Computers, Society, and the Lawsmb/talks/wpunj-ethics-law.pdf · people tend to get married more within the same ethnic group), other display lower correlation (e.g. Origin \ Is student).

$Page 1: Computers, Society, and the Lawsmb/talks/wpunj-ethics-law.pdf · people tend to get married more within the same ethnic group), other display lower correlation (e.g. Origin \ Is student).$
Computers,Society,andtheLawStevenM.Bellovin

DepartmentofComputerScience,ColumbiaUniversityhttps://www.cs.columbia.edu/~smb

smb

2

TheSpaceShuttleChallengerl NASAwantedtolaunchtheshuttleonacoldJanuaryday

l ThecrucialO-ringshadneverbeentestedatlowtemperatures,butsomeThiokolengineerssuspectedaprobleml RogerBoisjoly hadwarnedofitsixmonthsearlier

l AllanMcDonald,directorofthesolidrocketprogramatThiokol,opposedthelaunch

l NASA:“MyGod,Thiokol,whendoyouwantmetolaunch,nextApril?”

Engineersoftenknowthingsthatmanagersdon’tknowbutneedto

smb

3

SomeCasesareEasylVolkswagenandthe“defeatdevice”softwaretofoolemissionstests

lPrenda Lawanditsboguscopyrightinfringementlawsuitsl AjudgehearingonecasereferredthemattertotheFBI…

lViruses,ransomware,andthelike

smb

4

VotingMachinesl There’slongbeeninterestincomputerized(“DRE”—DirectRecordingElectronic)votingmachinesandInternetvoting

l Virtuallyallcomputerscientistsopposetheidea:“Don’tuseourtechnology!”

l But:“Webankonline;whycan’twevotethatway?”

smb

5

(PhotobyEdFelten)

ComputerScientistsandVotingSystemsl We knowhowbuggyandinsecuresoftwarecanbe

l We knowthatATMs,etc.,canhavelogfilesand(insomecases)wecan“unwind”problematictransactions

l But—anonymityandresultintegrityareextremely importantinvoting

l (Rerunningelectionsisproblematic.Iflastyear’selectionwerererunaweeklaterbecauseofcomputerproblems,whatwouldtheresultshavelookedlike?)

Howdowecommunicatethesoftwareissuestolegislators?smb

6

Encryptionl TheFBIclaimsthatthey’re“goingdark”becauseofincreasinguseofencryption

l Theywantsomesortof“exceptionalaccess”toletthemgetattheplaintext

l Mostcryptologiststhinkthatthisisdangerous,thatcryptographicprotocolsandmechanismsarefartoohardtogetright

l Why?

smb

7

HistoricalExample:TheWorldWarIIEnigmaMachine

Photo:publicdomainsmb

8


Youselecttheproperrotors


9


Adjusttherotorstotheir“groundsetting”


10


Settheplugboard

Photo:BobLord,viaWikiMedia Commonssmb

11


Photo:PaulHudson,viaFlickr

• Pickthreerandomlettersandencryptthemtwice,andsendthosesixlettersasthestartoftheencryptedmessage

• Resettherotorstothosethreeletters

smb

12

WhatCouldGoWrong?lSendingthesame,simplemessageeverydaywasafatalflaw

lPickingnon-randomletterswasafatalflaw

lSendingamessageconsistingofnothingbuttheletter“L”wasafatalflaw

lEncryptingthethreeletterstwice wasafatalflaw

smb

13

TheThreeLettersl Imaginethat“XJM”wasencryptedto“AMRDTJ”

l ThecryptanalystsrealizedthatA andD representedthesameletter,M andT werethesame,andR andJ werethesame

l Thisgaveawayvaluablecluestotherotorwiringandtherotororder!

