNING IS A PA R T I C U L A R A P P R O A C H T O BUILDING & TRAINING ARTIFICIAL NEURAL NETWORKS. THINK OF THEM AS DECISIONMAKING BLACK BOXES." – P e t e Wa rd e n , " W h a t i s d e e p l e a r n i n g , a n d w h y s h o u l d y o u c a re ? " @CCZONA


Slide 11

INPUT Array of numbers representing 
 words, concepts, or objects. EXECUTION Run a series of functions 
 o n t h e a r r a y. OUTPUT Prediction of properties useful for drawing intuitions about similar inputs. @CCZONA


Slide 12

INPUT Array of 
 numbers representing 
 words, concepts, or objects EXECUTION Run a series of functions 
 on the array OUTPUT Prediction of 
 properties useful for 
 drawing intuitions about 
 similar inputs


Slide 13

Deep Learning relies on an artificial neural network's automated discovery of patterns within its training dataset. It applies those discoveries to draw intuitions about future inputs. @CCZONA


Slide 14


Slide 15


Slide 16

WOULD Y OU LIKE T O P L AY A N I C E G A M E O F D AT A M I N I N G F A I L ?


Slide 17

# D AT A M I N I N G F A I L Algorithmic Profiling DeAnonymization Disparate Impact Consent Issues Unproven Methods Inadvertent Algorithmic Cruelty Deception Filtering No Recourse Creepy Stalkery Messing With Heads Accurate, But Not Right Shaming Diversity Fail Black Box Personally Human Replicates Bias Identifiable Info Complexity Fail Uncritical Assumptions Moved Fast, Broke Things High Risk False Neutrality Consequences Invasion Of Privacy Not How Valid Data Insecurity Research Works


Slide 18

TA R G E T


Slide 19


Slide 20

S H U T T E R F LY


Slide 21


Slide 22

“ T h a n k s , S h u t t e r f l y, f o r the congratulations 
 on my 'new bundle of joy'. I'm horribly infertile,
 b u t h e y, 
 I'm adopting 
 a kitten, so...” @CCZONA


Slide 23

"I lost a baby in November who would have been due this week.
 It was like hitting a wall all over again." @CCZONA


Slide 24

“THE INTENT OF THE E M A I L W A S T O TA R G E T CUSTOMERS WHO H A V E R E C E N T LY HAD A BABY" @CCZONA


Slide 25

" Yo u s t a r t i m a g i n i n g w h o 
 they'll become & dreaming of h o p e s f o r t h e i r f u t u r e . Yo u s t a r t making plans, and then they're gone. It's a lonely e x p e r i e n c e ." @CCZONA


Slide 26

FACEBOOK
 E M O TI O N A L C O N TA I G O N Algorithmic Profiling DeAnonymization


Slide 27


Slide 28

FACEBOOK
 YEAR I N R EV I EW Algorithmic Profiling DeAnonymization Consent Issues Unproven Methods Inadvertent Algorithmic Cruelty Deception Filtering No Recourse Creepy Stalkery Messing With Heads Accurate, But Not Right Shaming Diversity Fail Black Box


Slide 29

I N A D V E R T E N T A L G O R I T H M I C C R U E LT Y Inadvertent algorithmic cruelty is the result of code that works in the overwhelming majority of cases but doesn’t take other use cases into account. @ C C Meyer EricZ O N A


Slide 30

"The Year in Review ad keeps coming up in my feed, rotating through different fun-and-fabulous backgrounds, as if celebrating a death, and there is no obvious way to stop it."


Slide 31

INCREASE AWARENESS OF 
 A N D C O N S I D E R AT I O N F O R 
 T H E FA I L U R E M O D E S , T H E E D G E C A S E S , T H E W O R S T- C A S E S C E N A R I O S . –Eric Meyer @CCZONA


Slide 32

BE HUMBLE. WE CANNOT INTUIT I N N E R S TAT E , EMOTIONS, 
 P R I V AT E SUBJECTIVITY @CCZONA


Slide 33

F I TB I T S E X TR A C K I N G Algorithmic Profiling


Slide 34


Slide 35


Slide 36


Slide 37


Slide 38

UBER


Slide 39


Slide 40


Slide 41


Slide 42

consent is permission granted Informed freely (where "no" is consequence-free alternative and the default value) with informed appreciation and understanding (ahead of time) of the facts, implications, and consequences


Slide 43

Informed consent is permission freely granted (where "no" is consequencefree alternative and the default value) with informed appreciation & understanding (ahead of time) of the facts, implications, & consequences


Slide 44

GOOGLE ADWORDS


Slide 45


Slide 46

A BLACK IDENTIFYING NAME WAS 
 2 5 % M O R E L I K E LY T O R E S U LT I N A N A D T H AT I M P L I E D A N A R R E S T R E C O R D – H a r v a r d U n i v e r s i t y, D i s c r i m i n a t i o n i n O n l i n e A d D e l i v e r y s t u d y ( 2 0 1 3 ) @CCZONA


