cs 4001: computing, society & professionalism · 2019. 4. 1. · cs 4001: computing, society...
TRANSCRIPT
CS 4001: Computing, Society & ProfessionalismMunmun DeChoudhury |AssistantProfessor|SchoolofInteractiveComputing
Week 12: Algorithms and (Lack of) ControlMarch 26, 2019
A Lack of Control
• Techcompaniesowningprofuseamountsofdata§ Companiescontrolmonopolies(Amazon,Facebook,
andGoogle)
• Techcompaniesopaquelysharingdatawithpartners
• Algorithmsinfluencingourexperiencesonline
Some Examples
• Todayalgorithmscanshapewhatyoubuy,whereyoulive,whetheryougetajoborabankloan,andmanyotheraspectsofyourlife.
• Autocompletenowpredictsyourwordsintextmessages,Gmail,andsearchterms.
• EvenTinderiscontrolledbyalgorithms— didyoupickyourloveordidTinder?
• Doyoupickwhatyouwatchorbuyifmorethan80percentofwhatyouwatchonNetflixand30percentofpurchasesonAmazonaretheresultofanalgorithm?
Why?
• Naturevs.nurture§ Algorithmicdecisionshardwiredbyanengineer§ Now,machinelearning– wedon’thavetohardcode
inalltherules,letthesystemlearntherelevantrulesbylearningfromdata
• Natureisthehumancode,thecodethatisessentiallygiventothealgorithmorthat’spartofthealgorithm,liketheequivalentofgeneticcode.Soit’sthenatureofthealgorithm.Andnurtureisthedatafromwhichitlearns.
CaseStudy1
Impacting Real World Outcomes: The Positive Side
• Theuseofdigitalmediatodiscusspoliticsduringelectiontimeshasalsobeenthesubjectofvariousstudies,coveringthelastfourU.S.Presidentialelections(Adamic andGlance,2005;Diakopoulos andShamma,2010;Bekafigo andMcBride,2013;CarlisleandPatton,2013;DiGrazia,etal.,2013;Wang,etal.,2016)
• Mostworkfocusesonthepositiveeffectsofsocialmediasuchasincrementingvotingturnout(Bond,etal.,2012)orexposuretodiversepoliticalviews(Bakshy,etal.,2015)contributedtothegeneralpraiseoftheseplatformsasatooltofosterdemocracyandcivilpoliticalengagement(Shirky,2011;LoaderandMercea,2011;Effing,etal.,2011;Tufekci andWilson,2012;Tufekci,2014;Yang,etal.,2016)
Social bots distort the 2016 US Presidential Election Online Discussion
• Quantitativeinvestigationofhowthepresenceofsocialmediabots,definedasalgorithmicallydrivenentitiesthatonthesurfaceappearaslegitimateusers,affectedpoliticaldiscussionaroundthe2016U.S.Presidentialelection
• Data:over20milliontweetsgeneratedbetween16Septemberand21October2016byabout2.8milliondistinctusers;datapriortothethreePresidentialdebates
• Findings:§ OnefifthofTwitterconversationsrelatedtotheelection
generatedbybots§ Networkanalysisandembeddedness ofhumanandbot
connectionsrevealedthatbotshampereddemocratizeddiscussion
Ecosystem of social media bots
• SearchTwitterforphrases/hashtags/keywordsandautomaticallyandretweet them
• Automaticallyreplytotweetsthatmeetacertaincriteria• Automaticallyfollowanyusersthattweetsomethingwithaspecific
phrase/hashtag/keyword• Automaticallyfollowbackanyusersthathavefollowedthebot• Automaticallyfollowanyusersthatfollowaspecifieduser• Automaticallyadduserstweetingaboutsomethingtopubliclists• SearchGoogle(andotherengines)forarticles/newsaccordingto
specificcriteriaandpostthem,orlinktheminautomaticrepliestootherusers
• Automaticallyaggregatingpublicsentimentoncertaintopicsofdiscussion
• Bufferandposttweetsautomatically
The challenges of bots
• Botsarealmostentirelyanonymousandcanbeeasilyboughtinsecretfromcompaniesorindividualprogrammers
• Sourcecodeavailablefordevelopingyourownbot
• Canbeemployedaspartofanorganizedeffort
Results
Similar results
• Oxfordresearchersfoundthat“highlyautomatedaccounts— theaccountsthattweeted450ormoretimeswitharelatedhashtag andusermentionduringtheperiodbeforeelection— generatedcloseto18percentofallTwittertrafficaboutthepresidentialelection.”