Cryptographyishard…

smb

14

LegalIssuesl Sometimes,therearelegalissuesinvolvingcomputertechnologyl Today,almosteverythinginvolvescomputertechnology…

l Mostlegislatorsandjudgesknownothingofcomputers

l Howcantheyreachtherightanswer?

l Wemayknowtheanswers—butwehavetolearntospeaktheirlanguage:thelaw

smb

15

Example:WiretapLawandtheInternetl UnderUSlaw,phoneandemailconversationsarestronglyprotected—policeneedasearchwarrantbasedon“probablecause”toobtainthem

l However,informationthatisvoluntarilygiventoa“thirdparty”isonlyweaklyprotected;itcanbeobtainedifitis“likelytoberelevant”toanongoingcriminalinvestigation

l Phonenumbersarethird-partydata,obtainedbya“penregister”or“trap-and-tracedevice”

l Whataboutemailaddresses?

smb

16

SendingEmail

17

ISP

ISP ISP

ISP

OutboundMailServer

InboundMailServer

Email(Simplified)lMailgoesfromasender’sdevicetoan“outboundmailserver”

l Fromthere,itissenttotherecipient’s“inboundmailserver”

l Therecipientdownloadsitfromthatmachine

l ThemailserversaregenerallyISP- orenterprise-operated

It'sTooComplicated

18

SendingMyselfEmail19

220machshav.com ESMTPExim4.82Tue,11Mar201419:43:03+0000HELOeloi.cs.columbia.edu250machshav.com Helloeloi.cs.columbia.edu [2001:18d8:ffff:16:12dd:b1ff:feef:8868]MAILFROM:<[email protected]>250OKRCPTTO:<[email protected]>250AcceptedDATA354Entermessage,endingwith"."onalinebyitselfFrom:BarackObama<[email protected]>To:<[email protected]>Subject:Test

Thisisatest.250OKid=1WNSaS-0001z5-1dQUIT221machshav.com closingconnection

Message

ConversationWithAThirdParty20



Message

WhattheRecipientSees21



Message

CourtsHaveGottenThisWrong‘Thatportionofthe“header”whichcontainstheinformationplacedintheheaderwhichrevealsthee-mailaddressesofthepersonstowhomthee-mailissent,fromwhomthee-mailissentandthee-mailaddress(es)ofanyperson(s)“cc’d”onthee-mailwouldcertainlybeobtainableusingapenregister and/oratrapandtracedevice.’

(InreApplicationofUnitedStates,396F.Supp.2d45)

l Butthe“header”isn’tthird-partydata;it’scontent,whichcannotbeobtainedwithapen/traporder

l Ifyouthinkthat’shardtoexplaintoajudge,whataboutTCPportnumbers?

Paper:http://jolt.law.harvard.edu/assets/articlePDFs/v30/30HarvJLTech1.pdf

smb

22

TheInternet:ALayeredArchitecture

It'sTooComplicated

23

IsaSearchWarrantNeededtoTrackSomeone’sLocationviatheirCellPhone?l Lawenforcement:“No,you’reinpublic,andyou’vegivenyourlocationtothephonecompany”

l But—theFourthAmendmentbars“unreasonable”searches

l Legalacademics:ifyoutracksomeonefortoolong,youcanbuildupaveryfullpictureoftheirlife,whichis unreasonable(called“mosaictheory”)

l Rejoinder:Howlongis“toolong”?Howwillpoliceknowwhentheyneedawarrant?

smb

24

CanBigDataProvideanAnswer?lUsemachinelearningtomakepredictionsbasedonlocationdata

lWhenpredictionsareaccurateenough,amosaicexists

l Inotherwords,usecomputersciencetoanswerthequestion

smb

25

l Thetechnicalliteraturesupportsthebasicpremise:withenoughpoints,thewholeisgreaterthanthesumofitsparts

l Notethejumpinaccuracyat5weeksand28weeks

Figure 9. Predicting Significant Other over Time – we chose the significant other as the node with the maximum strength.

Figure 10. Predicting ethnicity using SMS social network over time (65 weeks) – after every week we analyze the graph with the same method as described at 3.4 (Louvain Algorithm). Figure 11 demonstrates the correlations among the learning process dynamics of several features. It was calculated using the Pearson product-moment correlation coefficient (a measure of the linear dependence between two variables X and Y, giving a value between +1 and −1). The correlation is defined as the covariance of the two variables divided by the product of their standard deviations. In general, variables of correlation higher than 0.5 are usually considered strongly correlated.

Figure 11. Pearson correlation between the learning process dynamics for three of the properties we predict. As might be expected, there are some strong correlation between the different evolution trajectories of the learning processes of the three features. However, notice that while some are very highly correlated (e.g. Origin \ Significant other), which might point out a strong correlation in the underlying data itself (i.e. people tend to get married more within the same ethnic group), other display lower correlation (e.g. Origin \ Is student).

5. DISCUSSION AND FUTURE WORK As reviewed in section 3.4, the Gompertz function is a well-known technique that has been used to model processes over time. Our analysis confirms that the evolving learning of social and individual features, as mobile phone sensing data accumulates over time, can also be fitted to the form of a Gompertz function. We see that this result is true for the prediction of different features, both social and individual, and for a set of different prediction methodologies, using a varying number of input signals, all collected via mobile phones in a field deployment.

Correlations between the evolution trends of the different learning process, as depicted in Figure 11, may imply underlying correlation between the raw data itself, and can hence be used as additional validation for correlated features and observations (such as the suggestion that people might have a higher tendency to marry within their own ethnic group, as has been widely observed [33,34]). In addition, this information could be used for informing the design of data collection configuration for an ongoing or future data collection initiative. For example, if we know of two features that are highly correlated in the same experiment, but one of them is very “cheap” to gather from a processing or battery power perspective, while the other is very expensive, we might decide that the cheaper one is sufficient (e.g. one requires just reading the phone’s built-in call-log database while the other requires battery-intensive GPS scanning). Alternatively, we might want to make sure that two correlated values are gathered in order to strengthen the result and help deal with noise.

MachineLearningand MosaicTheory

(GraphfromAltshuler etal.)

4/13/17

26

OneWeekisthelimitlExperimentsshowthatweek-to-weekmovementsareverypredictable(Sadilek &Krumm)

lWeekendmovementsaremore predictable,thoughofcoursedifferentthanweekdaymovement

lWithsevendaysofobservation,youhaveaverygoodpictureofsomeone’slife

4/13/17

27

WhereAreWe?

4/13/17

28

lFromatechnicalperspective,mosaictheoryiscorrect:youreallycanbuildaveryfullpictureofsomeonewithenoughdatapoints

lThelimitshouldbeaboutoneweek

lBut—movementsarestillinpublic

lBut—thereareotherlegalissuesthatmightariseinspecificcases,suchasthethirdpartydoctrine

ResultslThesciencealoneisn’tenough

lFundamentally,thisisalegalquestion,notatechnicalone.Wecansupplyfactsbutthecourtsdeterminethelaw.Gettingtherightanswerrequiresbothkindsofinput,legalandtechnical.

Paper:http://lawandlibertyblog.com/s/Hutchins.pdf

4/13/17

29

WhatDoWeDo?l Firstandforemost:decide tobeinvolved

l Beawareofsocietalissuesl Makeethicalchoicesaboutcareerpathsandon-the-jobbehavior

l Learnthelanguageoflawandpolicyl Youdon’thavetobealawyer—I’mnot—butyoudoneedtounderstandhowtotalktopolicymakers

l Getinvolved—spendtimeinTrentonorWashington

l Ifyoudon’tspeak,theycan’tlisten,eveniftheywantto

smb

30

Computers, Society, and the Lawsmb/talks/wpunj-ethics-law.pdf · people tend to get married more within the same ethnic group), other display lower correlation (e.g. Origin \ Is student).

Documents