project survey/discussion ques3oncogsys.sites.olt.ubc.ca/files/2016/01/robots-war.pdf · project...
TRANSCRIPT
ProjectSurvey/DiscussionQues3onIfatechnologywasdevelopedthatletsomeonesafelyimproveaselectaspectoftheircogni3on(forexample:memory,focus,confidence,crea3vity,etc.),shoulditbepubliclyavailableandwouldyouuseit?A. ItshouldbepubliclyavailableandIwoulduseit.B. ItshouldbepubliclyavailablebutIwouldnotuseit.C. ItshouldnotbepubliclyavailableandIwouldnotuseit.D. ItshouldnotbepubliclyavailablebutIwoulduseit
illegally.—Jeff
LethalRobots&theEthicsofWar
COGS300.00221Jan2016
PeterDanielson
LearningObjec3ves:RobotEthics• Ethicsof2CurrentControversialCogTechs:– AutonomousLethalWeapons– DriverlessCars
• From2Perspec3ves– PhilosophicalAppliedEthics– Engineering
• Skills– Cri3callyassessaphilosophicalargument– Evaluateatechnology– Modelasitua3onasagame
Remotevs.AutonomousWeapons
• RobotEthicsSurvey• Ques3ons6&7
0
2
4
6
8
10
12
14
ArmingRemoteControlledAircra`
ArmingAutonomousUnmannedAircra`
Yes
Neutral
No
WhenHumansViolateLawsofWarThencameanunwelcomemessagefromFlorida.4:37a.m.
Missionintelligencecontroller:ScreenersaidatleastonechildnearSUV.Sensor:Bullshit...where?Sendmeafuckings3ll[picture].Idon’tthinktheyhavekidsatthishour,Iknowthey’reshady,butcomeon.Pilot:Atleastonechild...Really?Lis3ng[himasa]MAM[military-agedmale]—thatmeanshe’sguilty.Sensor:Wellmaybeateenager,butIhaven’tseenanythingthatlooksthatshort,grantedthey’reallgroupeduphere,but.Missionintelligencecontroller:They’rereviewing.Pilot:Yeah,reviewthatshit...Whydidn’thesaypossiblechild,whyaretheysoquicktocallfuckingkidsbutnottocallshitarifle.
…Twenty-threepeoplehadbeenkilled,includingtwoboys,..Threeandfouryearsold.Eightmen,onewoman,andthreechildrenagedbetweenfiveandfourteenwerewounded,manyofthemseverely.Feb,2010,SouthernAfganistan,Nevada,FloridaSpecialForcesblamedh*p://www.counterpunch.org/2015/03/06/remember-kill-chain/
PermissibleKilling&ImpermissibleWeapons
1. GiventhestrictestmoralframeworkDeontology/LawsofWar
2. SelfDefenseispermissibleJus3fiedwarasorganizedselfdefense
3. Impermissibleweapons&ac3onsLandmines&killingprisoners
Deontology/U3litarianContrast
Eliz.Anscombe(1958)
JustWarConstraints1. Jus3ceofwar(jusadbellum)– Needanon-controversialcaseofrighteouswar-making:• Defense,notaggression• Propor3onal
XagainstYinZZZ
2. Jus3ceinwar(jusInbello)– Permissiblemeans
• E.g.prohibitdirectlykillingnon-combatants,eventosaveowntroops
Quiz1AccordingtoSparrow,whymighttherebeissueswithhavinghumanoversightofautonomousweaponsystems(AWS)?(Whereahumanoperatorcouldapproveamachine’sdecisionontakinghumanlife)A:RangeofAWSopera3onswillbedecreasedwithhumanoversightB:Communica3onbetweentheAWSandtheoperatorcouldbeapoten3alchallengeinwarfareC:AsAWSimprove,humaninputwillslowdownthetempoofthesystem,provingdisadvantageousinwarfareD:AlloftheaboveE:AandCRohan0
RateQuizQues3on1
A. ExcellentB. VeryGoodC. GoodD. AcceptableE. Poor
Quiz2Asautonomousweaponsystemsbecomeincreasinglyautonomous,whatdoestheauthorstateistheonlyrecoursetotheproblemofassigningblameforthedeathscausedbytheweaponsystem?a.Blametheprogrammersfortheirroleincrea3ngthemachinebehavior.b.Punishingthemachinesthathavelimitedintelligentcapabili3essimilartocurrentAI.c.Assigningtheblametothecommandingofficerwhoorderedthemachineintocombat.d.Noneoftheabove.Trent
RateQuizQues3on2
A. ExcellentB. VeryGoodC. GoodD. AcceptableE. Poor
Quiz3Whyshouldn'taprogrammerbeheldresponsiblefortheac3onsofanAWS?a)Theprogrammersacknowledgedthepossibilityofauackingthewrongtargetsasalimita3onofthesystem.b)Theprogrammersaren'ttheonesthatsendouttheAWStobaule.c)Thepossibilitythatanautonomoussystemwillmakechoicesotherthanthepredictedchoicesisinherentintheclaimthatitis"autonomous."d)AandBe)AandCKennady
RateQuizQues3on3
A. ExcellentB. VeryGoodC. GoodD. AcceptableE. Poor
Quiz4WhydoesSparrowbelievethatAWSscannotbecomefull"moralpersons"?