Slide 47

GOOGLE / FLICKR / APPLE I M A G E A N A LY S I S


Slide 48

iPhoto Face Detection @CCZONA


Slide 49

Flickr Auto-Tagging @CCZONA


Slide 50

Flickr Auto-Tagging @CCZONA


Slide 51

Google Photos 
 Auto-Categorizing @CCZONA


Slide 52

AFFIRM


Slide 53


Slide 54


Slide 55


Slide 56


Slide 57


Slide 58

F L I P P I N G T H E PA R A D I G M LESSONS FROM THE PROFESSIONAL ETHICISTS


Slide 59

AV O I D H A R M 
 TO OTHERS ACM Code of Ethics @CCZONA


Slide 60

AVOID HARM consider decisions' potential impacts on others


Slide 61

AVOID HARM project the likelihood of consequences to others


Slide 62

CONTRIBUTE TO 
 HUMAN WELL-BEING ACM Code of Ethics @CCZONA


Slide 63

CONTRIBUTE WELL-BEING minimize negative consequences to others @CCZONA


Slide 64

B E H O N E S T. B E T R U S T W O R T H Y. ACM Code of Ethics @CCZONA


Slide 65

H O N E S T. T R U S T W O R T H Y. Provide others with full disclosure of limitations @CCZONA


Slide 66

H O N E S T. T R U S T W O R T H Y. Call attention to signs of risk of harm to others @CCZONA


Slide 67

A C T I V E LY C O U N T E R 
 BIAS & INEQUALITY @CCZONA


Slide 68

A C T I V E LY C O U N T E R B I A S & I N E Q U A L I T Y • equality • tolerance • respect • justice • anti- discrimination @CCZONA


Slide 69

A C T I V E LY C O U N T E R B I A S & I N E Q U A L I T Y • Culture • Systems • Assumptions • Ignorance • Malice @CCZONA


Slide 70

A C T I V E LY C O U N T E R B I A S & I N E Q U A L I T Y • Unequal access to resources & power • Unfair outcomes @CCZONA


Slide 71

AUDIT OUTCOMES @CCZONA


Slide 72

WE'RE IN A RACE


Slide 73

DEEP LEARNING ARMS RACE • Facebook, Google, and Microsoft are making big bets @CCZONA


Slide 74

INSIGHTFUL ALGORITHMS ARE GROWING • More precise in correctness • More damaging in wrongness @CCZONA


Slide 75

E M PAT H E T I C C O D E R S Identify potential harms @CCZONA


Slide 76

E M PAT H E T I C C O D E R S Anticipate diverse ways to screwup @CCZONA


Slide 77

WE MUST HAVE DECISION-MAKING AUTHORITY
 IN THE HANDS OF H I G H LY 
 D I VEEM S S E R T A 
 @CCZONA


Slide 78

DIVERSE. OF DIFFERING KIND, FORM,AND CHARACTER @CCZONA


Slide 79

Tokenism is not diversity @CCZONA


Slide 80

The point of culture fit is to avoid disruption of groupthink @CCZONA


Slide 81

Unidimensional variety
 is not diversity @CCZONA


Slide 82

@CCZONA


Slide 83

E M PAT H E T I C C O D E R S Cultivate informed consent @CCZONA


Slide 84

E M PAT H E T I C C O D E R S Audit constantly @CCZONA


Slide 85

E M PAT H E T I C C O D E R S Recognize bias is inherent @CCZONA


Slide 86

E M PAT H E T I C C O D E R S Visionary about countering bias @CCZONA


Slide 87

E M PAT H E T I C C O D E R S Aim mining tools at public benefit consequences @CCZONA


Slide 88

E M PAT H E T I C C O D E R S Commit to transparency @CCZONA


Slide 89

PROFESSIONALS 
 APPLY EXPERTISE & JUDGEMENT ABOUT 
 HOW TO 
 SOLVE PROBLEMS @CCZONA


Slide 90

❌ ❌ ❌❌ ❌ ❌❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ Algorithmic Profiling DeAnonymization Black Box Personally Human Replicates Bias Identifiable Info Complexity Fail Uncritical Assumptions Moved Fast, Broke Things High Risk False Neutrality Consequences Invasion Of Privacy Not How Valid Data Insecurity Research Works Disparate Impact Unproven Methods Deception No Recourse Creepy Stalkery Accurate, But Not Right Shaming ❌ ❌ ❌ ❌ Consent Issues Inadvertent Algorithmic Cruelty Filtering Messing With Heads Diversity Fail


Slide 91

REFUSE TO 
 P L AY A L O N G . @CCZONA


Slide 92

THANK YOU. @CCZONA


Slide 93

CREDITS @CCZONA


Slide 94

I M A G E U N D E R S TA N D I N G : 
 DEEP LEARNING WITH CONVOLUTIONAL NEURAL NETS Roelof Pieters @graphific http://www.slideshare.net/ roelofp/python-for-imageunderstanding-deeplearning-with-convolutionalneural-nets @CCZONA


Slide 95

A H I P P O C R AT I C O AT H F O R D ATA S C I E N C E Roelof Pieters @graphific http://www.slideshare.net/ roelofp/a-hippocratic-oathfor-data-science @CCZONA