• Theyalsonotedthatbotstendtocirculatenegativenewsmuchmoreeffectivelythanpositivereports.
But we still don’t quite know if the bots really influenced election outcomes…. We will perhaps never know (don’t have data on a counter-factual situation)
ClassActivity1
Bots Generate False News
• Shaoetalidentifiedfalsenewssites:infowars.com,breitbart.com,politicususa.com,andtheonion.com.
• Authorsthenmonitoredsome400,000claimsmadebythesewebsitesandstudiedthewaytheyspreadthroughTwitter.Theydidthisbycollectingsome14millionTwitterpoststhatmentionedtheseclaims.
• Atthesametime,theteammonitoredsome15,000storieswrittenbyfact-checkingorganizationsandoveramillionTwitterpoststhatmentionthem.
• Next,theylookedattheTwitteraccountsthatspreadthisnews
• Socialbotsplayakeyroleinthespreadoffalsenews
Steps being taken
• GoogleannouncedinNov2016thatitwouldbanwebsitesthatpeddlefakenewsfromusingitsonlineadvertisingservice.
• Facebookafterinitialdenial,announcedupdatingthelanguageinitsAudienceNetworkpolicy,whichalreadysaysitwillnotdisplayadsinsitesthatshowmisleadingorillegalcontent,toincludefakenewssites.§ Currentlyasignificantresearchagendatoassess
veracityofinformationshownonNewsFeed
ClassActivity2
Legislationdoesnotovercomeinternationalborders.Giventherecentconjecturesandevidencearoundhowforeignpowershavemanipulatedthespreadoffalsenewspriortothe2016Presidentialelections,it’shardtoseehowthiswouldwork.
CaseStudy3
The Cambridge Analytica-Facebook Scandal
• Thedataanalyticsfirmusedpersonalinformationharvestedfrommorethan50millionFacebookprofileswithoutpermissiontobuildasystemthatcouldtargetUSvoterswithpersonalizedpoliticaladvertisementsbasedontheirpsychologicalprofile
• FacebookreceivedanumberofwarningsaboutitsdatasecuritypoliciesinrecentyearsandhadknownabouttheCambridgeAnalytica databreachsince2015,butonlysuspendedthefirmandtheCambridgeuniversityresearcherwhoharvesteduserdatafromFacebookearlierthismonth
Brexit and 2016 Presidential election links
• DuringtheBrexit referendum,adigitalservicesfirmlinkedtoCambridgeAnalytica receiveda£625,000paymentfromapro-Brexit campaignorganization
• Inthesummerof2016CambridgeAnalytica caughttractioninTrumpTower.OneofthetopcampaignofficialsreachedouttoCambridgeforhelpbuildingageneralelection-styledataoperation.§ Trumpson-in-lawJaredKushnersuggestedthatwas
athisdirectioninapost-campaigninterviewwithForbesmagazine.
Consequences
• BillionsofdollarshavebeenwipedoffFacebook’sstockmarketvaluationasagrowing#DeleteFacebookmovementandregulatoryfearshavespookedinvestors.
• FacebookisbeinginvestedbytheFTC.
• AdvertisersarepullingadsfromFacebook,companiesareeliminatingFacebooklog-infunctions.
Class Activity 2
Hosanagar’s Algorithmic Bill of Rights
• TheAlgorithmicBillofRightsaddressessomekeyprotectionsconsumerscanandshouldexpect
Hosanagar’s Algorithmic Bill of Rights
• Transparencyofdataappropriation§ UseofFacebookdatainhiring
• Transparencywithregardtotheactualdecisions§ Whywastheloandenied?
• Usercontrol§ Usersattheveryleastshouldhavesomeabilitytoturnonorturnoffsomeofthese
systems,forexample,tobeabletotellasmartspeaker‘Don’t listentomerightnow’ or‘Don’t listenuntilIsayI’mreadyforyouto listen.’
§ Thisthirdpillarisessentiallyaroundsomefeedbackloopwhereuserscanhavesomeimpactonalgorithmicchoice
• Formalaudits§ Forlargecompanies,beforetheydeploytheiralgorithmstheyactuallyshouldaudit