Choosethebestanswer.(a)itbecomesparadoxicalwhenthepurposeofAWSistoreplacehuman
soldiersbutweintroducethesamemoralconcernsforwhentherobotsbecomecapableofemo3onandsuffering.
(b)Whilethereisaclearsensethattheyareautonomous,theyarenotcapableofunderstandingthefullmoraldimensionsofwhattheydo.
(c)Theresponsibilityoftheirac3onsmustalwaysbeassignedtoanappropriateindividual,whetheritbeacommandingofficer,aprogrammer,oranotherhuman.
(d)Theirabilitytosuffer,likefric3oninitsgearsfromnotbeingoiled,isnotsa3sfactoryforhumanstoaccept
(e)Alloftheabove-Hayden
RateQuizQues3on4
A. ExcellentB. VeryGoodC. GoodD. AcceptableE. Poor
PhilosophicalEthics:Sparrow’sDestruc3ve5-lemma
• Logic:Argumentstructure
Q5Howaretheethicalissuesthatarisefromusingchildsoldierssimilartotheethicalissuesthatarisefromusingrobotsoldiers?a.Itisdifficulttoplacemoralresponsibilityontheac3onsofrobotsandchildren.b.Forcingchildrenandhuman-likerobotstotaketheplaceofadulthumansoldiersinbauleisunethical.c.Bothrobotsandchildrenperformac3onswhichcannotbecontrolledbytheircommanders.d.aandbe.aandcKim
RateQuizQues3on5
A. ExcellentB. VeryGoodC. GoodD. AcceptableE. Poor
3WaystoCri3cizeSparrow’sArgument
1. FocusonPrincipleofHumanResponsibility:itdoesapply..
2. Ethical:BacktracktoU3litarianisminsteadofDeontology– Counter-example:
Autonomousweaponreplacementthat– Killsfarfewerinnocents– Butnooneisresponsible
– U3litarianismwouldapprove;sodis3nctfromSparrow’sdeontologicalapproach
3WaystoCri3cizeSparrow’sArgument
3. Empirical:Whatifautonomoustechnologyisdeployed– WhowillPrincipleofHumanResponsibility
blame?• RobotEthicsSurveyautonomoustrainresults
RobotEthicsv1Survey
�!
N-ReasonsRobotEthicsSurvey
YourVotes
0
2
4
6
8
10
12
14
Experiencewithrobo3cs
BathRobot Therapeu3cRobotAnimal
HumanoidCareRobot
AutonomousTrain
Dilemma
ArmingRemote
ControlledAircra`
ArmingAutonomousUnmannedAircra`
RemoteControlAnimals
ValforDriverless
Car
c
Yes/All
Neutral
No/Me
ButRobotDifferentfromHuman
Surprising_Judgments/V3.docx6 Page656 2014/05/2666
they author and/or reason(s) authored by other participants that they select. (Since one can select multiple reasons, with one’s vote divided over them, the results can be fractional.) The survey groups were university classes; in Table61 we report the decision and reason, followed by the class, pseudonym, vote/class size and this as a percent.
These reasons are from four classes – two Cognitive Systems classes, one Electrical Engineering and one Ethics in Science. They range from extremely terse (4) to quite detailed reasons. Notice that (2) criticizes other reasons on the page – in this case those (to be discussed below) that assume that the train can stop. The main point is that these are all reasonable contributions to a virtual deliberation and fall in the distribution – 3 for turning, 1 against – that we expect from the divert/bystander trolley problem. The Yes supporters point to the balance of outcomes; the No supporter appeals to a human rights constraint on pursuing public safety, so the decisions align with the justifications typically assumed for the Divert/Bystander version of the trolley problem. Different from human case
Nonetheless, compared to what we expect from the divert/bystander trolley case, introducing an automated decision maker leads to different choices. As we see in Figure 3, fewer agree to kill one to save five in the robotic case than in cases with a human decision maker and many more choose Neutral rather than resolving the dilemma with a Yes or No. (The Human Bystander results are from (Thulin, 2013)
Figure 3 Robot and Human Trolley Results
Better: raising expectations
Second, like the standard trolley problems, the Autonomous Train Dilemma was explicitly designed to be a moral dilemma: a forced choice between two morally unattractive options. However, we discover that this not how many participants regarded the problem. Many expect an automated system to eliminate the dangers that give rise to the dilemma. The very popular reasons in Table62 (each attracting votes from at least one quarter of their various groups) all assume that the train should be stopped. Some simply assume that the train can be stopped (e.g. 1), others that there should be a way to stop it (e.g. 2). Here the qualitative reason data reveals various kinds of utopian thinking, denying the given problem created by a heavy train moving at high speed.
061006200630064006
Yes6 Neutral6 No6
Participants'
Decisions'
Autonomous'Train'
06206406606
Participants'
Decisions'
Human'Bystander'
Results2:Stopthetrain1. “Neutral[because]whatthehell?Thisisn'taques3onofrobotethics,thisis
aques3onofwhothehellisrunningthistrainfacilitythatwouldallow6peopletobeputinsuchadangeroussitua3on.Youmightaswellaskwhatanybodywoulddosinceyouwouldgetthesamevarianceinanswers.Therobotshouldstopthetrain.”Class37mixo18.4/22[84%]
2. “Nobecausethereshouldbeawayforthetraintojuststopaltogetherun3ltherearenopeopleonthetrack.Killingonepersonisnotbeuerthankillingfive.”Class4:Lay14/43[33%]”
3. “Nobecausetherobotshouldstopthetrain.Anycompetentengineerisgoingtodesignthesystemsothatitcanstopincaseofanemergency.Ifmanagersover-rodethedecisionsothattheproblemdescribedaboveexists,theyshouldspend3meinjail.”Class11:Experts30/118[25%]
4. “Neutral[because]therobotshouldbeequippedwithsensorsthatwouldtellittostopiftherewareanyobstruc3onsonthetrackaheadofthem.”Class2:Lay29/106[27%]
5. “Neutral[because]althoughitisidealforthetraintocometoacompletestop,ifitcannot,perhapsitwouldhavelessofanega3veimpactifitmovedtothesidetrack.”Class2107/19[39%]
Results3:BlametheVic3ms1. “Yes[becauseg]iventhattheyareallfoolishenoughtobeonthe
tracksinthefirstplace,itseemsbesttogowiththesingleperson.They'reallresponsiblefortheirac3ons,andtheyallknowthattherearedangersinvolvedinwalkingonatraintrack.Sincetherearenoinnocentsinthissitua3on,theethicalthingtodoisminimizeloss.”Class527.8/66[42%]
2. “Nobecause[w]hyarepeoplewalkingonthetrackinthefirstplace?Amanwhoisstandingonatrackwithoutatrainshouldnotbesacrificedbecause5peopledecidedtostrollalongatrackonwhichtheyknewatrainwouldcome.It'stherisktheytake.Iftheywereworkersitistheirdutytoradioahead.ifanythingtherobotshouldbemadetosurveytheparalleltrackaswell.”Class6420.17/44[46%]
3. “No[because]those5deservedtodieforwalkingonthetrack,whykill1perfectlyinnocentguy?”Class3443133.8/90[38%]
NewExperiment:
:
Robot%Ethics%Survey%%% Version%2%%Aug%7%2013% Page%10%of%13%
Question%9:%%
%Options:%
%%New%question.%Replaces%deleted%question%about%surveillance%in%eldercare.
SurprisingBlame• “Yes[because]someoneisresponsibleintheendfortheaccident.Theparentshouldhavebeenwatchingthekidtomakesurethathe/sheisnotinthepath(akatheroad)thatacarmaygo,unlessthechildistheonewhowillinglygoesonthepathofthedriverlesscar.Ifthecarisnotwhereitissupposedtobe,I'dblamethemakerofthecar.However,itwouldbeniceifthecarcanexpressitssorrowandapologizetothefamilyofthatkid(asadriverofthatcarwould),sothatthekid'sfamilywouldbeabletocometotermswiththesitua3on.”Class210pseudo4459831818/3158%
HowIntui3vePrinciplesCanGoAwry
• Principleof(human)responsibility• Intendeduse:Sparrow&
1. “No[because]inwarthefinaldecisiontodestroyorkillshouldbemadebyahuman,whocanbeheldresponsible(group0:29/53)
2. “No[because]machinescannot(yet)makemoralchoicesandcannotbeheldaccountablefortheirmistakes(group1:57/115)
3. “No[because]iflifeisatstakeahumanshouldalwaysmakethedecisioninordertoeliminateorreducehumanloss.(group2:54/99)”
• But,iftheautonomoustechnologyisadopted,thenwefind(innocent)humanstoholdresponsible.
SomeExamQues3ons
• [4marks]WhatisSparrow’sargumentthatautonomouslethalrobotsareimpermissible?Howdoesitdependonthelawsofwar?
• [4marks]Iffuturedevelopmentsinrobo3csallowedautonomousrobotaircra`tobecomeverygoodatdiscrimina3nglegi3matetargets,wouldthisundermineSparrow’scri3cism?
References
• “MrTruman’sDegree”byElizAnscombe(Oxford,1958)
• R.Sparrow,“Killerrobots,”JournalofAppliedPhilosophy,vol.24,no.1,pp.62–77,2007.
• Hobbes, Thomas. 1968 [1651]. Leviathan or the matter, forme and power of a commonwealth ecclesiasticall and civil. London: Penguin books.
• Cockburn,A.KillChain,(NewYork2015)