Slide 96

THE ETHICS OF BEING A PROGRAMMER Kate Heddleston @heddle317 https://youtu.be/ DB7ei5W1eRQ @CCZONA


Slide 97

IMAGES • Kate Heddleston: PyCon SE 2015 video • Roelof Pieters: Hendrik https://twitter.com/hen_drik/status/612653056421982208 • Code is made by people: Chris Eppstein https://twitter.com/chriseppstein/status/611612633335095296 • Recipe cards: Olivia Juice & Co. http://olivejuiceco.typepad.com/my_weblog/2006/04/easter_recap_an.html • Crochet pattern & shawl: http://www.abc-knitting-patterns.com/1129.html • Wargames font: http://www.urbanfonts.com/fonts/wargames.htm • Space Invaders: http://www.caffination.com/backchannel/shooting-for-the-high-score-3942/ • Wheat: https://www.flickr.com/photos/paule92/9506023990/ • Eye shadows: https://www.flickr.com/photos/niallb/5300259686/ • Colored pencils: https://www.flickr.com/photos/jenson-lee/6315443914/ • Buttons: https://www.flickr.com/photos/[email protected]/3616793654/ • Cookie face: https://www.flickr.com/photos/amayu/4462907505/in/[email protected]/ • Facebook emotion marionettes: http://altlaw.info/2014/07/emotional-contagion-subliminal-messaging-wheredoes-it-end-for-facebook/ @CCZONA


Slide 98

ARTICLES • Association for Computing Machinery Code of Ethics http://www.acm.org/about/code-of-ethics • The 10 Commandments of Egoless Programming http://www.techrepublic.com/article/the-tencommandments-of-egoless-programming-6353837/ • Disparate Impact Analysis is Key to Ensuring Fairness in the Age of the Algorithm http:// www.datainnovation.org/2015/01/disparate-impact-analysis-is-key-to-ensuring-fairness-in-the-ageof-the-algorithm/ • On Algorithmic Fairness, Discrimination and Disparate impact. http://fairness.haverford.edu/ • Big Data's Disparate Impact http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2477899 • White House https://www.whitehouse.gov/sites/default/files/docs/ big_data_privacy_report_may_1_2014.pdf • Algorithmic Accountability & Transparency http://www.nickdiakopoulos.com/projects/algorithmicaccountability-reporting/ • What is Deep Learning and why should you care? http://radar.oreilly.com/2014/07/[email protected] learning-and-why-should-you-care.html


Slide 99

CASE STUDIES • FitBit: http://thenextweb.com/insider/2011/07/03/fitbit-users-are-inadvertently-sharing-details-of-their-sex-lives-with-the-world/ • Uber: http://www.forbes.com/sites/kashmirhill/2014/10/03/god-view-uber-allegedly-stalked-users-for-party-goers-viewingpleasure/ & http://valleywag.gawker.com/uber-allegedly-used-god-view-to-stalk-vip-users-as-a-1642197313 • Affirm: http://recode.net/2015/05/06/max-levchins-affirm-raises-275-million-to-make-loans/ & http://time.com/3430817/paypallevchin-affirm-lending/ & http://www.nytimes.com/2015/01/19/technology/banking-start-ups-adopt-new-tools-for-lending.html • Shutterfly: http://jezebel.com/shutterfly-thinks-you-just-had-a-baby-1576261631 & http://www.adweek.com/adfreak/shutterflycongratulates-thousands-women-babies-they-didnt-have-157675 (Zuckerberg on miscarriage: https://www.facebook.com/ photo.php?fbid=10102276573729791) • Target: http://www.nytimes.com/2012/02/19/magazine/shopping-habits.html • Facebook Emotional Contagion Study: http://www.theguardian.com/science/head-quarters/2014/jul/01/facebook-cornellstudy-emotional-contagion-ethics-breach & http://psychcentral.com/blog/archives/2014/06/23/emotional-contagion-onfacebook-more-like-bad-research-methods/ • Facebook Year in Review: http://meyerweb.com/eric/thoughts/2014/12/24/inadvertent-algorithmic-cruelty/ & http:// www.theguardian.com/technology/2014/dec/29/facebook-apologises-over-cruel-year-in-review-clips • Google AdWords: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2208240 & http://freakonomics.com/2013/04/08/howmuch-does-your-name-matter-full-transcript/ • Flickr: http://www.theguardian.com/technology/2015/may/20/flickr-complaints-offensive-auto-tagging-photos • Google Photos: https://twitter.com/jackyalcine/status/615329515909156865 & https://www.yahoo.com/tech/[email protected] mislabels-two-black-americans-as-122793782784.html


Slide 100

VIDEOS • MarI/O Machine Learning for Video Games https://youtu.be/qv6UVOQ0F44 • Algorithmic Accountability Workshop by Tow Center https://vimeo.com/125622175 @CCZONA


Slide 101

THANK YOU♥ Noah Kantrowitz Mike Foley Heidi Waterhouse Estelle Weyl VM Brasseur Chris Hausler Yoz Grahame
 We So Crafty @CCZONA


Slide 102


×

HTML:





Ссылка: