d4.1: state-of-the-art review...david wright, trilateral research ltd philip brey, university of...

D4.1: State-of-the-art Review

[WP4–AI&Robotics]

Leadcontributor PhilipJansen,UniversityofTwente

Email:[email protected]

Othercontributors StearnsBroadhead,TrilateralResearchLtd

RowenaRodrigues,TrilateralResearchLtd

DavidWright,TrilateralResearchLtd

PhilipBrey,UniversityofTwente

AliceFox,UniversityofTwente

NingWang,UniversityofTwente&UniversityofZurich

OwenKing,UniversityofTwente(reviewer)

RajaChatila,SorbonneUniversité(reviewer)

VirginiaRomano,UppsalaUniversity(reviewer)

Duedate 31March2018

Deliverydate 13April2018

Type Report

Disseminationlevel PU=Public

Keywords Artificialintelligence,AI,robotics,state-of-the-art,literaturereview,presentapplications,futureapplications,socio-economicimpactassessment,socialimpacts,economicimpacts,environmentalimpacts

TheSIENNAproject-Stakeholder-informedethicsfornewtechnologieswithhighsocio-economicandhumanrightsimpact-hasreceivedfundingundertheEuropeanUnion’sH2020researchandinnovationprogrammeundergrantagreementNo741716.©SIENNA,2018ThisworkislicensedunderaCreativeCommonsAttribution4.0InternationalLicense

741716–SIENNA–D4.1Deliverablereport

2

Abstract Thisreportdescribesthecurrentstateoftheartinthefieldsofartificialintelligence(AI)androbotics.Inthisreport,weestablishdefinitionsanddemarcationsforbothfieldsthatwillbeusedinsubsequentSIENNAresearchonAIand robotics.The reportdiscusses the fields in termsof theirbackgrounds,positions,challenges,andpresentandexpectedapplications.Thisreport includesasocio-economicimpactassessment(SEIA)thatexaminesthecurrentandexpectedimpactsofAIandrobotics.

Document history

Version Date Description Reasonforchange Distribution

V.01 19March2018 Draftforreview Expertreview Reviewers

V.02 16March2018 Draftforreview Consortiumreview Consortium

V.03 13April2018 FinalVersion Finalapproval ECsubmission

V.04 13June2019 Post-reviewversion Minoredits Public

Information in this report that may influence other SIENNA tasks

Linkedtask Pointsofrelevance

Task3.1–State-of-the-artreviewofhumanenhancement

ThereviewofAIandroboticstechnologiesandapplicationsforhumanenhancement

Task4.2–AnalysisofthelegalandhumanrightsrequirementsforHMIinandoutsidetheEU

ThereviewofAIandroboticsapplicationsandsocio-economicimpactassessmentmayinformlegalandhumanrightsinquiries

Task4.3–SurveyofRECapproachesandcodesforHMI

ThereportdefinitionsmayhelptoidentifyconsistenciesordifferencesfromexistingRECapproachesandcodes

Task4.4–EthicalassessmentofHMI ThereviewofAIandroboticsapplicationsandsocio-economicimpactassessmentandwillinformethicsinquiries


3

Table of Contents

Abstract ................................................................................................................................................... 2Table of Contents .................................................................................................................................... 3

Executive summary ................................................................................................................................. 5

List of figures ........................................................................................................................................... 9

List of tables ............................................................................................................................................ 9

List of acronyms/abbreviations .............................................................................................................. 10

Glossary of terms .................................................................................................................................. 10

1. Introduction ........................................................................................................................................ 12

1.1 Artificial intelligence ..................................................................................................................... 121.2 Robotics ...................................................................................................................................... 13

1.3 Structure of the report ................................................................................................................. 15

1.4 Scope and limitations .................................................................................................................. 15

2. Defining the field ................................................................................................................................ 16

2.1 Definitions of artificial intelligence and robotics ........................................................................... 16

Definition of artificial intelligence .................................................................................................... 16

Core concepts, approaches and methods in artificial intelligence ................................................. 17Definition of robotics ....................................................................................................................... 19

Core concepts, approaches and methods in robotics .................................................................... 20

2.2 Subfields of artificial intelligence and robotics ............................................................................. 22

Subfields in artificial intelligence .................................................................................................... 22

Subfields in robotics ....................................................................................................................... 24

2.3 Present and future developments and challenges in AI and robotics ......................................... 27

Artificial intelligence ........................................................................................................................ 27

Robotics ......................................................................................................................................... 293. Present and future applications ......................................................................................................... 31

3.1Artificial intelligence ..................................................................................................................... 31

Transportation ................................................................................................................................ 31

Infrastructure .................................................................................................................................. 32

Healthcare ...................................................................................................................................... 33

Finance and insurance ................................................................................................................... 34

Security .......................................................................................................................................... 36

Retail and marketing ...................................................................................................................... 37Media and entertainment ............................................................................................................... 39

Science .......................................................................................................................................... 40

Education ....................................................................................................................................... 41

Manufacturing ................................................................................................................................ 41

Agriculture ...................................................................................................................................... 42

3.2 Robotics ...................................................................................................................................... 42


4

Transportation ................................................................................................................................ 42Security .......................................................................................................................................... 44

Infrastructure .................................................................................................................................. 45

Healthcare ...................................................................................................................................... 46

Companionship .............................................................................................................................. 48

Industry .......................................................................................................................................... 49

Discovery ........................................................................................................................................ 50

Service ........................................................................................................................................... 51

Environment ................................................................................................................................... 53Agriculture ...................................................................................................................................... 54

4. Social and economic impacts of AI and robotics ............................................................................... 56

4.1Approach ..................................................................................................................................... 56

Preparation ..................................................................................................................................... 57

Impact identification ....................................................................................................................... 57

Consultations ................................................................................................................................. 58

Assessment .................................................................................................................................... 594.2 Social, economic and environmental impacts of AI and robotics: identification .......................... 60

Social impacts ................................................................................................................................ 60

Economic impacts .......................................................................................................................... 62

Environmental impacts ................................................................................................................... 64

Limitations and challenges in impact identification ........................................................................ 65

4.3 Assessment of impacts ............................................................................................................... 66

Category of impacts – AI, robotics or AI and robotics .................................................................... 66

Nature of the impacts ..................................................................................................................... 67Applications and sectors of AI and robotics with the most positive (beneficial) or negative (adverse)

impact (described by interviewees) ................................................................................................ 69

Timing of impacts and starting points (described by interviewees) ................................................ 70

Geographical reach of the impacts ................................................................................................ 72

Effects of impacts on values .......................................................................................................... 73

Parties affected by the social, economic and environmental impacts ............................................ 78

Resilience and vulnerability of the affected parties to the impacts (described by interviewees) .... 82

Costs of the impacts ....................................................................................................................... 83Mitigation measures for negative impacts ...................................................................................... 86

4.4 Conclusions ................................................................................................................................. 87

5. Conclusion ......................................................................................................................................... 89

References ........................................................................................................................................ 90

Annexes .............................................................................................................................................. 102

Annex 1: Interview questions .......................................................................................................... 102


5

Executive summary Thisdeliverable isastate-of-the-artreviewofthefieldsofartificial intelligence(AI)androbotics. Itoffersathoroughanalysisofbothfieldsintermsoftheircentralconcepts,theirhistory,theirpresentandexpectedtechnologiesandapplications,aswellasasocio-economicimpactassessmentofcurrentandexpected impactsof their technologies. Theanalysis isbasedona thorough literature review,interviews,andcommentaryondraftsofthisreportbyfieldexperts.Intheremainderofthissummary,wepresentthecontentofeachofthemainsectionsinthisdeliverable.

Insection1,webrieflyintroducethefieldsofAIandrobotics.Insection2,weofferdefinitionsofAIandrobotics,layoutthescopeofthesefields,andexplaintheircoreideas,concepts,approaches,andmethods. In addition, we describe the most important subfields in AI and robotics, including theparticularconceptsandtechniquesusedinthem,andwediscusspresentandfuturechallengesforAIandroboticsandpotentialfuturedevelopmentsinthesefields.

Our analysis has found that AI can be defined as the science and engineering of machines withcapabilitiesthatareconsideredintelligentbythestandardofhumanintelligence.Here,intelligenceisconceivedofasageneralcognitiveabilityencompassingmorespecificabilities,includingtheabilitiestoreason,solveproblems,plan,comprehendideas,uselanguage,andlearn.OuranalysishasfurtherfoundthatAIresearchcanbedividedintotwomainareasofstudy:“strongAI”and“weakAI”.StrongAIiscapableofcarryingoutanymentaltaskthatcanbecarriedoutbyahumanbeing,andisbelievedtobecapableofhavingcognitivestatessimilartothosefoundinhumanminds.WeakAI,ontheotherhand,canperformingintelligenttasksonlyinspecificdomains(e.g.,thevisualclassificationofdigitalimages).WeakAIresearchhasbecomedominantoverresearchonstrongAIinrecentdecades.

BesidesafocusoneitherstrongAIorweakAI,itwasfoundthatthereexisttwomainparadigmsorapproachesinAIresearch(besidesseveralancillaryones).ThefirstissymbolicAI,whichisatermfora group of methods in AI research that are based on high-level, “symbolic” representations ofproblems, concepts, objects, events, and so on. Symbolic AI approaches have been particularlyinfluentialintheresearchonstrongAIandarebestsuitedforthemodellingofhighercognitivetasks,suchasabstractreasoningandproblemsolving.ThesecondmajorresearchparadigmisconnectionistAI,whichholdsthatmentalphenomenacanbestbedescribedbyinterconnectednetworksofsimpleandoftenuniformunitssimilartothosethatexistinthebiologicalbrain.Connectionistapproaches,inparticular neural networks, have recently become very popular, and are currently being used inmachinelearningapproaches,whereAIsystemsarebeing“taught”torecognisepatternsanddeviserulesthroughprocessingof“trainingdata”andfeedbackonperformance.

ImportantsubfieldsofAIwerefoundtoinclude:knowledgerepresentationandautomatedreasoning,artificialneuralnetworksandmachinelearning,computervision,computeraudition,naturallanguageprocessing, expert systems, data mining, intelligent agent systems and automated planning,evolutionarycomputation.

Forrobotics,ouranalysishasfoundthatthisfieldcanbedefinedasthedesign,developmentanduseofelectro-mechanicalmachineswithsensorsandactuatorsthatcanmoveintheirenvironmentandperformintendedtasksautonomouslyorsemi-autonomously.Autonomy,here,canbedefinedasthecapacity to operate in a real-world environment without external control. Important concepts inroboticsandrobotsincludesensing,actuation,anddecision-makingability.Sensorsareneededsothattherobotcanobtaininformationfromitsenvironment;actuatorsaretheretogivetherobottheabilitytomoveandexertforcesonitsenvironment;andanon-boardcomputationalcapacityisrequiredfortherobottohavesignificantautonomy.


6

Our analysis has further found that there aremanydifferent approaches and areas of research inrobotics.Whereassomeof theresearch focuseson investigatinganddevelopingnewfundamentaltechniques in robotics, alternative ways to design and interact with robots, and new ways tomanufacture robots, other research is geared towards developing robots for very specific kinds ofapplications (e.g., research for consumer-market unmanned aerial vehicles [UAVs]). An importanttrendinroboticsisthatrobotsareincreasinglybeingdevelopedtoresembleandinteractwithhumans.Importantnotionsherearethoseofhumanoidrobotandsocialrobot.Humanoidrobotsarerobotsthat resemblehumanbeings in termsappearance,mechanicsandbehaviour,andsocial robotsarerobotsthatarecapableofinteractingwithhumansthroughsocialbehaviourandadherencetorulesattachedtotheirsocialrole.Differentaspectsoftheinteractionbetweenhumansandrobots,aswellasaspectsofthedesignofrobotsinrelationtosuchinteraction,arestudiedinthefieldofhuman-robotinteraction.

Importantsubfieldsofroboticswerefoundtoinclude:robotmechanics,robotsensing,robotcontrol(including many subareas, such as robot learning, adaptive control, developmental robotics,evolutionaryrobotics,cognitiverobotics,behaviour-basedrobotics, roboticmappingandplanning),robot locomotion, bio-inspired and soft robotics, humanoid robotics,microrobotics, nanorobotics,beam robotics, cloud robotics, swarm robotics, telerobotics, social robotics and human-robotinteraction.

Insection3,wediscussimportanttechnologiesandapplicationsthathavebeendeveloped,ormayinthefuturebedeveloped,withinthefieldsofAIandrobotics.ThetimehorizonsetforinclusionoffutureAIandroboticsapplicationswassetat20yearsfrompresent.Thiswasconsideredneitherapointintimetoofar intothefuture(makingtheanalysistoospeculative),noronetooclosetothepresent(decreasingtheanticipatoryvalueoftheanalysis).

For AI, itwas found that significant present and expected future applications are in the followingapplication domains: transportation, infrastructure, healthcare, finance and insurance, security(militaryandlawenforcement),retailandmarketing,mediaandentertainment,science,education,manufacturingandagriculture. In transportation, significantpresentapplicationsofAIare insmartschedulingfortransportationservices,andexpectedfutureapplicationsareinpredictivemodellingofindividual user characteristics for transportation services. In infrastructure, significant presentapplicationsofAIareinenergymanagement,optimizationanddistribution,andsensingandpredictiveanalysis forwatermanagement, andexpected futureapplicationsare inpredictiveneighbourhoodanalysisforurbanplanningactivities.Inhealthcare,significantapplicationsofAIareinclinicaldecisionsupport, patient monitoring and coaching, preventative medicine and management of healthcaresystems, and expected future applications are in automated image interpretation, personalisedpreventativemedicinedevices,andpersonaliseddiagnosisandtreatmentinvolvingDNAanalysis.Infinance and insurance, significant present applications of AI are in algorithmic trading and high-frequencytrading,automatedfinancialadviceandportfoliomanagement,underwritingforcreditandinsuranceindustries,andinsuranceclaimsprocessing,andexpectedfutureapplicationsareinbigdataanalysisformarketanalysisandautomatedtrading,conversationalsystemsforpersonalfinanceandinsurance,andsocialmediascanningforunderwritinginfinanceandinsurance.Insecurity,significantpresentapplicationsofAIareinautomatedanalysisofcamerafootageforsurveillanceandpredictivepolicing, detectionof financial fraudandevidencegathering frompersonal electronicdevices, andexpected futureapplicationsare inadvancedcyberdefenceandweapons systems,and large-scalesurveillancesystemsusingwide-areaimagery.Inretailandmarketing,significantpresentapplicationsofAIareinrecommendersystemsforpurchasedecisionsandprogrammaticadvertising,andexpectedfutureapplicationsareinintegrationofbiometricdatainretail,cross-platformmarketingandretail


7

practices,visualsearchsystemsforfindingproductsandconversationalsystemsincustomerservice.Finally,inmediaandentertainment,significantpresentapplicationsofAIareintargetedadvertising,smartfiltersandrecommendersystemsforoptimizingandascertaininguserengagement,discoveryofnewtargetaudiences,andnon-playercharactersinvideogames,andexpectedfutureapplicationsarein(co-)creationbycomputersanddevicesofmediaandentertainmentcontent,andmediaandentertainmentfeaturesofadvancedvirtualassistantsinthehome.

Forrobotics,itwasfoundthatsignificantpresentandexpectedfutureapplicationsareinthefollowingapplication domains: transportation, security, infrastructure, healthcare, companionship, industry,discovery,service,environmentandagriculture.Intransportation,significantpresentapplicationsofroboticsarein(semi-)autonomouscars,wheeled(andlegged)cargorobotsandexoskeletalrobotsfortransporting heavy objects, and expected future applications are in unmanned aerial vehicles fordrone-delivery services, and air taxis and pilotless passenger jets. In security, significant presentapplications of robotics are in drones formilitary reconnaissance and target engagement, groundrobots for the handling and destruction of hazardous objects, and robots for search, rescue andrecovery,andexpectedfutureapplicationsare inmilitaryexoskeletons,surveillancedronesfor lawenforcement, and arrest and detainment robots. In healthcare, significant present applications ofroboticsareinsurgicalcollaborativerobots,sociallyassistiverobots(e.g.,forwelcomingpatients)andcare robots for patients and elderly people, and expected future applications are in roboticexoskeletons fordisabledpeople. In thecompanionship domain, significantpresentapplicationsofroboticsareinroboticpetsandsexrobots,whichareexpectedtobecomemorelife-likeinthefuture.Inindustry,significantpresentapplicationsofroboticsareinmaterialhandling,welding,finishingandassembly,intelligentassistdevices,andcollaborativerobots,andexpectedfutureapplicationsareinrobotic exoskeletons. In the service sector, significant present and expected future applications ofroboticsareincookingrobots,deliveryrobots,cleaningrobotsandentertainmentrobots.Finally,indiscovery, significantpresentandexpectedfutureapplicationsof roboticsare inspaceexploration,deep-seaexplorationandinvestigationofhazardousareasonland.

Insection4,wepresentoursocio-economicimpactassessment(SEIA)ofAIandrobotics.SEIAmeansan analysis used to identify and assess the social, economic and environmental impacts of AI androbotics on society. Impacts refer to (potential) changes –whether positive or negative, direct orindirect,inwholeorinpart–causedbyorassociatedwiththetechnologicalfieldunderconsideration.The SEIAexamines social impacts (i.e., those related to society in general, specific groups and theindividual,suchassocialinequalities,protectionofparticulargroups,personaldataandgovernance),economicimpacts(i.e.,thoseassociatedwitheconomiceffectsontheindividual,groupsandsocietyin general, such as employment, the conduct of business and operating costs) and environmentalimpacts (i.e., those related to thenatural environment, such as energyuse,water quality and theclimate).

Thecentral researchquestionof theSEIA is:Whataresomeof themaincurrentandfuturesocial,economicandenvironmentalimpactsofAIandroboticsasreportedbyrecentliteratureandexpertsinthetopic?Toanswerthisquestion,wefollowedamultiple-stepapproachandabidedbyaprescribedmethodology. First, we prepared the SEIA plan. Second, we identified social, economic andenvironmentalimpactsthroughaliteraturereview(andstakeholderinterviews).Third,weconsultedwithexpertsthroughinterviews.Fourth,weassessedtheimpactsbasedonourliteraturereviewandexpertopinions.Fifth,weformulatedconclusionsaboutmitigatingnegativeimpactsonthebasisoftheprecedingsteps.Finally,wereviewedtheresults(withpublicationoftheSEIAtofollow).

WebrieflysummariseheresomeofthehighlightsandfindingsoftheSEIA.TheSEIAidentifiedavarietyofpositive(beneficial)andnegative(adverse)social,economicandenvironmentalimpacts.Basedon


8

theliteratureandinterviews,autonomousweaponssystemswillhavethemostnegativeimpact,whilemedicaldiagnosticapplicationsofAIand/orroboticswillhavethemostpositiveimpacts.Thesectoror fieldwiththemostnegative impactwillbemilitarydefence,whilethemostpositive impactwillcomefromthehealthcaresector.Theliteraturereviewandinterviewsalsoindicatethatmostimpactsoccurnowandwillcontinuetodosointothefuture,withintervieweesofferinginsightsintostartingpointsandtimerangesforimpacts.

Theliteraturereviewandinterviewsrespectivelyhighlightthatthemajorityofsocial,economicandenvironmental impacts are (orwill be) global in reach (i.e., theywill occur throughout theworld).Separately,socialimpactsmostcommonlyaffectthevaluesofwell-being,trustandprivacy.Economicimpacts most commonly affect values such as competiveness, efficiency and productivity andconsumer choice. Environmental integrity stands out as the most commonly affected value ofenvironmentalimpacts.Althoughtheliteraturereviewandinterviewspointedouttheimpactsaffecteveryone (i.e.,allparties), theyalsohighlight specificaffectedparties.Social impactsaffectpartiessuch as the elderly, patients and healthcare providers. Economic impacts affect, in particular,enterprises,workers and consumers. Environmental impacts affect, forexample, energyproviders,societyandscientists.

Ourinterviewsrevealedthatcertainaffectedpartiesaremoreresilientormorevulnerablethanotherstosocial,economicandenvironmentalimpacts.Forexample,intervieweesbelievethatcommunitiesbased on strong unifying values are more resilient to impacts, while poor countries are morevulnerable.Byandlarge,accordingtotheliteratureandinterviews,society(ingeneral)willbearthecosts of negative impacts. To reduce the negative impacts of AI and robotics, the literature andinterviewsproposevariousmitigationmeasures,suchaspolicyandregulatorymeasures,technologicalorindustrymeasuresandsociety-levelmeasures.

Weultimately setout conclusions abouthow tominimisenegative impacts andmaximisepositiveones, drawing from what the literature review and interviews revealed about the impacts. Ourconclusionsfortheshort-termtomedium-term(fiveto10years)areasfollows.Theeffectsofnegativeimpacts withoutmitigation, especially for extremely destructive applications such as autonomousweapons,willpotentiallyefface thepositive impacts fromAIand robotics. If the timingof impactspresentedabove is accurate, then regulationandothermitigationmeasuresneed implementationnowtomitigatenegativeimpactscurrentlyoccurringandprojectedtodosointhefuture.Socialcostsof negative impactsmerit special attention, and all efforts to limit them should begin as soon aspossible.Joblossesandthethreattosocialandeconomicstabilityforworkersneedprioritisationinregulatoryandpolicydiscussions.EducatingindividualsaboutthepositiveimpactsofAIandroboticscould aid in their acceptance and reduce impacts associated with social strife, isolation anddisaffection. However, presenting only positive attributes and impacts of AI and robotics may beperceived as disingenuous or dishonest. Consequently, increasing societal acceptance of AI androboticsshouldalsoincludediscussionsoftheirnegativeimpacts(andmitigationmeasurestakentoreducethem).Ourconclusionaboutthe long-term(20yearsandafter) is:Recognisethatdecisionsoccurring intheshort-andmedium-termswill largelyconstructthestateof theaffairs inthe long-term.Asaruleofthumb,avoidanyapplicationofAIandroboticswhoseexpectedlossesoutweightheprobablebenefitsinthelongrun.


9

List of figures • Figure1:StepsintheSIENNAAIandroboticsSEIA

List of tables • Table1.Listofacronyms/abbreviations• Table2.Glossaryofterms• Table3.Searchtermexamples• Table4.Socialimpactsidentifiedintheliterature• Table5.Socialimpactsidentifiedintheliterature• Table6.Economicimpactsidentifiedintheliteraturereview• Table7.Economicimpactsidentifiedintheinterviews• Table8.Environmentalimpactsidentifiedintheliteraturereview• Table9.Environmentalimpactsidentifiedintheinterviews• Table10.Thenatureofsocialimpactsasdescribedintheliterature• Table11.Thenatureofsocialimpactsasdescribedintheinterviews• Table12.Thenatureofeconomicimpactsasdescribedintheliterature• Table13.Thenatureofeconomicimpactsasdescribedintheinterviews• Table14.Thenatureofenvironmentalimpactsasdescribedintheliterature• Table15.Thenatureofenvironmentalimpactsasdescribedintheinterviews• Table16.Responsesto“Whichimpactscanyouforeseeinthenext5-10years?”• Table17.Responsesto“Whichimpactscanyouforeseeinthenext20years?”• Table18.Socialimpacts–effectsonvalues–asappearingintheliterature• Table19.Socialimpacts–effectsonvalues–asappearingintheinterviews• Table20.Economicimpacts–effectsonvalues–asappearingintheliterature• Table21.Economicimpacts–effectsonvalues–asappearingintheinterviews• Table22.Environmentalimpacts–effectsonvalues–asappearingintheliterature• Table23.Environmentalimpacts–effectsonvalues–asappearingintheinterviews• Table24.Resilienceandvulnerabilityofaffectedparties• Table25.Negativesocialimpacts–cost-bearers–asdescribedintheliterature• Table26.Negativesocialimpacts–cost-bearers–asdescribedintheinterviews• Table27.Negativeeconomicimpacts–cost-bearers–describedintheliterature• Table28.Negativeeconomicimpacts–cost-bearers–describedintheinterviews• Table29.Negativeenvironmentalimpacts–cost-bearers–fromtheliteraturereview• Table30.Negativeenvironmentalimpacts–cost-bearers–fromtheinterviews


10

List of acronyms/abbreviations Abbreviation ExplanationAI ArtificialintelligenceANN ArtificialneuralnetworksAUV AutonomousunderwatervehicleBEAM Biology,electronics,aestheticsandmechanicsCAD ComputeraideddesignCobot CollaborativerobotDoA DescriptionofActionEC EuropeanCommissionGPS GlobalPositioningSystemIoT InternetofThingsITS IntelligenttutoringsystemMAV MicroaerialvehicleMEMS MicroelectromechanicalsystemNLP NaturallanguageprocessingNPC Non-playercharacterR&D ResearchanddevelopmentSAR SociallyassistiverobotSEIA Socio-economicimpactassessmentUAV UnmannedaerialvehicleWP Workpackage

Table1:Listofacronyms/abbreviations

Glossary of terms Term ExplanationActuator Adevicemoduleorsubsystemforperformingactionsinanenvironment.Algorithm “[A]precisely-definedsequenceofrulestellinghowtoproducespecified

outputinformationfromgiveninputinformationinafinitenumberofsteps.”1

Artificialintelligence Thescienceandengineeringofmachineswithcapabilitiesthatareconsideredintelligent(i.e.,intelligentbythestandardofhumanintelligence).

Artificialneuralnetwork

Aninterconnectednetworkofsimpleandoftenuniformunitssimilartothosethatexistinthebiologicalbrain,whichcanbeimplementedinintelligentcomputingsystems.

Autonomy “[A]capacitytooperateinareal-worldenvironmentwithoutanyformofexternalcontrol,oncethemachineisactivatedandatleastinsomeareasofoperation,forextendedperiodsoftime.”2

Bigdata Extremelyvoluminousdatasetsthatrequirespecialistcomputationalmethodstouncoverpatterns,associationsandtrendsinthem.

Computervision AnapplicationofAIthatgivesacomputersystemthecapacitytoacquire,processandanalyse(numericalorsymbolic)informationaboutthecontentpresentedindigitalimagery.

1Knuth.Donald.“ComputerScienceandItsRelationtoMathematics,”AmericanMathematicalMonthly,Vol.81,No.4,1974,pp.323-343.2Lin,Patrick,KeithAbneyandGeorgeA.Bekey,“CurrentTrendsinRobotics:TechnologyandEthics,”RobotEthics:TheEthicalandSocialImplicationsofRobotics,MITPress,2012.


11

Term ExplanationConnectionistAI AgroupofmethodsinAIresearchthatutiliseinterconnectednetworks

ofsimpleandoftenuniformunitssimilartothosethatexistinthebiologicalbrain.

Datamining Theprocessofdiscoveringpatternsinlargedatasetsinvolvingdatabasesystems,statisticalanalysis,andintelligentmethodssuchasmachinelearning.

Deeplearning Anapproachtomachinelearningthatappliesartificialneuralnetworkswithhiddenlayersandthebackpropagationmethod,incombinationwithpowerfulcomputersystemsandvoluminoustrainingdata.

Drone Synonymouswith“unmannedaerialvehicle”;anaircraft without a human pilot aboard.

Expertsystem Acomputersystemthatcanmimicahumanexpert’sdecision-makingabilitywithinaparticularfieldbyreasoningthroughalargeamountoffield-specificknowledgecontainedinadatabase.

Humanoidrobot Arobotthatresemblesahumanbeingintermsofappearanceand/orbehaviour.

Impact Apotentialchange–whetherpositiveornegative,directorindirect,inwholeorinpart–causedbyorassociatedwiththetechnologicalfieldunderconsideration.

Intelligence Ageneralcognitiveabilityencompassingseveralmorespecificabilities,includingtheabilitiestoreason,solveproblems,plan,conceptualise,uselanguage,andlearn.

Intelligentagent Anartificiallycreated,autonomousentitythatcanperceiveitsenvironmentbymeansofsensors,actuponthisenvironmentthroughtheuseofactuators,anddirectitsactivitiestowardsreachinggoals.

InternetofThings(IoT)

TheinterconnectionviatheInternetofobjectsinthephysicalworld–devices,vehicles,persons,buildingsandotheritems–allowingthemtosendandreceivedata.

Machinelearning AsetofapproacheswithinAIwherestatisticaltechniquesanddataareusedto“teach”computersystemshowtoperformparticulartasks,withoutthesesystemsbeingexplicitlyprogrammedtodoso.

Naturallanguageprocessing

AnapplicationofAIthatgivesacomputersystemthecapacitytounderstandhumanlanguageinwrittenorspokenform.

Robotcontrolsystem Asystemthatusesarobot’ssensordatatocalculateandsendappropriatesignalstotherobot’sactuators.

Robotics Thefieldofscienceandengineeringthatdealswiththedesign,construction,operation,andapplicationofrobots.

Robot Electro-mechanicalmachineswithsensorsandactuatorsthatcanmove,eitherentirelyorapartoftheirconstruction,withintheirenvironmentandperformintendedtasksautonomouslyorsemi-autonomously.

Socio-economicimpactassessment

Theanalysisusedtoidentifyandassessthesocial,economicandenvironmentalimpactsofAIandroboticsonsociety.

Sensor Adevice,moduleorsubsystemfordetecting(andsendinginformationabout)eventsorchangesinanenvironment.

Socialrobot Arobotthatiscapableofinteractingwithhumansthroughsocialbehaviourandadherencetorulesattachedtotheirsocialrole.

SymbolicAI AgroupofmethodsinAIresearchthatarebasedonhigh-level,“symbolic”representationsofproblems,concepts,objects,events,etc.

Table2:Glossaryofterms


12

1. Introduction Thisdeliverable isastate-of-the-artreviewofthefieldsofartificial intelligence(AI)androbotics. Itoffersathoroughanalysisofbothfieldsintermsoftheircentralconcepts,theirhistory,theirpresentand anticipated technologies and applications, aswell as a socio-economic impact assessment (ofpresentandexpectedimpacts)oftheirtechnologies.Theanalysis isbasedonathoroughliteraturereviewandcommentaryondraftsofthisreportbyfieldexperts.Intheremainderofthisintroduction,we offer brief introductions to the fields of AI and robotics, and then present an outline for theremainderofthisdeliverable.

Beforewecontinue,letusbrieflyconsidertherelationbetweenAIandrobotics.AIisoftenreferredtoasthescienceandengineeringofmachineswithcapabilitiesthatareconsideredintelligent,thatis,intelligent by the standard of human intelligence. Robotics can be defined as the science andengineering of programmable electro-mechanical machines that can perform human tasksautonomouslyorsemi-autonomously.Asisapparentfromthesedefinitions,thereexistsadegreeofoverlap between AI and robotics. Artificially intelligent machines may or may not be physicallyembodiedand(semi-)autonomous(i.e.,theyarerobots);androbotsmayormaynotuseAItechniquesas a part of their control systems. The fields of AI and robotics come together in the science andengineeringofartificiallyintelligentrobots.Wediscusssuchrobotsinthepartsofthisdeliverablethatarefocusedonrobotics.

1.1 Artificial intelligence

Artificial intelligence is a field of computer science that emerged in the 1950s and has roots inphilosophy, mathematics, computation, biology and psychology. Its subfields include machinelearning,neuralnetworks,evolutionarycomputation,AIforrobotics,expertsystems,planning,speechprocessing,naturallanguageprocessing,computervisionandothers.TheprimaryaimsofAIresearcharetosystematicallystudythephenomenonofintelligence,andtodevelopusefulprogramsandtoolsthat can execute tasks normally requiring intelligence. Here, the notion of intelligence is usuallyconceivedofasageneralcognitiveabilityencompassingseveralmorespecificabilities,includingtheabilities to reason, solve problems, plan, conceptualise, use language, and learn. Research oftenfocusesonaspecificabilityanddevelopingcomputerprogramsthatarecapableofperforminglimitedtasksinvolvingthisability.

As of present, AI capabilities are wide-ranging and include, for example, interpretation of humanspeech,interpretationofimageandvideodataforobjectavoidanceandface-recognition,controlofrobotandautonomousvehiclebehaviour,andintelligentroutingincontentdeliverynetworks.

Major applicationsofAI technologyare in transportation, education, finance, industry, healthcare,marketing,management,telecommunications,entertainmentanddefence,amongstotherfields.Forinstance,AItechniquesarecurrentlybeingusedincarswithdriverassistfeatures(e.g.,self-parking,advancedcruisecontrols),inintelligenttutoringsystemsinhighereducation,inalgorithmictradinginfinancialmarkets,inindustrialrobots,inclinicaldecisionsupportsystemsformedicaldiagnosis,andinvirtualassistantsonmobilephones.Inthefuture,suchapplicationsareexpectedtobecomemorecommonplaceandadditionalapplicationsmayemerge,includingfullyautonomousroadvehicles,AI-basedsystemsforimprovedcybersecurity,andchatbotsforcomplexcustomerinteractions.

AItechnologyisexpectedtohave,andtosomedegreeisalreadyhaving,anumberofprofoundsocial,economic,andethicalimpacts.Theseinclude:increasesineconomicefficiency,economicoutput,andsocietal wealth; decreases in the time spent on repetitive and burdensome tasks for individuals;potential increases in the quality of and time spent on leisure activities; the transformation of


13

workplacesandthenatureofwork;thepotentialforrisingunemploymentacrosslow-skilledlaboursectorsandprofessionalsectors;thepotentialforrisingsocialinequalitiesandpolarisationasaresultof unequal distribution of benefits of AI technology; the potential for resistance to the use of AItechnology; the potential diminishment of individual autonomy due increased reliance on AItechnology;andthepotentialforabuse,misuseanddual-useofAItechnology.

ThefieldofAIhashadatumultuoushistory.Ithasexperiencedseveralwavesofoptimism,whichwerefollowedbydisappointmentandlossoffundingduringwhatareknownasthe“AIwinters”intheearly1970s,late1980s,andearly1990s.3Inthetwenty-firstcentury,AIexperiencedaresurgencefollowingadvancesincomputingpower,massdatastorageandavailability,andtheoreticalunderstanding.ThisallowedAItechniquestobecomeanessentialpartofthetechnologyindustry,helpingtosolvemanychallengingproblemsincomputerscience.4

Inrecentdecades,therehasbeenashiftinfocusfromAIasasciencethatexaminesthephenomenonofintelligencetoAIasanengineeringdisciplinethatdevelopspracticalprogramsandtoolsusingAItechniques.5TheinitialgoalofAIresearchhadbeentoconstructasystemwithcognitiveabilitiesakintothoseofahumanbeing.EarlyresearchersinAIoftenheldthebelief,basedonadvancesmadeindigitalcomputingandkeybreakthroughsinthefieldsofinformationtheoryandformallogic,thatthisgoalwouldbereachedwithinjustafewdecades.Theycontendedthatintelligentcomputersystemscouldbeusedtomodelhumanthoughtprocessesandthuscouldgeneratecrucial insights intotheworkingsofhumancognition.Thisviewformedthebasisofthe“strongAI”approachinAI6andisalsoknownasthecognitivesimulationapproach.Astrongartificialintelligencecansuccessfullyperformanyintellectualtaskahumancanandisbelievedtopossesshuman-likecognitivestates.

Whereasmany researchers inAIembraced the strongAIapproach,othershaddoubtswhether itspromisescouldbeeasilyfulfilled,andquestioned,forexample,whetherahumancognitioncouldreallybe simulatedwith computational techniquesbeingused.Many in the latter groupwantedonly todevelop computer programs capable of performing intelligent tasks in specific domains. Theirapproachhasbeencalled“weakAI”.TheweakAIapproachhasbecomedominantinrecentdecades.TheapproachhascombinedwithotherfieldsbeyondAI,integratingitsmethodsandtechniquesintosuch areas as robotics, computer networking, data mining, computer vision, human-computerinteraction,agenttechnology,embeddedsystemsandubiquitouscomputing.

1.2 Robotics

Roboticsisaninterdisciplinaryfieldofengineeringandsciencethatemergedinthe1950sandinvolves(andoverlapswith)mechanicalengineering,electricalengineering,computerscience,mechatronics,nanotechnology,andbioengineering.Itconcernsthedesign,developmentanduseofrobots,includingthecomputersystemsneededfortheircontrolandtheoutputandprocessingofsensoryinformation.Itssubfieldsaremanyandincludethosethatstudymorefundamentalcharacteristicsofrobots(e.g.,robot control, robot locomotion, microrobotics) and those that study more application-orientedaspectsofrobots(e.g.,aerialrobotics,underwaterrobotics,industrialrobotics).

3Guice,Jon,“DesigningtheFuture:TheCultureofNewTrendsinScienceandTechnology,”ResearchPolicy,Vol.28,No.1,1999,pp.81–98.4Russell,Stuart,andPeterNorvig,ArtificialIntelligence:AModernApproach,3rdEdition,PrenticeHall,2010.5Brey,Phlip,andJohnnyHartzSøraker.“PhilosophyofComputingandInformationTechnology,”inA.Meijers(Ed.),PhilosophyofTechnologyandEngineeringSciences(pp.1341-1408),VolIX,inD.Gabbay,P.ThagardandJ.Woods(Eds.),HandbookofthePhilosophyofScience,Elsevier,Amsterdam,2009.6Searle,JohnR.,Minds,brains,andprograms,BehavioralandBrainSciences,Vol.3,No.3,1980,pp.417–457.


14

Currently, thereexistmanydifferentkindsof robotswithgreatlyvaryingphysicalappearancesandcapabilities. Robots range from the very small (e.g., insect-sized drones) to the very large (e.g.,automotive industrial robots).Theymaybe fullyautonomous, semi-autonomous, tele-operated,orattachedtoanddirectlyoperatedbythehumanbody(e.g.,prostheticdevices).Theycanhaveavarietyof sensors, including cameras, microphones, thermometers, infrared and vibration sensors.Furthermore,theycanhaveavarietyofactuators,includingspeakers,displays,mechanicalarmsandgrippers, guns, andmechanisms for locomotion (includingwalking, driving, rolling, swimming, andflying).Therearerobotsthatcanperformcomplextasksthatinvolveadvancedartificialintelligencetechniques(e.g.,semi-autonomouscars),androbotsthatperformrelativelysimple,repetitivetasksthroughalimitedsetofpre-programmedactions(e.g.,packagingrobots).Finally,therearerobotsthatresemble (tovaryingdegrees)humansoranimals inappearance,mechanics,andbehaviour. In thenear future, robots are expected to be able to perform tasks that are increasingly complex (e.g.,advancedobjectdetectionandavoidance in fully autonomousvehicles)due toadvances inAI andcomputingpower,andtheyareprojectedtobecomesmallerinsizeandmoreenergyefficient.

Major applications of robots are in transportation, industry, healthcare, education, entertainment,spaceexploration,defence,retail,companionship,housekeepingandotherareas.Robotsareoftenusedfortasksthatarerepetitive,dirtyand/ordangerous.Forexample,articulatedroboticarmsareusedinindustryforapplicationssuchaswelding,materialhandling,andpainting.Indomesticsettings,roboticvacuumcleaners,poolcleaners,guttercleanersarebeingused.Surgeryrobotsandcarerobots(e.g.,liftingrobots)havefounduseinhealthcare.Inmilitarysettings,specializedbombdisposalrobotsandreconnaissancedronesarebeingused.Semi-autonomousspacecraftandlandingvehiclesareusedin spaceexploration. In the future,manynewapplicationsof robotsare likely toemerge,possiblyincluding fully autonomous taxis, fully autonomous policing/surveillance robots, fully autonomouslethalmilitaryrobots,roboticexoskeletons7inindustryandconversationalcarerobotsinhealthcare.

Similar toAI technology, robotsareexpected tohaveanumberofprofoundsocial,economic,andethicalimpacts.Theseinclude:increasesineconomicefficiency,economicoutput,andsocietalwealth;decreasesinthetimespentonrepetitiveandburdensometasksforindividuals;potentialincreasesinthequalityofandtimespentonleisureactivities;thetransformationofworkplacesandthenatureofwork;thepotentialforrisingunemploymentacrosslow-skilledlaboursectorsandprofessionalsectors;thepotentialforrisingsocialinequalitiesandpolarisationasaresultofunequaldistributionofbenefitsofrobots;thepotentialforresistancetotheuseofrobots;thepotentialdiminishmentof(experienced)privacyduerobotsurveillance,tothepotentialdiminishmentofindividualautonomydueincreasedrelianceonrobots;andthepotentialforabuse,misuseanddual-useofrobots.

LikethefieldofAI,themodernfieldofroboticshashadacomparativelyshorthistory.Whiletheveryideaofcreatingautonomouslyoperatingmachinesdatesbacktoclassicaltimes,researchintothebasictechnicalfunctionalityandpossibleapplicationsofroboticsystemsdidnotsubstantiallygrowuntilthemiddleofthetwentiethcentury.Itwasonlyin1948thattheprinciplesofcybernetics,asformulatedbyNorbertWiener,laidthegroundworkforpracticalrobotics.Thefirstrobotsemerginginthe1950sand1960swerefairlyrudimentaryvehicleswithsimplebehaviourguidedbylightsensors,anddigitallyoperatedandprogrammableroboticarmswithlimiteddegreesoffreedom.

During the 1970s and 1980s, research into robotics expanded greatly andmore advanced robotsemerged.Japaneseroboticsresearchatbothacademicinstitutionsandcompaniesrosetoprominenceduringthisperiod.Notablewasthedevelopmentofthefirstmobilerobots(e.g.,ShakeytheRobot)

7Ahuman-wearablemachinethatallowsforlimbmovementswithincreasedstrengthandendurance.


15

thatwerecapableofreasoningabouttheirsurroundingsthroughtheintegrationofAIandmultiplesensorinputs,includingcameras,laserrangefinders,andcollisionsensors.Thefirstfull-scalehumanoidrobots and androids were also introduced, which could walk in a rudimentary fashion, grip andtransportobjectswithartificiallimbs,measuredistancesanddirectionstoobjectsandcommunicatewithhumansinverybasicways.Furthermore,industrialrobotsfirstsawsignificantuseinindustryandbecamemoresophisticated,withmoredegreesoffreedomandthecapacityforveryfinemovements.

From the 1990s onwards, robots diversified greatly and became more sophisticated in terms ofmechanics and intelligence. New research fields opened up, including evolutionary robotics, bio-inspiredrobotics,andBEAMrobotics.Fromapredominantly industrial focus, roboticsexpandedtotakeon the challenges of the humanworld in healthcare, services, entertainment, education, andothersectors.Themostadvancedhumanoidrobotshaveimprovedtothepointwheretheycanwalk,run,communicatewithhumans,analysetheirenvironment,recognisefacesandvoices,andinteractwiththeirenvironment–albeitnotyetatthelevelofhumans.Someintelligentrobots,suchasSony’sAibo pet dog and iRobot’s Roomba vacuum cleaner, were created for private use in the home.Autonomousroadvehicleshaveenteredastagewheretheybegantobetestedinsmallnumbersonpublicroads,andtele-operatedandsemi-autonomousrobots(notablyunmannedaerialvehicles)haveseen wide use in the military environment. Finally, the International Federation of Robotics hasestimatedthatin2016therewere1.2millionindustrialrobotsintheworld,andthatbytheyear2019thisnumberwillhaveincreasedto2.6million.8

1.3 Structure of the report

Inthefollowingsections,weofferabroadanalysisofthestate-of-the-artinAIandrobotics.Insectiontwo,wewilldefinethefieldsofAIandrobotics,whichwill includeademarcationanddefinitionofthesefields,explanationofcore ideas,descriptionof theirsubfields,anddiscussionofpresentandfuturechallenges.Insectionthree,wewilldeliverasystematicdescriptionofproductsandapplicationsthathavebeendevelopedorareexpectedtosoonbedevelopedwithinthefields.Insectionfour,wewilldeliveran initial socio-economic impactassessmentonAIand robotics. Finally, in five,wewillconcludewithasummaryandrecommendationsforfurtherstudy.

1.4 Scope and limitations

Thisreport laysthegroundworkforfurtherSIENNAworkontheethicalandlegalaspectsofAIandroboticstechnologies,productsandapplications.Assuch,itprovidesnoconclusionsorcommentaryonanyethicalandlegal issueswithcertainAIandroboticsapplicationsthatmaybecomeapparentthroughthisanalysis.

Further,inordertoprovidethemostusefulinputforthedevelopmentofpracticalrecommendationsinlaterSIENNAreports,itishelpfultosetalimitontheinclusionofpotentialdevelopmentsinAIandroboticsthatmayonlyoccuroverlargertimescales.Intheanalysisofpotentialfuturedevelopmentsin AI and robotics research and applications, we therefore restrict ourselves to discussingdevelopmentsthatarereasonablypossiblewithinapproximatelytwentyyearsfornow,withthemostemphasisputondevelopmentsfivetotenyearsfromnow.Weconsideratimehorizonoftwentyyearstobeneitherapointintimetoofarintothefuture(makingtheanalysistoospeculative),noronethatistooclosetothepresent(decreasingtheanticipatoryvalueoftheanalysis).

8InternationalFederationofRobotics.“WorldRoboticsReport2016,”IFR,2016.https://ifr.org/news/world-robotics-report-2016


16

2. Defining the field It is important to have working definitions and informative descriptions of the fields of artificialintelligenceandroboticspriortolayingoutthelandscapeofcurrentandpotentialfutureapplicationsoftechnologiesinthem.Inthissection,wethereforeofferdefinitionsofAIandrobotics,layoutthescopeof these fields,andexplaintheircore ideas,concepts,approaches,andmethods (subsection2.1);wedescribethemostimportantsubfieldsinAIandrobotics,includingtheparticularconceptsandtechniquesused in them(subsection2.2);andwediscusspresentandfuturechallenges forAIandroboticsandpotentialfuturedevelopmentsinthesefields(subsection2.3).

2.1 Definitions of artificial intelligence and robotics

In this subsection, we define and demarcate the fields of AI and robotics, and explain their keyconcepts,approaches,andmethods.WebeginbyofferingadefinitionanddescriptionforAI,andthenproceedtodothesameforrobotics.

Definition of artificial intelligence

In the introductionof this report,artificial intelligence isdefinedas thescienceandengineeringofmachineswithcapabilitiesthatareconsideredintelligentbythestandardofhumanintelligence.Here,intelligenceisconceivedofasageneralcognitiveabilityencompassingmorespecificabilities,includingtheabilitiestoreason,solveproblems,plan,comprehendideas,uselanguage,andlearn.

AlthoughourdefinitionofAIisquitegeneral,itisnotaconsensusdefinition.Overtheyears,alargenumberofdefinitionshavebeenproposed.Themajorityofthesecanbecategorisedintofourgroupsonthebasisoftwocriteria.9Afirstdistinctioncanbemadebetweendefinitionsthatemphasisetheability of AI systems to think and reason intelligently and those that emphasise the capacity of AIsystemstoactintelligently.DefinitionsoftheformerkindcanbeseeninthelightofAI’shistoricalaimtogenerateabetterunderstandingof theconceptofhuman intelligence (i.e.,humanthinkingandreasoning), while definitions of the latter kind can be seen as focussing more on the practicalapplicationofAItechniques.AseconddistinctioncanbemadebetweendefinitionssuggestingthatAIsystemsareaimedatemulatinghumanintelligencetosomedegree,anddefinitionssuggestingthatAIsystemssimplythinkorbehaveinaccordancewithanidealconceptofintelligence,whichwemaycallrationally.10Here,too,definitionsoftheformerkindmaybesomewhatassociatedwithAI’saimofcreatingabetterunderstandingoftheconceptofhumanintelligence,whiledefinitionsofthelatterkindaremorefocusedonthepracticalapplicationofAI.

Ourdefinitionofartificial intelligenceisphrasedsoastorefertoAIasboththeabilitytothinkandreasonintelligentlyandtheabilitytoactintelligently.ThiswayitcoversbothofthetwomainaimsofAIresearch:generatingabetterunderstandingofhumanintelligence,anddevelopingtoolsthatcanautomateintelligentbehaviour.Furthermore,ourdefinitionreferstoAIsystems’capacityforhuman-likeintelligence(andnotjustthecapacityforrationalthoughtorbehaviour),sincespecificfacetsofhuman intelligence (e.g., planning, learning, and communicating) are the yardsticks AI systems’intelligenceistypicallymeasuredby.

9Russell,Stuart,andPeterNorvig,op.cit.,2010.10Asystemisrationalifitchoosestoperformtheactionwiththeoptimalexpectedoutcomeforitselffromamongallfeasibleactions.


17

Core concepts, approaches and methods in artificial intelligence

TheprimaryaimsofAIresearchare,firstly,tosystematicallystudythephenomenonofintelligence,andsecondly,todevelopprogramsandtoolsthatcanautomateintelligentbehaviour,whichincludesinformationgathering,detecting,planning, learning,communicating,andmanipulating. In termsofresearchoutput,thesecondaimhasbeenreceivingincreasedattentionofthepastdecades.

Artificial intelligence can be broadly divided into strong AI andweak AI.11 Strong AI is capable ofcarryingoutanymentaltaskthatcanbecarriedoutbyahumanbeing,andisbelievedtobecapableofhavingcognitivestatessimilartothosefoundinhumanminds.Onsomephilosophicalaccounts,itisevenheldtoexperienceconsciousness(artificialconsciousness).StrongAIistheobjectofstudyofsomeartificialintelligenceresearchandacommontopicinsciencefictionandforesightstudiesonAI.Onepopularapproach inAIresearchforachievingstrongAIaction iswholebrainemulation.12Thisapproachreliesonbuildingalow-levelbrainmodelbyscanningandmappingabiologicalbrainindetailandcopyingitsstateintoacomputersystem.Thissystemthenrunsasimulationmodelsofaithfultothebiologicalbrainthatitwillbehaveinessentiallythesamewayasthebrain,andforallpracticalpurposes,beindistinguishablefromit.ThepromisesofstrongAIandwholebrainsimulationarebuilton a belief in the computational theory of mind, which is the view that the human mind is aninformationprocessingsystemandthatthinkingequalscomputation.13

Totestwhetheracomputersystemisasintelligentasahumanbeing,theTuringtestwasdeveloped.14Developedin1950byAlanTuring,thistestwasbuiltonthepremisethatamachinemustbeconsideredintelligentifitcanfoolhumanbeingsintothinkingitisahuman.Duringthetest,ahumanbeingandacomputerareplacedbehindascreen,andanevaluatorwhoisontheothersideofthescreen(andwhothuscannotseeboth)thenposesquestionstobothsoastoascertainwhichoneofthetwoisthehumanbeing.Iftheevaluatorisunabletotellwithinareasonableamountoftimewhichoneisthehumanbeing,thenthecomputerissaidtohavepassedthetestandtopossessgeneralintelligence.TheTuringtestisstillfrequentlyreferencedasatestforgeneralintelligence,buthasnotbeenwithoutcriticism.15

ComparedtostrongAI,weakAI’sgoalshavebeenmoremodestandpractical.ThefocusofweakAIresearch is on developing computer programs that are capable of performing intelligent tasks inspecific domains.Weak AI research has become dominant over research on strong AI research inrecentdecades.Goodprogresshasrecentlybeenmadeasaresultofgrowthincomputerprocessingpower,availabilityof largevolumesofdata,anddevelopmentofmachine learningtechniques.TheweakAIapproachhasmergedwithotherfieldsbeyondAI,integratingitsmethodsandtechniquesinto

11Overthelastfewyears,asimilardistinctionhasbeeninuseamongAIscholarsbetween“artificialgeneralintelligence”(AGI)and“weakAI”/”narrowAI”.AGIisconsidereddistinctfromstrongAIinthatitisfocusedonlyonapparentintelligentbehaviourthatisequivalentindepthandscopetohumanperformance,anddoesnotsuggestthatactualthinkingorconsciousnessisinvolved.12Sandberg,Anders,andNickBoström.“WholeBrainEmulation:ARoadmap,”TechnicalReport#2008-3,FutureofHumanityInstitute,OxfordUniversity,2008.http://www.fhi.ox.ac.uk/Reports/2008-3.pdf13Pylyshyn,ZenonW.,ComputationandCognition:TowardaFoundationforCognitiveScience,MITPress,Cambridge,MA,1989.14Turing,Alan.“ComputingMachineryandIntelligence,”Mind,Vol.49,1950,pp.433-460.15Moor,James,TheTuringTest:TheElusiveStandardofArtificialIntelligence,KluwerAcademicPublishers,Dordrecht,2003.


18

suchareasasrobotics,statistics,appliedmathematics,computernetworking,datamining,computervision,human-computerinteraction,artificialagents,embeddedsystemsandubiquitouscomputing.

BesidesthefocusoneitherstrongAIorweakAI,thereexisttwomainresearchparadigmsinAI.ThefirstissymbolicAI,whichisatermforagroupofmethodsinAIresearchthatarebasedonhigh-level,“symbolic”(i.e.,abstractandhuman-readable)representationsofproblems,concepts,objects,events,andsoon.Fromaboutthemid-1950suntilthelate1980s,symbolicAIwasthedominantparadigminAIresearch.16Itsessentialclaimisthatintelligence,bothinhumansandinmachines,isamatterofmanipulatingsymbols through theuseof formal rules,anassumptiondefinedbyAllenNewellandHerbert Simon as the “physical symbol systems hypothesis”.17 Symbolic AI approaches have beenparticularly influential in the researchon strongAIandarebest suited for themodellingofhighercognitivetasks,suchasabstractreasoningandproblemsolving.

ThesecondmajorresearchparadigmisconnectionistAI,whichemergedinthelate1980s.18Rejectingthe view that intelligent behaviour is the result of symbol manipulation through formal rules,connectionismholdsthatmentalphenomenacanbedescribedbyinterconnectednetworksofsimpleandoftenuniformunitssimilartothosethatexistinthebiologicalbrain.19Connectionistapproaches,inparticularneuralnetworks,haverecentlybecomeverypopular.Theyarecurrentlybeingused inmachinelearningapproaches,whereAIsystemsarebeing“taught”torecognisepatternsanddeviserulesthroughprocessingof“trainingdata”andfeedbackonperformance.Theresultofdevelopmentsin machine learning is that AI systems can be self-learning, be more adaptive, and have greaterautonomy.Connectionistapproacheshaveturnedouttobeextremelygoodatcarryingoutparticulartypes of intelligent tasks, such as pattern recognition, categorization, and the coordination ofbehaviour.

ImportantnotionsinapplicationsofAIarethoseofintelligentagentandexpertsystem.Anintelligentagentisanautonomousdecision-makerthatcansenseandact.Moreaccurately,itcanbedefinedasanautonomous,artificiallycreatedentitythatcanperceiveitsenvironmentbymeansofsensors,actuponthisenvironmentthroughtheuseofactuators,anddirectitsactivitiestowardsreachinggoals.Intelligent agents have found widespread use in the automation of processes in softwareenvironments(assoftwareagents),inrobotsandinotherdevicesandmachines.Anexpertsystemisacomputersystemthatcanmimic(andsometimesoutdo)ahumanexpert’sdecision-makingabilitybyreasoningthroughanextensivebodyof(human)knowledgerepresentedinadatabase.Itcanbeusedbyhumanoperator tohelp solvecomplexproblems, suchas in themedicaldiagnosisof rarediseases.ExpertsystemsareoneofthefirstformsofAIsoftwarethatweretrulysuccessful.20

CloselyrelatedtothescientificfieldofAIarethephilosophyofAIandtheethicsofAI.Importantideasandconceptsinthelatterfieldsthatareofsignificantrelevancetotheformerincludethepossibilityandnatureofartificialconsciousnessandsuperintelligence,amongstothers.ArtificialconsciousnessisafieldofresearchthatquestionsthepossibilityandtheorisesaspectsofemulatinghumanconsciousexperiencesbyAIsystems.Asuperintelligenceisahypotheticalintelligentagentorproblem-solvingsystem that possesses intelligence far surpassing that of the brightest human minds. A

16Haugeland,John,ArtificialIntelligence,theVeryIdea,MITPress,Cambridge,1987.17Newell,Allen,andHerbertA.Simon.“ComputerScienceasEmpiricalInquiry:SymbolsandSearch,”CommunicationsoftheACM,Vol.19,No.3,1976,pp.113–126.18BreyandSøraker,op.cit.,2009.19Medler,DavidA.“ABriefHistoryofConnectionism,”NeuralComputingSurveys.Vol.1,1998,pp.61–101.http://www.blutner.de/NeuralNets/Texts/Medler.pdf20Russell,Stuart.,andPeterNorvig,op.cit.,2010.


19

superintelligencemayresultfromtherecursiveself-improvementbyafuturestrongAI(or,artificialgeneral intelligence), and is associated with runaway technological growth leading to a so-calledtechnologicalsingularity.

Definition of robotics

Intheintroductionofthisreport,roboticsisdefinedasthefieldofscienceandengineeringthatdealswiththedesign,construction,operation,andapplicationofelectro-mechanicalmachineswithsensorsandactuatorsthatcanmove(eitherentirelyorapartoftheirconstruction)intheirenvironmentandperform intended tasks autonomously or semi-autonomously. This, however, is not a commonlyagreeddefinitionforrobotics.Therecurrentlystillexistsalackofconsensusamongroboticistsonhowtheobjectoftheircraftshouldbedefined.

Thedefinitionofroboticsdependsmostonhowwedefinearobot.Intheroboticsliterature,thereexistanumberofdefinitionsforrobot,withmanybeingproblematictosomedegree.Forexample,arobothasoccasionallybeendefinedas“acomputerwithsensorsandactuatorsthatallowittointeractwiththeexternalworld”,21oras“anymachinethatdoesworkon itsown,automatically,after it isprogrammedbyhumans”.22Under suchdefinitions, devices like thermostats, coffeemachines andsmart speakers – not tomention personal computers –may qualify as robots, an implication fewroboticistswoulddefend.Forthisreason,agooddefinitionofroboticsmoreshouldbepreciseandbeabletodifferentiaterobotsfromcomputersandotherelectronicorelectromechanicaldevices.

The ISO (the International Organization for Standardization) defines a robot as an “actuatedmechanism programmable in two or more axes with a degree of autonomy, moving within itsenvironment, to perform intended tasks”.23 Another definition, by the Robot Institute of America,holds that a robot is “a reprogrammable,multifunctionalmanipulator designed tomovematerial,parts, tools,or specializeddevices throughvariousprogrammedmotions for theperformanceofavarietyoftasks”.24Bothdefinitionsunderscorethecapabilityofrobotstomoveintheirenvironment.However, it is not immediately clear that robots need to be programmable – at least in theconventionalsenseofthenotionassociatedwithcodingalgorithms–sincerobots’controlsystemscanemploysomeformofAI that isnotexplicitlyprogrammed. Inaddition, therequirement in the ISOdefinition that robotsbeprogrammable “in twoormoreaxes”wouldexclude so-called single axisrobotsthatareoftenusedinindustry.

Philosophicalapproachestendtodescriberobotsasmachinesthat“sense”,“think”(or“decide”)and“act”.25Sensorsareneededsothat therobotcanobtain information from itsenvironment;anon-boardcomputationalcapacityisrequiredfortherobottohavesignificantautonomy;andactuatorsaretheretogivetherobottheabilitytomoveandexertforcesonitsenvironment.Thequestionofwhethertheabilitiestosense,think,oractcanaccuratelybeascribedtoamachineisstillamatterofsignificant contention. Thismatteraside, a crucial element in thephilosophicalunderstandingof a21Lin,Patrick,KeithAbneyandGeorgeA.Bekey,op.cit.,2012.22ElectroTrainingandServicingInstitute,“RoboticsTechnology,”ETASI,n.d.https://etasieducation.com/course/robotics-technology/23InternationalOrganizationforStandardization,“ISO8373:2012(en):Robotsandroboticdevices—Vocabulary,”n.d.https://www.iso.org/obp/ui/#iso:std:iso:8373:ed-2:v1:en24RobotInstituteofAmerica.NBS/RIAroboticsresearchworkshop:ProceedingsoftheNBS/RIAWorkshoponRoboticResearchheldatGaithersburg,MD.,1979.25Asaro,Peter.“HowJustCouldaRobotWarBe?,”inAdamBriggle,KatinkaWaelbersandPhilipBrey(eds.),CurrentIssuesinComputingAndPhilosophy,IOSPress,Amsterdam,2008,pp.50-64.


20

robot is the idea that it exhibits some degree of autonomy, meaning that it is at least semi-autonomous.Autonomy,here,canbedefinedas“acapacitytooperateinareal-worldenvironmentwithout any formof external control, once themachine is activatedandat least in someareasofoperation,forextendedperiodsoftime”.26Dependingonwhatdegreeofautonomyisacceptedforrobots, tele-operated and remote-controlled robots canbe includedor excluded from the class ofrobots

Basedonthisanalysisofexistingdefinitions,itseemsthatagoodworkingdefinitionofarobotwouldbe:anelectro-mechanicalmachinewithsensorsandactuatorsthatcanmove(eitherentirelyorapartof its construction) in its environment and perform intended tasks autonomously or semi-autonomously.Thisdefinitiongenerallycoversanddoesnotexceedtherangeofthingsthatpeopleintuitively consider to be robots. It rules out simple stationary devices such as toasters, coffeemachines, and printers, which are clearly not robots. It also rules out any fully remote-controlledmachines,suchassimpleremote-controlledchildren’stoys,sincethosedevicesarenotautonomousorsemi-autonomous.(Thatis,suchmachinesdonotmakeanydecisionsforthemselves,astheyarefullydependentonhumaninput.)ManyUAVs,however,wouldqualifyasrobotsunderthisdefinition,becauseeventhoughtheyaretele-operatedbyhumans,theyoftenmakeatleastsomenavigationaldecisionsontheirown(e.g.,collisionavoidanceandreturn-to-homefeatures).

Core concepts, approaches and methods in robotics

Robotsareoftendevelopedtosubstituteforhumansandreplicatehumanactions,therationalebeingthattheycaninmanyapplicationsperformtasksmoreeffectivelyandefficiently,andreducestrainon,andsafetyrisksfor,humans.Inaddition,theycanextendthereachofhumaninfluencetoplacesthathumanshithertoneverhadaccessto,suchasthesurfacesofotherplanetsinthesolarsystem.

Asstatedabove,importantconceptsinroboticsandrobotsincludesensing,actuation,anddecision-making ability. Sensing is performed through sensors, which are devices,modules, or subsystemswhosepurposeitistodetecteventsorchangesintheirenvironment,andtosendinformationabouttheseeventsorchangestootherelectronicsforprocessing.Sensorsallowarobot’scontrolsystemstoreceive information about the robot’s environment and its internal components, which is thenprocessed and used to calculate an appropriate response, and/or relayed to human operators orsupervisors. Robots can have a wide variety of sensors that include cameras, microphones,accelerometers,thermometer,infrared,radar,sonar,andvibrationsensors.

Actuatorsarethemeansbywhicharobotperformsactions in itsenvironment.Theyuseenergytoproduce movement, sound, vibration, light or chemical reactions. Widely used actuators includeelectricmotorsthatproducetorquetorotatewheelsorgears,andlinearactuatorshatcreatemotioninastraightline.However,robotscanhaveavarietyofotheractuatorsaswell, includingspeakers,displays,LEDs,lasers,anddifferenttypesofmovement-producingactuators(e.g.,pneumaticartificialmuscles,piezoelectricmotors,electroactivepolymers).Bymeansofits(electro)mechanicalactuators,a robot can drive its other mechanical components and achieve complex motions with multipledegrees of freedom that are useful for object manipulation (e.g., through mechanical arms andgrippers)and locomotion(e.g.,walking,driving, rolling,swimming, flying).The“hand”ofarobot isusuallyreferredtoasan(end)effector,whilethe“arm”isreferredtoasamanipulator.Roboticmotionisstudiedinthefieldsofrobotkinematicsandrobotdynamics.Here,essentially,robotkinematicsis

26Lin,Patrick,KeithAbneyandGeorgeA.Bekey,op.cit.,2012.


21

thestudyofthegeometryofmotionofarobot’smechanicalparts,androbotdynamicsisthestudyoftheforcesthatareresponsibleforthismotion.

Themechanicalstructuresofrobotshavetobecontrolledtoenablethemtoperformtasks.Robotcontrol systems take sensor data as input and calculate the appropriate signals to be sent to theactuators. These systems use techniques from (robot) control theory and can range greatly incomplexity.Atareactivelevel,theymaytranslaterawsensorinformationintoactuatorcommandsinarelativelyquickandsimplefashion.However,atlongertimescalesorwithmoresophisticatedtasks,theymayneedtouseartificialintelligenceandreasonwithcognitivemodels,whichareintendedtorepresenttherobot,itsenvironment,andtheinteractionsbetweenthetwo.Furthermore,robotsmayusepatternrecognitionandcomputervisiontotrackobjects,techniquesinroboticmappingtobuildmapsoftheworldandlocalizethemselveswithinthesemaps,andtechniquesinmotionplanningtofigureouthowtheyshouldmoveefficiently,withouthittingobstaclesorfallingover.

A robot’s control systems determine its capacity for autonomous behaviour. Robots can range inautonomyfromfullyautonomoustosemi-autonomous.Ifarobotissemi-autonomous,itcanbelargelytele-operated,orbeattachedtoanddirectlyoperatedbythehumanbody(e.g.,prostheticdevices).Fullyautonomousorsemi-autonomousbehaviour in robotscanrange fromverybasic, suchas thebehaviour displayed by packaging robots that use simple pre-programmed algorithms, to verysophisticated,suchasthebehaviourshownbysemi-autonomouscarsthatemploytechniquesfromtheAIsubfieldsofcomputervisionandmachinelearning.

Robots come inmany forms, but they are increasingly being developed to resemble humans andanimals inappearance,mechanicsandbehaviour.Humanoidrobots (oranthrobots)arerobotsthatresemblehumanbeingsin importantways.Suchrobotsmaybedevelopedtoenablemorenatural,effective and efficient interactions with humans, so as to increase the likelihood of a robot’sacceptanceincertainapplicationswheretheyreplacehumanbeings. Ithasfurtherbeensuggestedthat advanced humanoid robotics research will facilitate the enhancement of ordinary humans.27Humanoid robots can be designed to replicate human locomotion, speech, gestures, cognition, orfacialexpressions,amongstothercharacteristics.

Somewhat related to humanoid robotics are the emerging areas of bio-inspired robotics andbiomimicry,whichstudyconceptsfrombiologicalsystems,simplify,enhanceandthenapplythemtothedesignofnewrobots.Bothfieldshavecontributedtothedevelopmentofabranchofroboticscalled soft robotics, which deals with constructing robots out of highly compliant (i.e., flexible)materials,similartothosefoundinlivingorganisms.Bio-inspiredroboticistsdrawinspirationfrom,forexample,biosensors(e.g.,thehumaneye),bio-actuators(e.g.,muscles),biomaterials(e.g.,spidersilk)anddifferentmodesofanimallocomotion(e.g.,insectwingmotions).

Asaresultofthecontinuousprogressinintegratedcircuitdesignandmanufacturingtechnologyofthepast few decades, robots and their subsystems (i.e., their sensors, actuators, and control andcommunicationssystems)havebeengreatlyminiaturizedanddecreasedincost.Thisdevelopmenthasmadepossibleentirelynewcategoriesofrobots,suchassmallconsumer-orientedUAVs.Morerecentprogressinso-calledmicroelectromechanicalsystems(MEMSs)holdspromiseforthedevelopmentofmicrobots,whichareverysmallrobotswithcharacteristicdimensionsoflessthan1mm.

27Kazerooni,H.,“HumanAugmentationandExoskeletonSystemsinBerkeley,”InternationalJournalofHumanoidRobotics,Vol.04,No.03,2007,pp.575–605.


22

Muchoftheresearchinroboticsdoesnotfocusondevelopingrobotsforspecifickindsofapplications,but on investigating and developing new fundamental techniques in robotics, alternative ways todesignandinteractwithrobotsandnewwaystomanufacturerobots.Adistinctioncanthereforebemadebetweenfundamentalrobotics,whichstudiesessentialaspectsofrobotdesignandbehaviourindependentfromrobotapplications,andapplication-orientedrobotics,whichfocusesonrobotdesignanddevelopmentforspecificapplications.

In recent decades, robotics research, education and product development have seen increasinginterest inthedevelopmentofopen-sourcesoftwareandfreetouseblueprintsandschematicsforrobotics hardware. These open-source blueprints and schematics for robotics hardware generallymake use of standard, commonly available, components. Related to this, the popularity of do-it-yourselfroboticshasbeenincreasing,withseveralcompaniesnowproducingkitsforbuildingsimplerobots.

Finally,animportantrecentdevelopmenthasbeentheintroductionof3Dprinters,whicharguablyfallunderthedefinitionofrobotsusedinthisreport.3Dprintingallowsobjectsofalmostanyshapetobecreatedthroughaprocesscalledadditivemanufacturing.Ithasbeensuggestedthat3Dprintingwillrevolutionisemanufacturingbyallowingindividualstodotheirownmanufacturinginsteadofbuyingphysicalproductsmanufacturedbycompanies.

2.2 Subfields of artificial intelligence and robotics

This section offers brief descriptions of the most important subfields of artificial intelligence androbotics,includingsomeofthemainconceptsandtechniquesusedinthem.

Subfields in artificial intelligence

ThefollowingisalistingofarguablythemostimportantsubfieldsinAI.Itshouldbenotedthatmanyofthesesubfieldsoverlaptoasignificantdegree(i.e.,sharingimportanttechniquesandmethodswithothersubfields)orareinterdisciplinarycombinationswithnon-AIfields.

Knowledge representation and automated reasoning. Knowledge representation deals with thefundamentalchallengesfacedinrepresentinginformationabouttheworldinaformthatacomputersystemcanutilizetosolvecomplextasks,suchasdiagnosingamedicalconditionorhavingadialoginanatural language. Things thatanAI systemmayneed to represent include: concepts, categories,objects,properties,situations,events,states,time,andcausesandeffects.Knowledgerepresentationgoes hand in hand with the capacity for automated reasoning about that knowledge. Tools andtechniquesofautomatedreasoningincludeclassicallogics,fuzzylogic,Bayesianinference,reasoningwithmaximalentropy,andalargenumberoflessformaladhoctechniques.Knowledgerepresentationandautomatedreasoningarefoundationaltothedevelopmentofexpertsystems(asubfieldthatwediscusslater).

Machinelearning.MachineLearningisabroadsubfieldwithinAIwherestatisticaltechniquesanddataareusedto“teach”computersystemshowtoperformaparticulartask,withoutthesesystemsbeingexplicitlyprogrammedtodoso.Forexample,inimagerecognition,computersystemsmightlearntoidentifyimagescontainingcatsbyanalysingexampleimagesthathavebeenmanuallylabelledas"cat"or"nocat",withoutaprioriknowledgeaboutdefiningfeaturesofcats.Thefieldiscloselyrelatedtocomputationalstatisticsandhasmanydifferentapproaches,whichincludeartificialneuralnetworks,Bayesiannetworks,reinforcementlearningandothers.IthasbeenakeycontributorinthesuccessofAI applications for product recommendations, image understanding, speech recognition, frauddetection,andothertaskwhichonlyhumanscouldperformnotsolongago.Machinelearningtaskor


23

approachescanbedividedintotwobroadcategories:supervisedlearningandunsupervisedlearning.Insupervisedlearning,asystemisprovidedwithexampleinputsanddesiredoutputsfortheseinputsand tasked to learn through iteration the general rule(s) that map the inputs to the output. Inunsupervisedlearning,nopre-specifiedoutputisgiventothelearningsystem,thusleavingittofindstructureintheinputonitsown.

Artificialneuralnetworks.Artificialneuralnetworks(ANNs)arecomputingsystemsvaguelyinspiredbythebiologicalneuralnetworksthatconstitutehumanandanimalbrains.Suchsystemshavefoundapplicationinmachinelearning,wheretheirusehasbecomeverycommon.Inmachinelearning,theuseofadvancedANNswithmorethanonehidden“layer”andamethodcalled“backpropagation”28,incombinationwithpowerfulcomputerhardwareandlargeamountsoftrainingdata,iscalleddeeplearning, which in some areas of application produces results comparable to, and in some casessuperiorto,humanexperts.29Animportantlimitationofdeeplearningmethodsandneuralnetworksingeneral,however,isthelowinterpretabilityoftheinferencemodelsthatresultfromthelearningprocess. Artificial neural networks and machine learning have been applied to subfields such ascomputervision,machinehearingandnaturallanguageprocessing,wheretheyhavebeenshowntoproducecutting-edgeresults.

Computervision.Computervisionistheabilityofacomputersystemtotakein,processandgainahigh-levelunderstandingofcontentpresentedindigitalimagesorvideos.High-levelunderstandinginthiscontextmeansthecreationofsymbolicdescriptionsofwhatisvisuallypresentedusingmodelsconstructed with the aid of geometry, physics, statistics, and learning theory.30 Sub-domains ofcomputer vision include scene reconstruction, event detection, video tracking, object or facerecognition,3Dposeestimation,learning,indexing,motionestimation,andimagerestoration.

Computeraudition.Computerauditionisthecapacityofasystemtoregisterandprocessaudiodata(e.g.,music,speech),andtofocusselectivelyonaspecificsoundagainstmanyothercompetingsoundsandbackgroundnoise(thelatterabilityiscalled“auditorysceneanalysis”).Thefieldhasawiderangeofapplicationthatincludesspeechsynthesis,speechrecognition(innaturallanguageprocessing)andmusicrecordingandcompression.

Natural language processing. Natural language processing gives machines the ability tounderstand human language in written or spoken form. Sufficiently powerful systems for naturallanguage processing enable the use of natural language user interfaces and the extraction ofknowledge directly from speech or human-written sources (e.g., email correspondence). Somestraightforward applications of natural language processing include information retrieval, textmining,questionanswering,conversationalagents,andmachinetranslation.Acommonmethodofprocessingandextractingmeaning fromnatural language in text is called latent semanticanalysis,whichworkswiththeassumptionthatwordsthatarecloseinmeaningwilloccurinsimilarpiecesoftext.

28Backpropagationisanabbreviationfor“backwardpropagationoferrors”.Thisisacommonmethodusedtotrainartificialneuralnetworks.Itistypicallyusedincombinationwithanoptimizationmethodsuchasgradientdescent.Inbackpropagation,thegradientofalossfunctioniscalculatedwithrespecttoallweightsinthenetwork.29Thomsen,Michael,“Microsoft'sDeepLearningProjectOutperformsHumansinImageRecognition,”Forbes,ForbesMagazine,February19,2015.https://www.forbes.com/sites/michaelthomsen/2015/02/19/microsofts-deep-learning-project-outperforms-humans-in-image-recognition/#4fb4d93d740b30Forsyth,DavidA.,andJeanPonce,ComputerVision:AModernApproach,PrenticeHall,UpperSaddleRiver(NewJersey),2003.


24

Expertsystems.Expertsystemsarecomputersystemsthatemulatethedecision-makingabilityofahuman expert by reasoning through an extensive body of human knowledge represented in adatabase.Theyarebuiltbyapplyingknowledgeengineeringtechniques,andaretypicallycomposedoftwosubsystems–theknowledgebaseandtheinferenceengine.Theknowledgebaseisanorganisedcollectionoffactsandrulesrelevant inaparticulardomain,andtheinferenceengineappliestheserulestothesefactstologicallydeducenewfacts.Expertsystemscanbeusedbyhumanoperatorstosolve complex problems, and were among the first truly successful forms of artificial intelligencesoftware.

Datamining.Dataminingistheprocessofdiscoveringpatternsinlargedatasetsinvolvingdatabasesystems,statisticalanalysis,andintelligentmethodssuchasmachinelearning.Theoverallgoalistoextractinformationfromlargedatasetsandtransformitintoanunderstandablestructureforfurtheruse.Applicationsofdataminingareinmanufacturing,marketing,finance,insurance,healthcare,andtransportation,amongstothersectors.

Intelligentagentsystems,automatedplanning.Asexplainedinsubsection2.1,intelligentagentsareautonomous,artificiallycreatedentitiesthatperceivetheirenvironmentthroughsensors,actupontheirenvironmentusingactuators,anddirecttheiractivitytowardsachievinggoals.Foragentstobeabletosetgoalsandachievethem,theyneedtobeabletomakepredictionsonthefuturestateoftheirenvironment,onhowpotentialactionswillchangeit,andonwhatwouldbethebestcourseofaction.Automatedplanningandschedulingistheareawheresolutionstotheseproblemsaredevised.Inmulti-agentplanning, cooperationandcompetitionbetweenmanyagentsareused toachieveagivengoal.Multi-agentplanningformsthebasisforswarmintelligence.

Evolutionary computation. Evolutionary computation is a subfield of AI that studies algorithms forglobaloptimizationthatareinspiredbybiologicalevolution.Techniquesinevolutionarycomputationcan produce highly optimized solutions for a wide range of problems, making them popular incomputerscience.Essentially,aninitialsetofcandidatesolutionsisgenerated,whichistheniterativelyupdatedthroughanevolutionaryprocess.Eachnewgenerationisproducedbystochasticallyremovinglessdesiredsolutions,andintroducingsmallrandomchanges.Inbiologicalterminology,apopulationof solutions is subjected to natural selection (or artificial selection) andmutation. As a result, thepopulationwillgraduallyevolvetoincreaseinfitness.

(Intelligent)Informationretrieval.Informationretrievalisasubfieldofcomputersciencethatstudieseffective andefficient algorithmicmethods for searchprocesses inwhichauser tries to identify asubset of information within a large amount of information. The problems of how to properlyrepresent information in indexes and how to match imprecise representations has led to theapplicationofAItechniques.

Subfields in robotics

Thesubfieldsinroboticsaremany.Theycanbeusefullydividedintotwomaincategories:fundamentalroboticsandapplication-orientedrobotics.Fundamental roboticsstudiesessentialaspectsof robotdesignandbehaviour,largelyindependentfromrobotapplications.Ontheotherhand,application-oriented robotics is focused on robot design and development for specific (types of) applications.Below,alistingofsubfieldsisonlyprovidedforfundamentalroboticsresearchandinnovation,andnotforapplication-orientedrobotics,sincerobotapplicationsinvariousdomainsarealreadydescribedatlengthinsubsection3.2.Itshouldbenotedthatthelistingpresentedbelowisanattemptatdescribingthemostimportantfundamentalroboticssubfields,butthat,giventhemanyspecializedsubfieldsthatexist,itmaynotbecomplete.


25

Robotmechanics.Robotmechanics isasubfieldofroboticsthatcoversthemechanicalengineeringaspectsofrobotdesign.Principlesofrobotmechanicsareinvolvedinthedevelopmentofallrobots.The subfield includes the studyof factors such as inertia, stress, load-carrying ability anddynamicresponse in thedesignof robotarmsandgrippers, transmissionsystems,hydraulicandpneumaticcouplingsandrelateddesignareas.

Robotsensing.Roboticsensingisasubfieldthatstudiesanddevelopstechniquestogiverobotssensingcapabilities(i.e.,theabilitytodetecteventsorchangesintheirenvironment,andtosendinformationabout these events or changes to other electronics for processing). Such techniques can relate tosensordesignand (pre-)processingof sensordata.Robots canhaveawide varietyof sensors thatinclude cameras,microphones, accelerometers, thermometer, infrared, radar, sonar, and vibrationsensors.

Robotcontrol.Robotcontrolisabroadsubfieldofroboticsthatdealswiththeapplicationofmethodsandtechniquesfromcontroltheoryincontrollingthemechanicalstructureofrobots.Robotcontrolsystemscanvarygreatlyincomplexity.Belowaredescriptionsofanumberofimportantsubfieldsofrobotcontrol.

• Robotlearning.Robotlearningisasubfieldofrobotcontrolthatcombinesmachinelearningandrobotics. It studies techniques thatallowrobots toacquirenewskillsoradapt to theirenvironmentthroughapplicationoflearningalgorithmsand/orneuralnetworks.

• Adaptivecontrol.Adaptivecontrolisasubfieldthatstudieshowcontrolmethodsemployedbyacontrolsystemcanadapttoparameterswhicharechangeableorareuncertainat thestart. For instance, in a flying aircraft, total mass will decrease over time due to fuelconsumption,thusnecessitatingacontrollawthatcanadapttochangingconditionslikethis.

• Developmentalrobotics.Developmentalroboticsisasubfieldthatstudiesthedevelopmentalmechanisms,architecturesandconstraintsthatallowforlifelongandopen-endedacquisitionofnewskillsandknowledgeinrobots.Itisdirectlyinspiredbythedevelopmentalprinciplesandmechanismsresponsibleforcognitivedevelopmentinchildren.

• Evolutionary robotics. Evolutionary robotics is a subfield that studiesmethods that enableautomaticincrementalimprovementsinautonomousrobotsthroughaprocessakintonaturalselection.Robotsinthisfieldareseenasartificialorganismsthat,throughcloseinteractionwiththeirenvironmentandwithouthumaninterference,candeveloptheirownuniqueskills.The best or fittest of these robots are iteratively selected and used as a basis for furtherdiversification.

• Cognitiverobotics.Cognitiveroboticsisasubfieldthatstudiesmethodstoendowrobotswithintelligentbehaviourthatborrowfromanimalcognitionmodelsratherthanmoretraditionalartificialintelligencetechniques.

• Behaviour-basedrobotics.Behaviour-basedrobotics isasubfieldthatstudieshowtocreaterobotsthatarecapableofexhibitcomplex-appearingbehavioursdespitehavinglittleinternalvariablestatetomodelitsimmediateenvironment.Robotsdevelopedthiswayareadaptableinthattheyrequirenopre-setcalculationstodealwithasituation,andtheyarereactiveinthattheycancorrecttheiractionsdirectlyviasensory-motorlinks.

• Roboticmappingandplanning.Roboticmappingstudiestechniquesforrobotstobuildmapsoftheworldandlocalizethemselveswithinthesemaps.Thesemapsareusedinrobotic(path)planning,whichappliestechniquesinmotionplanningtodeterminehowrobotsshouldmoveefficientlyintheirenvironment,withouthittingobstaclesorfallingover.


26

Robotlocomotion.Humanoidroboticsisasubfieldthatstudiesthevariousmethodsthatrobotsusetotransportthemselvesfromplacetoplace.This involvesthedesignofbothmechanicalsystemsandcontrol systems. A major goal in this subfield is the development of capabilities for robots toautonomouslydecidehow,when,andwheretomove.Differenttypesoflocomotionthatarebeingstudied include bipedal walking, walking with more than two legs, running, rolling, hopping,metachronalmotion(i.e.,wavymotion),slithering,swimming,brachiating(i.e.,armswinging).

Bio-inspiredandsoftrobotics.Bio-inspiredroboticsisasubfieldthatstudiesconceptsfrombiologicalsystems, simplify, enhance and then apply them to the design of new robots. Both fields havecontributed to the development of a branch of robotics called soft robotics, which deals withconstructing robotsoutofhighlycompliant (i.e., flexible)materials, similar to those found in livingorganisms.

Humanoidrobotics.Humanoidroboticsisasubfieldfocusedondevelopingandstudyingrobotsthatresemblehumansintermsofappearance(bodyshape)and/orbehaviour.Robotsmaybedesignedtohavehumanresemblanceforfunctionalpurposes,suchasforpurposesofinteractingwithhumantoolsand in human social environments, or for experimental reasons, such as the study of humanlocomotion. Usually, humanoid robots have a torso, a head, two arms and two legs, while somehumanoidrobotsmaymimiconlyparticularpartsofthehumanbody(e.g.,onlytheupperbody).Someadvanced humanoid robots have heads with faces that can replicate human facial features (e.g.,humaneyesandmouth)anddisplayhuman-likebehaviour(e.g.,speech,facialgestures).

Microrobotics.Microroboticsisasubfieldthatfocusesoncreatingmobilerobotswithcharacteristicdimensions of less than 1mm,which are sometimes called “microbots”. Thesemicrobots usuallyfeaturesimpledesignsandlackthecomputingpoweroflargerrobots.Oneofthemajorchallengesinmicroroboticsistoachievemotionusingaverylimitedpowersupply.Duetotheirsimplicityandsmallsize,however,microbotsarepotentiallyverycheapandcouldbeused in largenumberstoexploreenvironments that are too dangerous for people or too small for people and larger robots. It isexpectedthatmicrobotswillbeuseful inapplications insearchandrescue,andmedicine,amongstotherfields.

Nanorobotics. Nanorobotics is an emerging subfield that focusses on the creation of robots thecomponents of which are at the scale of a nanometer or close to it. Presently, research anddevelopmentofsuchnanomachinesislargelystillintheearlystages,althoughsomesimplemolecularmachines, includingnanomotorsandnanosensors,havealreadybeentested.Usefulapplicationsofnanomachines are expected to be in nanomedicine (amongst other fields), where such minisculerobotscouldbeinjectedinaperson’sbodytoidentifyandeliminatecancercells.

BEAMrobotics.BEAM(Biology,Electronics,AestheticsandMechanics)roboticsisasubfieldthatusessimple analogue circuitsmimicking anetworkof biological neurons, inorder toproduceunusuallysimple robot designs. Generally, BEAM robots trade flexibility for robustness and efficiency inperformingthetasksforwhichtheyaredesigned.

Cloudrobotics.Cloudroboticsisanemergingsubfieldthatutilisescloudtechnologiescentredontheconvergenceofinformationandcommunicationinfrastructuresandsharedservices–includingcloudcomputing,cloudstorageandrelatedinternettechnologies–inthedevelopmentofroboticsystems.Whenconnectedtodatacentresinthecloud,robotscanbenefitfromthesecentres’powerful(andrelativelyinexpensive)storage,computationandcommunicationresourcesintheprocessingofdataandtheexchangeofinformationwithotherrobots.Thismakesitpossibletobuildlighter,cheaperandmoreintelligentrobotsthathavetheir“brains”inthecloud.


27

Swarm robotics. Swarm robotics is a subfield focused on the development of methods for thecoordinationof largenumbersof(usually)simplerobots inmulti-robotsystems.Researchers inthefieldworkontheassumptionthatprogrammingindividualrobotsforinteractionwithoneanotherandtheirenvironmentcanproducecoherentanddesiredcollectivebehaviours.SwarmroboticsbuildsontheAIsubfieldofartificialswarmintelligence,aswellasonthebehaviouralstudyofinsectsandotherfieldswhereswarmbehaviourcanbeseen.

Telerobotics.Teleroboticsisasubfieldconcernedwiththecontrolofsemi-autonomousrobotsfromadistancebymeansofwirelessorwiredcommunication.Thefieldcanbesubdividedintwootherfields:teleoperationandtelepresence.Teleoperationreferstotheoperationofrobotsatadistance,whiletelepresencereferstocapabilityofrobotsofallowingapersontofeelorappeartobepresentataplacethatisnottheirtruelocation.

Socialroboticsandhuman-robotinteraction.Socialroboticsisafairlyrecentsubfieldfocusedonthestudyanddevelopmentofautonomousrobotsthatarecapableofinteractingwithhumansthroughthedisplayofsocialbehaviourandadherencetorulesattachedtotheirsocialrole.Onabroaderlevel,human-robotinteractionisthestudyofhowhumansinteractwithrobots,andhowbesttodesignandimplementrobotscapableofaccomplishinginteractivetasksinhumanenvironmentsinwaysthatareeffective,safeandconvenientforhumans.

2.3 Present and future developments and challenges in AI and robotics

Inthissubsection,wediscusssomeimportantpresentandfuturedevelopmentsandchallengesforAIandrobotics.

Artificial intelligence

Currently,thefieldofAIisdevelopinginseveralimportantways.Oneofthemostsignificantoveralldevelopmentsisthecontinuedadvancementoftheconnectionist,data-drivenparadigm.Inparticular,we are witnessing the maturation of neural-network-based machine learning methods andarchitectures,whichareoftensupportedbycloudcomputingresourcesand large-scale,web-baseddatagathering.Itisexpectedthatsystemsbasedonthesemethodsandarchitectureswillevolvetobecome increasingly human-aware.31 This means that they are specifically designed for, and canaccurately model, the characteristics of the people they will interact with (e.g., their mentalcapabilities,theirpersonalities).

Toanimportantdegree,machinelearningisbeingpropelledforwardbydeeplearning(seesubsection2.2).Muchofthecurrentresearcheffortindeeplearning(andmachinelearningingeneral)focusesonscalingalgorithmstoworkwithverylargedatasets.32Whilenormalsizeddatasetsallowfortheapplicationoftraditionaldataanalysismethodsinwhichseveralcompletepassesaremadeoverthedata,withextremelylargedatasetstheuseofsuchmethodsbecomesimpractical.Oftensuchsetscanonlybeanalysedusingsublinearmethods,thatis,methodsthatonlyconsideracomparativelysmallportionofthedata.Inaddition,deeplearningismakingsignificantinroadsintoareasbesidescomputervision(e.g.,object/activityrecognition,videolabelling),suchasaudio,speechandnatural language

31IBM.“AAAI-17Workshop:Human-AwareArtificialIntelligence(HAAI-17),”IBM,July25,2016.https://researcher.watson.ibm.com/researcher/view_group.php?id=744632Grosz,Barbara,AlanMackworth,RussAltman,TomMitchell,EricHorvitz,DeidreMulligan,andYoavShoham,“ArtificialIntelligenceandLifein2030,”September2016.https://ai100.stanford.edu/sites/default/files/ai_100_report_0831fnl.pdf


28

processing.Furthermore,aspecifictypeofmachinelearningcalledreinforcementlearningismakingstrides,inpartowingtodevelopmentsindeeplearning.Reinforcementlearningallowsmachinesandsoftwareagentstoautomaticallydeterminetheidealbehaviourwithinaspecificcontext,inordertomaximizetheirperformance.TherecentsuccessofGoogleDeepmind’sAlphaGoAIsystem,whichbeatahumanGochampioninafive-gamematch,wasinlargepartattributedtotheuseofreinforcementlearning.33

Withinmachineperception,computervisioniscurrentlyoneofthemostprominentareas.Incomputervision, the interplay of large-scale computing, the availability of large datasets (e.g., throughcrowdsourcingon the internet)andmore refinedneuralnetworkalgorithmshasbroughtdramaticperformance improvements in benchmark tests.34 Only recently have some AI systems becomecapableofperformingsomevisualclassificationtasks(narrowlydefined)betterthanhumanbeings.Muchoftheresearchthatiscurrentlygoingonisfocusedonautomaticimageandvideocaptioning,andgreatprogressisbeingmadeinthisarea.35

Naturallanguageprocessing(oftencoupledwithautomaticspeechrecognition)isanotherveryactiveareaofresearch inmachineperception.Googleannounced in2016thatroughlytwentypercentofpresentmobilewebsearchqueriesweredonebyvoice,36andrecentexperimentshaveshowcasedthefeasibilityofreal-timemachinetranslation.Focusisnowshiftingtowardsthedevelopmentofsystemsthatmovebeyondtheabilitytoreacttostylizedrequestsandcanusenaturalconversationaldialogtointeractwithpeople.

Further,agrowingbodyofresearchisdevotedtothestudyofmodelsandalgorithmsforcollaborativesystemscomposedofartificialagentsandhumans.37Applicationswherethecomplementarystrengthsofmachinesandhumansarebroughttogetherhavebeenanareaofsignificantinterest.Humansmayhelp AI systems to overcome their limitations,while intelligent agents and systemsmay augmenthumanabilitiesandactivities.

SignificantresearchisbeingdoneonAImethodstoanalyseandusethehugeamountsofdatathatresult frominterconnecteddevicesthatareapartof the InternetofThings (IoT).Suchdevicescaninclude appliances, vehicles, buildings, cameras, and other things. Currently, these devices use abewilderingarrayofincompatiblecommunicationprotocols,andAImethodsareexpectedtobeabletohelpinmakingthedevicescommunicatemoreeffectivelyandefficiently.

With the success of AI methods based on neural architectures, some researchers are pursuingalternativemodelsof computing thatareabettermatch for thesemethods in termsofefficiency.Presently,connectionistAImostlyemploystraditionalVonNeumann-basedcomputerarchitecturesinwhichtherearedifferentmodulesfor inputandoutput,memoryandinstruction-processing. Inthe

33Silver,David,andDemisHassabis.“AlphaGo:MasteringtheancientgameofGowithMachineLearning.”GoogleResearchBlog,January27,2016.https://research.googleblog.com/2016/01/alphago-mastering-ancient-game-of-go.html34Grosz,Barbara,etal.,op.cit.,2016.35PetaPixel,“Google'sImageCaptioningAICanDescribePhotoswith94%Accuracy,”PetaPixel,September23,2016.https://petapixel.com/2016/09/23/googles-image-captioning-ai-can-describe-photos-94-accuracy/36Sterling,Greg.“Googlesays20%ofmobilequeriesarevoicesearches,”SearchEngineLand,May18,2016.https://searchengineland.com/google-reveals-20-percent-queries-voice-queries-24991737Grosz,Barbara,etal.,op.cit.,2016.


29

future,however,wecouldseeneuralnetworksbeing trainedandexecutedondedicatedso-called“neuromorphic”hardwarethatareinspiredbywhatisknownaboutbiologicalneuralnetworks.38

Robotics

Thefieldof roboticswillalsoseeanumberof importanttrendsanddevelopmentswithinthenexttwenty years.On a global level, these include the increased use of AI techniques in robots, lowerdevelopmentandproductioncostsofrobotsandsmallerdimensionsforsomecategoriesofrobots.Aspreviouslymentioned,AIsystemsareevolvingtobecomeevermorecapable,anditislikelythatfutureAI advanceswill also find theirway into robots and enhance their capabilities. For example, next-generationmachine-learningalgorithmsareexpectedtosignificantlyimprovethenavigationalabilitiesofmobile robots,39 andgivehumanoid social robotsmorehuman-likeexpressions and reactions.40Lowerdevelopment,productionand implementationcostsareexpectedasa resultofadvances incomputationalpowerandsoftwaredevelopmentandnetworkingtechniques41,andwilllikelyenablethedeploymentofrobotsinnewapplicationareas.,42

IncreaseduseofsophisticatedAItechniques(e.g.,machinelearning,computervision)inrobotswillallowthemtogainevermoreautonomy,especiallyintermsofmobilityandnavigation.Inthemediumterm,wemaysee thematurationof fullyautonomouscarsandunmannedaerialvehicles thatarecapable of transporting people and goods in public environments. AI techniques may allow forsophisticated coordination among robots themselves (e.g., in swarms of small robots deployed inindustry,themilitaryanddisasterrelief)andbyrobotsintheinteractionwithhumans.Robotsactinginpublicwillbedesignedtoanticipate,tosomedegree,thebehaviourofhumansaroundthemandtotakeexplicitdecisionsinsituationsthatposeamoraldilemma(e.g.,inautonomouscars).

A significant development in robotics is the increasing use of cloud storage43 and computation inrobotics.44Asexplained in subsection2.2, robots canbenefit fromexternaldata centres’powerfulstorage,computationandcommunicationresources in theprocessingofdataand theexchangeofinformationwithotherrobots.Cloudroboticsallowsforthedevelopmentoflighter,cheaperandmoreintelligentand flexible robots, thecreationofnetworksof intelligent,autonomous robots,and theanalysisoflargedatasetsproducedbysensorrichrobots(e.g.,autonomoussurveillanceaircraft).

Anotherkeyareaforgrowthinroboticsaresocialrobots.AdvancesinAIwillallowforthedevelopmentof robots that display increasingly complex social behaviours, such as recognizing, following andassistinghumansandengagingtheminnaturalconversation.45Thesesocialrobotsarelikelytofind38Dent,Steve,“IntelUnveilsanAIChipThatMimicstheHumanBrain,”Engadget,September26,2017.https://www.engadget.com/2017/09/26/intel-loihi-neuromorphic-chip-human-brain/39TharsusGroup.“2018TrendsinRobotics,”Tharsus,n.d.https://www.tharsus.co.uk/2018-trends-in-robotics/40DeloitteDevelopmentLLC,“FutureofRoboticsTechnology,”Government2020,n.d.http://government-2020.dupress.com/driver/robotics-technology/41Gupta,SatyandraK.SixRecentTrendsinRoboticsandTheirImplications,”IEEESpectrum,September8,2015.https://spectrum.ieee.org/automaton/robotics/home-robots/six-recent-trends-in-robotics-and-their-implications42Tilley,Jonathan.“Automation,robotics,andthefactoryofthefuture,”McKinseyInsights,September2017.https://www.mckinsey.com/business-functions/operations/our-insights/automation-robotics-and-the-factory-of-the-future43Gupta,SatyandraK.,op.cit.,2015.44DeloitteDevelopmentLLC,op.cit.,2015.45KPMGAdvisory.“SocialRobots,”KPMG,2016.https://assets.kpmg.com/content/dam/kpmg/pdf/2016/06/social-robots.pdf


30

increaseduseinarangeofdomains, includinghealthcare,education,retailandentertainment.46 Intheshortterm(5yearsfromnow),theirsocial“dexterity”mayimprovetoanextentthattheyareableto sense a person’s emotional state and act as a “teammate” in solving problems in structuredsituations,whileinthemediumtolongtermsomemayhavethecapacitytounderstandaperson’sfeelingsandactasacoachtohumans.47

Acategoryofrobotsthathasoverlapwithsocialrobotsandisexpectedtodevelopsignificantlyinthecomingyearsisthatofcollaborativerobots(orco-bots).Theserobotsarestartingtobeintroducedinindustry,where theyare transformingproductionandmanufacturingbyworkingalongsidehumanworkersinacollaborativefashion.48Theyaredesignedtobesmall,safe,flexibleandtrainable,andcanhelpin,forexample,theassemblyorpackagingofproductsalongaconveyorbelt.Collaborativerobotswithsignificantsocialcapabilitiescanalsobeconsideredsocialrobots.

46Ibid.47Ibid.48Bernier,Catherine,“WhatIsaCollaborativeRobot?,”RobotIQ,September20,2013.https://blog.robotiq.com/bid/66463/What-is-a-Collaborative-Robot


31

3. Present and future applications Inthissection,wediscussmanyoftheproductsandapplicationsthathavebeendevelopedormaybedevelopedwithinthefieldsofAIandrobotics.First,subsection3.1offersanoverviewofpresentandfuture developments in AI applications based on the domain of application (e.g., transportation,finance,healthcare,education,security).Then,subsection3.2providesasimilaroverviewforroboticsapplications.

ManydevelopmentsinroboticsandAIfallintoboththeroboticscategoryandAIcategory;thatis,theyareexamplesofartificiallyintelligentroboticsapplications.Toavoidrepetition,suchapplicationsareexclusivelydiscussedinsubsection3.2onroboticsapplications.

It should be noted that discussing future applications of emerging technologies is a speculativeendeavour,andthisisespeciallytrueforsuchcomplex(relativelypoorlyunderstood)andpervasivetechnologiesasAIandrobotics.Longeranticipatorytimeframesincreasethelevelofuncertainty.Someoftheanticipatedapplicationswehavereferencedmaynotmaterialisewithinthegiventimeframes–ifatall–evenincaseswhereatpresentthereseemstobeaconsensus.Inaddition,thisreviewofapplicationsmayomit some important future applications that are currently hard to foresee.Oneshouldthereforekeepinmindthatthisreviewlistsfutureapplicationsthatatthispointintimeseempossibleandsometimesquitelikely,butarenotguaranteed.

ThetimehorizonsetforinclusionoffutureAIandroboticsapplicationsinthisstudyissetlooselyat20yearsfrompresent.Thisisconsideredneitherapointintimetoofarintothefuture(makingtheanalysistoospeculative),noronethatistooclosetothepresent(decreasingtheanticipatoryvalueoftheanalysis).

3.1 Artificial intelligence

This subsection offers an overview of important present and expected future applications of AI.Applications are discussed for the following application domains: transportation, infrastructure,healthcare, finance and insurance, security (military and law enforcement), retail and marketing,mediaandentertainment,science,education,manufacturingandagriculture.

Transportation

AIuseinautonomousvehiclesisdiscussedinsubsection3.2.

Presentapplications.CitieshavebeeninvestinginAIsystemsconnectedtosensornetworksthatareaimedatimprovingtheuseefficiencyofthelimitedresourcesinthetransportationnetwork.49NewYorkCityhasstartedusingasystemthatcombinesnetworksofvideocameras,microwavesensorsandpassreaderstodetectvehicletrafficwithinthecity.50OthersensorsthatcanbeusedinsuchnetworksincluderadarandGPS.AImethodsareusedinvariouswaystooptimisetransportationservicessuchas bus and subway schedules and to track traffic conditions, enabling, for example, dynamicallyadjusted speed limits and smart pricingon theuseof highways, high-occupancy vehicle lanes and

49Lohr,Steve,“BringingEfficiencytotheInfrastructure,”TheNewYorkTimes,April29,2009.http://nytimes.com/2009/04/30/business/energyenvironment/30smart.html50AccessScienceEditors,“ActiveTrafficManagement:AdaptiveTrafficSignalControl,”AccessScience,2014.accessscience.com/content/active-traffic-managementadaptive-traffic-signal-control/BR0106141.


32

bridges.51Inaddition,trafficlighttimingcanalsobeoptimizedusingAIsystemsandsensornetworksforimprovedtrafficflow.52On-demandpeer-to-peerridesharingservices,suchasprovidedbyUber53and Lyft54, have emerged as another significant application of AI systems (and connected sensornetworks),55 with algorithms that canmatch passengers to drivers in specific locations by varioussuitabilityfactors.

Futureapplications.AI systemsare likely tohave increasingly significant impactson transportationsystemsinmajorcities.Predictivemodelsthataccuratelydescribethemovementsbyindividuals,theirpreferencesandtheirgoalsareexpectedtoemergewiththegrowthinsensinganddataprocessingcapabilities,whichmaycontributetoimprovedmanagementoftransportationservicesandnetworksinreal-time.56

Infrastructure

Presentapplications.EnergycompaniesareincreasinglymakinguseofAItechnologieslikemachinelearning as increased competition and advances in new energy technologies require solutions forproblemsinenergymanagement,energyoptimization,energydistributionandbuildingautomation.57Thedeploymentofsuchtechnologiesintotheenergysectorhasresultedinsophisticateddevicesforauto-detection of energy demand and precise sensors on energy infrastructure. In addition, AItechnologies are being used inwatermanagement. For example,UrbanFlood is an Internet-basedearly-warningsystemdesignedtogatherandanalysedataproducedbyverylargesensornetworksinflooddefenceslikeembankments,levees,dikesanddams.ThissystemusesAIandreal-timesensordata toquicklycalculate the likelihoodofdike failure,aswellaspotential scenarios regardingdikebreaching and flood spreading.58 Further, AI technologies are being used for crowd simulation formajor events and urban planning. For example, AI-based simulation programs have been used toensurethesmoothflowoftrafficintheGrandMosqueinMecca.59

MajorcompaniesincludingMicrosoftandnVidiaarecurrentlyofferingsoftwareplatformsthatuseAItechnologytoaidcities indealingwithurbanplanningchallenges includingshrinkingrevenuesandbudgetcuts,agingcitizens,limitednaturalresources,outdatedorinefficientroadinfrastructure,rising

51Jang,Kitae,KoohongChung,andHwasooYeo,“ADynamicPricingStrategyforHighOccupancyTollLanes,”TransportationResearchPartA:PolicyandPractice,No.67,2014,pp.69–80.52Sims,ArthurG,andKennethWDobinson,“TheSydneyCoordinatedAdaptiveTraffic(SCAT)SystemPhilosophyandBenefits,”IEEETransactionsonVehicularTechnology.Vol.29,No.2,1980,pp.130–137.53Seehttp://uber.comformoreinformation54Seehttp://lyft.comformoreinformation.55Meyer,Jared,“UberandLyftAreChangingtheWayAmericansMoveAboutTheirCountry,”NationalReview,June7,2016.http://nationalreview.com/article/436263/uber-lyft-ride-sharing-services-sharing-economy-are-future56Grosz,Barbara,etal.,op.cit.,2016.57Wood,Laura,“ArtificialIntelligenceMarketinEnergy&Utilities-Forecast(2017-2022)-ResearchandMarkets,”BusinessWire,December29,2017.http://businesswire.com/news/home/20171229005229/en/Artificial-Intelligence-Market-Energy-Utilities---Forecast58Pengel,Bob,LudolphyWentholt,ValeriaKrzhizhanovskaya,etal.,“TheUrbanfloodEarlyWarningSystem:SensorsandCoastalFloodSafety,”2010.http://edepot.wur.nl/19711659Thankappan,Jeevan,“SaudiUsesAIforCrowdControlatGrandMosqueinMecca,”TahawulTech,December5,2017.http://tahawultech.com/news/saudi-uses-ai-crowd-control-grand-mosque-mecca


33

energydemandsandregulatoryrequirements.60,61Amajorpracticalapplicationisthemeasurementby such platforms of pedestrian flow and vehicle traffic, using camera data and computer visiontechniques,todynamicallysetappropriatetrafficlightbehaviour,thushelpingtopreventcongestionandreducepollution.

Futureapplications.Inthefuture,advancedcomputervisiontechniquesmaybeusedtogaininsightsthat can serve as input for urban planning activities. For example, MIT’s Media Lab have beendevelopingacomputervisionsystemthat,whentrainedwithhuman-produceddata,candeterminehowsafeaneighbourhoodwouldappearforhumanobserversbyanalysingphotostakenofitatstreet-level.62Itisexpectedthatsystemssuchasthiswillbeusedbycitiestodetectearlysignsofurbandecayandwillleadtomoreeffectiveinterventions.

Healthcare

AIuseinroboticsurgeryisdiscussedinsubsection3.2.

Presentapplications.SignificantapplicationsofAIinhealthcareareinclinicaldecisionsupport,patientmonitoringandcoaching,preventativemedicine,andmanagementofhealthcaresystems.56However,currentuseofAIsystemsintheseareasisstillataminimallevel.Somerecentsuccesses,suchassocialmediaminingandmachinelearningtoinferpeople’shealthrisks,AI-basedpersonalhealthmonitoringdevices (e.g.,healthandfitness functionsonApple’s iWatch),andtele-roboticstosupportsurgery,haveclearlydemonstratedthepotentialofusingAIsystemsinthehealthcaredomain.

Future applications. It has been suggested that in the next fifteen years advances in AI – whencombined with sufficient data and well-targeted systems – will lead to significant change in thecognitivetasksassignedtoat leastsomegroupsofhealthcareprofessionals.63Currently,physiciansusuallyrequestthatpatientspresentthemselves inpersonandprovideverbaldescriptionsof theirsymptoms.Thephysicianswillsubsequentlycorrelatethepatternsagainsthisorherknowledgeoftheclinical presentation of known diseases. Through the use of AI systems (e.g., virtual assistants orchatbots,equippedwithnaturallanguageprocessing,knowledgemanagementandsentimentanalysiscapabilities64),thephysiciancouldinsteadsupervisetheprocess,guidingpatientinputintothesystemthroughapplicationofhisorherexperienceandintuitionandthenevaluatingthesystem’soutput.Thiswouldhavethepotentialtounburdenoverworkedphysiciansandgeneratesignificantcostandtimeefficiencyimprovements.Limitationsandchallengesfortheuseofsuchsystemsarepresentedbythecontinuedneedforhands-onsupervisionbyphysicians,whichwouldalwaysremaincritical,andtheintegrationofthehumandimensionsofcarewithautomatedreasoningprocesses.

Inthefuture,smartdevicessuchassmartphonesandsmartwatchesmaybeequippedwithmorethanjustbasichealthandfitnessmonitoringcapabilities.Smartwatchescancurrentlybeusedtomonitoraperson’shearthrate,activitylevels,andenergyexpenditure.Soon,however,theymayusemachine-

60Pleaseseehttp://enterprise.microsoft.com/en-us/industries/citynextformoreinformation61Pleaseseehttp://nvidia.com/en-us/deep-learning-ai/industries/ai-citiesformoreinformation.62Hardesty,Larry,“WhydoSomeNeighbourhoodsImprove?DensityofHighlyEducatedResidents,RatherthanIncomeorHousingCosts,PredictsRevitalization,”MITNews,July6,2017.http://news.mit.edu/2017/highly-educated-residents-neighborhoods-improve-070663Ibid.64Siwicki,Bill,“AIChatbotsMightbetheMoney-SaversHospitalsareLookingFor,”HealthcareITNews,September18,2017.http://healthcareitnews.com/news/ai-chatbots-might-be-money-savers-hospitals-are-looking


34

learningalgorithmstopredictonthebasisofrelativelyfewvariables(includingheartrate,bloodsugar,steps, etc.), whether a person has or is at risk of developing certain health conditions, includingdiabetes,highbloodpressureandsleepapnoea.65Someof thesedevicesmayalsobeusedtogivetractiontoAIdiagnosis.AsIBM’sWatsonisalreadyabletomatchrecommendedtreatmentswith99%of the cases it has been tested on, increasing the data available to this AImay very well lead toincreased accuracy and faster diagnosis times compared to conventionalmethods of diagnosis. Ifeffective,thiswillalsohelptoreducecostandresourceconstraintsonexistinghealthcaresystems66.

Another promising area in healthcare thatwill gainmomentum in the coming years is automatedimageinterpretation.Theuseoflargeamountsofmedicalimagedatacontainedinelectronicpatientrecord systems will enable the application of large-scale machine learning techniques (e.g., deepneuralnetworks)totrainAIsystemsinmedicalimageinterpretation.Academicresearchinthisareahasshownthatdeepneuralnetworks,whentrainedbylargemedicalimagedatasets,canproducebasicradiologicalfindingswithhighreliability.67Itisnotexpectedthatthenexttwentyyearswillbringfullyautomatedradiology,butsecond-levelcheckingbyAIsystemswilllikelyleadtosignificantcostandtimeefficiencyimprovementsinmedicalimaging.

AIalsoholdspromiseintheareaofpopulationhealth.FutureAIsystemscanmineandanalysedatacontained in largeelectronicpatient record systems toenablemore fine-grainedandpersonalizeddiagnosesandtreatments.68Theymayhavethecapabilitytosuggesttreatmentoptionsbasedontheanalysisofamedicallysimilarcohortofpatients.Somewhatfartherintothefuture,theymayalsoallowforthediscoveryofnumerousgenotype-phenotypeconnections,evenatthelevelofindividuals,thusenablingevenmoretargetedandpersonalizedtreatmentoptions.Machinelearningmaybeusedtoexaminehowpatientsrespondtomedicinesonageneticlevelandwhatmedicinalchangesneedtobemadetoenhancepatientwell-being.

Finance and insurance

Presentapplications.Inthefinancialindustry,AIiscurrentlybeingusedinanumberofways.Itisusedbylargeinstitutionalinvestorsinalgorithmictradingandhigh-frequencytrading,whichinvolvestheuseofhighlycomplexAIsystemsthatcanperformmillionsof(low-margin)tradesadaywithouthumanintervention.69 Inaddition,AI isused for financialmarketanalysis. Large financial institutionshaveinvestedinAIsystems(e.g.,Aladdin70,Sqreem71,Kensho72)toassistwiththeir investmentpracticesandthoseoftheirclients.Suchsystemsusebigdata,machinelearningandnaturallanguageprocessingtechniquestogatherandanalysefinancialnews,brokerreports,socialmediafeeds,andothersources,in order to assign ratings to potential investments. AI is also used in so-called robo-advisors that

65Simonite,Tom,“AICanHelpAppleWatchPredictHighBloodPressure,SleepApnea,”Wired,November13,2017.http://wired.com/story/ai-can-help-apple-watch-predict-high-blood-pressure-sleep-apnea/66Galeon,DomandKristinHouser,“IBM’sWatsonAIRecommendsSameTreatmentasDoctorsin99%ofCancerCases,”Futurism,October28,2016.http://futurism.com/ibms-watson-ai-recommends-same-treatment-as-doctors-in-99-of-cancer-cases/67Shin,Hoo-Chang,HolgerR.Roth,MingchenGao,etal.,“DeepConvolutionalNeuralNetworksforComputer-aidedDetection:CNNArchitectures,DatasetCharacteristicsandTransferLearning,”IEEETransactionsonMedicalImagingVol.35,No.5,2016,pp.1285–1298.68Ibid.69Foramoredetaileddefinitionseehttp://investopedia.com/terms/a/algorithmictrading.asp70FormoreinformationonAladdinsee:http://blackrock.com/aladdin/offerings/aladdin-overview71FormoreinformationonSqreemsee:http://home.sqreem.com/home/?72FormoreinformationonKenshoGlobalAnalyticssee:http://kensho.com/


35

provideautomatedfinancialadviceandinportfoliomanagementservices.TheAIsystemsherecantailortheiradviceandmanagementservicestotheinvestmentgoalsandthelevelofrisktoleranceofafinancialcompany’sclientsandcanadjustinreal-timefashiontochangesinthemarketandmodifyportfoliosaccordingly.73Furthermore,AIisbeingusedforunderwritingpurposesinthecreditindustry.Lenders are using machine learning techniques (e.g., Zest Automated Machine Learning byZestFinance74)toanalysealargevarietyofvariables–frompurchasetransactionstothemannerinwhichacustomerfillsoutaform–inordertodevelopriskmodelsandassignscorestoborrowers.Finally,inpersonalfinance,severalproductshaveemergedthatutilizeAItoassistpeoplewiththeirpersonalfinances,includingoptimizationoftheirspendingandsavingpractices.75

InsuranceprovidershavealsobeguntouseAIsystems.Theyhaveautomatedsomeaspectsoftheirclaims processes to reduce costs, improve underwriting, improve customer experience and fightfraudulent claims.76 Instead of relying on humans to manually comb through reports to catchfraudulentclaims,insurersnowoftenemployAIalgorithmsthatcanidentifypatternsinclaimsdataandrecognizeattemptsatfraud.

Futureapplications.Inthefinancialindustry,mostmarketparticipantsexpectthatAIwillbeadoptedfurther.77 In the future, the financial industry will see continued use of sophisticated algorithmictrading algorithms on the stockmarket involving an ever-greater percentage ofmarket volume, ifcurrenttrendsareexpectedtocontinue.78Inaddition,AIsystemswillincreasinglybeusedtogeneratenewsstoriesabouteconomicstatisticsandcompanyearningsresultsasthesearereleased;andthisinformationwillincreasinglyformdirectfeedsintootherAIsystemsthatwilltradeonit.AIsystemsinfinancemayalsousenewsourcesofinformation,includingsatellitedataandotherbigdata,inmakingmarket predictions.79 Furthermore, theuseofAI in fraud andmoney-launderingdetection, capitaloptimisation and portfolio management is expected to become more commonplace and developfurther. In personal finance, conversational AI systems or chatbots (based on natural languageprocessing and deep neural networks) will find use on people’s smartphones, supporting naturallanguageflowssimilartowhatpeoplehavewiththeirpersonalbankers.80

73Faggella,Daniel,“MachineLearninginFinanace—PresentandFutureApplications,”TechEmergence,March27,2018.http://techemergence.com/machine-learning-in-finance-applications/74Arvizo,Sarah,“ZestFinanceIntroducesMachineLearningPlatformtoUnderwriteMillennialsandOtherConsumerswithLimitedCreditHistory,”BusinessWire,February14,2017.http://businesswire.com/news/home/20170214005357/en/ZestFinance-Introduces-Machine-Learning-Platform-Underwrite-Millennials75Kaushik,Preetam,“IsArtificalIntelligencethewayForwardforPersonalFinance,”Wired.http://wired.com/insights/2014/02/artificial-intelligence-way-forward-personal-finance/76Morgan,Blake,“HowArtificialIntelligenceWillImpacttheInsuranceIndustry,”Forbes,July25,2017.http://forbes.com/sites/blakemorgan/2017/07/25/how-artificial-intelligence-will-impact-the-insurance-industry/#5255ab2e653177FinancialStabilityBoard,ArtificialIntelligenceandMachineLearninginFinancialServices,November1,2017.http://fsb.org/wp-content/uploads/P011117.pdf78Glant,Morton,andRobertKissell,Multi-AssetRiskModeling:TechniquesforaGlobalEconomyinanElectronicandAlgorithmicTrading,AcademicPress,2014,pp.258.79Faggella,Daniel,“FromPasttoFuture,TracingtheEvolutionaryPathjofFinTech—AConversationwithBradBailey,TechEmergence,December2,2017.http://techemergence.com/from-past-to-future-tracing-the-evolutionary-path-of-fintech/80Chowdhry,Amit,“HowArtificialIntelligenceisGoingtoAffecttheFinancialIndustryin2018,Forbes,February26,2018.http://forbes.com/sites/amitchowdhry/2018/02/26/clinc-artificial-intelligence/#80f299f481f9


36

Intheinsuranceindustry,chatbotswillfindincreaseduseindealingwithclients’basicquestionandclaims, aswell as in the sellingofproducts andensuringproper coverageof clients.76 In addition,insurancecompanies’AIbotswillbeusedtoscancustomers’socialprofilestofindpatternandtrendsthatcan improvetheunderwritingprocess (fromthe insurerspointofview).81For instance,suchasystemcouldrecommendlowercarinsurancepremiumsforapersonthathasahealthylifestyleanda steady job, since these factors canbe linked tobeinga saferdriver. Finally, telematics (e.g., thewirelesscommunicationandsubsequentanalysisofdrivingdata)isexpectedtobeasignificantareaofgrowth.82

Security

Presentapplications.CurrentapplicationsofAIinsecurityarefocusedaroundnationaldefenceandlawenforcement.Forexample,AIisalreadybeingutilizedinlawenforcementandbordercontrolbymany federalagencies,and it ispredicted that theuseofAI in theseareaswillonlybecomemoreubiquitous.CurrentexpansionstotheuseofAIinsecurityincludecamerasanddronesforsurveillanceandbiomonitoring,algorithms todetect financial fraud,andpredictivepolicing.Additionally, thereseemstobemuchinterestintheinvolvementofprivatecorporationsandR&DprojectsinvolvingthedatagatheredfromAIinsecurity.83AfewofthemostprominentprivateChinesecompaniesinthiscasebeingBaidu,AlibabaandTencent,andadynamicstart-upecosystem,includingiFlytek–aleaderin speech recognition technology, and SenseTime – focusing on innovative computer vision, havebecomegloballeadersinAI84.Baidu,forinstance,spentbillionsonR&DinAI,successfullylaunchingitssuperintelligentvoiceinteractionsystemDuerOS,andanadvancedautonomousdrivingplatformApollo.85 As aworld leader in technology-surveillance, it is no surprise that China provides an aptexamplefordominatingusesofAIinsecurity86.

In termsof lawenforcement, scientists inEuropearecurrentlydevelopingasystemcalledVALCRI,Visual Analytics for Sense-Making in Criminal Intelligence Analysis. VALCRI combines notable AIapplicationssuchasfacialrecognition,objectrecognition,andsentimentanalysistomakemorein-depthconnectionsbeyondwhathumaninvestigatorscanmanage87.Canada-basedMagnetForensicsprovidesamore in-depth lookatMachineLearning indigital forensicswiththeir InternetEvidenceFinder(IEF).IEFisasoftwareplatformthathelpsinvestigatorsfindandanalysedigitalevidenceacrossmultiple sources. Its newest offering isMagnet.AI—a contextual tool that usesMachine Learningtechniquestosiftthroughdatafromcomputersandmobiledevices.Morespecifically,Magnet.AIis81Ibid.82Ibid.83Kania,ElsaB.,“BattlefieldSingularity:ArtificialIntelligence,MilitaryRevolution,andChina’sFutureMilitaryPower”,CenterforNewAmericanSecurity,November2017.http://s3.amazonaws.com/files.cnas.org/documents/Battlefield-Singularity-November-2017.pdf?mtime=2017112923580484Muncaster,Phil,“SayHello,or你好,toChina’sSiri”,MITTechnologyReview,November16,2012.technologyreview.com/s/507416/say-hello-or-to-chinas-siri/.FormoreinformationonSenseTime,seetheirwebsite:http://sensetime.com/85Hempel,Jessi,“HowBaiduWillWinChina’sAIRace—and,Maybe,theWorld’s,”WIRED,August9,2017.http://wired.com/story/how-baidu-will-win-chinas-ai-raceand-maybe-the-worlds/86Deahl,Dani,“SuspectCaughtinChinaatMusicConcertAfterBeingDetectedbyFacialRecognitionTechnology,”TheVerge,April12,2018.http://theverge.com/2018/4/12/17229530/china-facial-recognition-technology-suspect-music-festival87Nanalyze,“8CompaniesUsingAIforLawEnforcement,”Nanalyze,November12,2017.http://nanalyze.com/2017/11/8-companies-ai-law-enforcement/


37

designedspecificallytoassist investigatorsworkingonchildexploitationcases.88Theseapplicationsraisetheethicalissuesofinnocentpeoplebeingunjustifiablymonitored,systematizinghumanbiases,andprotectionofcivilliberties,amongothers.

Future Applications. The future AI security applications are clearly geared towards the so-called“machine militaries”, profoundly transforming the international landscape of security innovation.While we already see the replacement of humans by machines in life-threatening situations, likeExplosive Ordinance Device (EOD) robots detecting and disabling IEDs, or Improvised ExplosiveDevices; in the future, humanswill likely delegate all dangerous,war-fighting tasks to robotic andautonomoustechnologieswithoutnecessarilyinterveningtocommanddecisions-making.89KeyfutureapplicationsofAI—includingvulnerabilitymonitoring,threatanalysis,andcyber-defencebolstering—will be critical to the future development of cyber security.90 Cyber offensewill also experience asignificantimprovementbyAI.NationsarmedwithcomplexAIcyberweapons,likeamachinelearningAdvancedPersistentThreat(APT)cyber-attack,willquicklyhuntsystemweaknessesandexecutehackand attack functions.91 Again looking at China for an example, the Chinese PLAwill leverageAI toenhanceitsfuturecapabilities,includinginintelligentandautonomousunmannedsystems;AI-enableddata fusion, informationprocessing,and intelligenceanalysis;war-gaming, simulationand training;defence, offense, and command in information warfare; as well as super intelligent support tocommanddecision-making.

Onthelawenforcementfront,anillustrativefutureapplicationcanbecapturedbytheIntel-poweredAI system which can help to find missing and exploited children. Intel has already designed anenterprisedatastrategyusingabasicmachinelearningmodelthatallowedthemtointegratemassivevolumeofdatafromhundredsofdifferentdatabasesanddatasources. Intel ishopefulthat futuremodelswill predictwhere to send reports to and continue to improveby usingMachine Learningalgorithmstoextrapolatedatafrompastcases,inconjunctionwithIPaddresses,phonenumbers,andtextstofindsuspectsmoreefficiently.92Ifpredictionalgorithmswillbeagreatstrideforjustice,secretexperimentsusingthesametechnologydefinitelywillnotbe.Scandals,suchastheNewOrleansPoliceDepartment,whoquietlyusedaSiliconValleycompanytopredictcrimes,alertedthatwhileittakestimeforpredictivetechnologiestoberefinedandadvanced,predictivemachinelearningtoolswithaccess to large amount of private and sensitive data is certainly not a far-fetched futuristic sci-fiscenario.

Retail and marketing

Presentapplications.AIapplicationsinretailandmarketinghavemanyoverlaps.Withsharedgoalsofcost-reduction, optimization, personalization, and performativity, it is no surprise that the mainoverlap iswithintheuseofrecommendersystems.Recommendersystemsareutilized insituationswhereusersarerequiredtomakeselectionsfromvastamountsofpotentialoptions—beitmovies,

88Ibid.89Piazza,Bailey,“MachineMilitaries:TheFutureofArtificialIntelligenceandNationalSecurity,”DiplomaticCourier,January23,2018.http://diplomaticourier.com/machine-militaries-future-artificial-intelligence-national-security/90Ibid.91Bailiey,op.cit,2018.92Intel,“Intel-poweredAIHelpsFindMissingChildren”,Intel,http://intel.com/content/www/us/en/analytics/artificial-intelligence/article/ai-helps-find-kids.html


38

products,events,etc.93Withinretail, recommendersystems implementingcognitivecomputingareused in chatbotsproviding concierge-like services for clientson thephoneoronline tohelp guidecustomers through purchasing decisions.94 Formarketing, the use of recommender systems takesplacemorewithinthedigitalenvironment—providinguserswithpersonalizedsuggestionsbasedoffof the content they have consumed on the platform as well as their browsing habits: rewinding,replaying,skipping,etc.95

OthernotableusesofAIinretailarefordelivery,customerservice,andpayment.AIisbeingusedtohelpsmoothandoptimizetheentireretailprocessfromplacingorderstodelivery.MachinelearningandlanguageprocessingtechniquesenablingAItoactassalesassistants,combiningwithuserdataandhabits to findnewwaysof re-engagingwithcustomersorcross-sellingproducts.Furthermore,combinations of these technologies, as is the casewithAmazon’s Echo, enables users to not onlyreceiverecommendationsfromanAI,butalsomakepurchasesandscheduledeliveries.Andwiththepopularization of instant-delivery (drones, autonomous delivery units, etc.), the assemblages ofvarious AI enable consumers to have a personalized, digital shopping experience from their ownhome.96

OneofthemostnotablecurrentapplicationsofAIinmarketingistheuseofprogrammicadvertising.ProgrammicadvertisingreliesuponAItobuyandselladvertisingdataandactasamiddlemanbetweenadvertisersandpublishers.TheseAIareabletoanalyse,buy,sell,andexchangeinrealtime,provinginvaluable to producers in optimizing the use of consumer data. Another application that aidsprogrammicadvertisingisthatofdataandtrendforecasting—whichAIalsoplaceacriticalroleinformarketers.AnotherofthemostrecognizableapplicationsisthatofsearchenginesoreCommerce—usingusersearchdatacombinedwithrecommendersystemstocustomizeandpushcertainadvertsat blanket or target demographics. AI used in conversational commerce has also generatedmuchattentionwithinrecentyears:usingvariousapplicationsofchatbotstoletusersmakepurchasesbyspeakingtomachine(AmazonEcho),chat(FacebookMessenger),orvirtualassistant(TheNorthface).95

Futureapplications.Thefutureofretailandmarketingseemtobefurtherintertwinedintothefuturewithpushesonbothendsforincreasedintegrationwithbiometricdataandincreasingcross-platformmarketingandretailexperiences.Toexplainmoreindetail,thefutureofretailandmarketingwilllikelyseefurtherintegrationbetweenconsumer’sbrowsingandshoppinghabitsacrossvariousplatforms(i.e.,AmazondatainfluencingadvertisinginPrimark).Additionally,cross-platformdatacompositeswillbeusedtopersonalizestorefrontadvertisingandshoppingguidancetoeachuser—recognizingusersthroughbiometricdataandin-storesurveillance.97

Further advances in marketing will be improvements in hitting target-audiences, reaching newconsumers,andmakingmarketingdecisionsonanexecutiveleveltofreeupthetimeofhumanCEOs.

93TolearnmoreaboutRecommendersystems,seethisexplanation:https://medium.com/recombee-blog/recommender-systems-explained-d98e8221f46894Faggella,Daniel,“ArtificialIntelligenceinRetail—10PresentandFutureUseCases,”TechEmergence,2018.http://techemergence.com/artificial-intelligence-retail/95Faggella,Daniel,“ArtificialIntelligenceinMarketingandAdvertising—5ExamplesofRealTraction,”TechEmergence,2018.http://techemergence.com/artificial-intelligence-in-marketing-and-advertising-5-examples-of-real-traction/96Sonsev,Veronika,“WillAIbetheFutureofRetail?”ForbesTech,December15,2017.forbes.com/sites/veronikasonsev/2017/12/15/predicting-the-future-of-retail-a-vc-perspective/#6c9941387ff197PSFKLabs,“5AITrendsRedefiningtheRetailExperience,”PSFK,2018.http://psfk.com/2018/01/5-ai-retail-trends-redefining-experience.html


39

This will potentially open up a much broader field of eConsultancy.98 Another much-anticipatedadvancement withinmarketing and advertising applications of AI is that of machine vision/imagerecognition forsearches.Thiswouldallowusers tosearchand findproductpricesand informationonlinebysnappingapictureoftheobjectwiththeirphone,ratherthanscanningabarcode.

Future advances anticipated in retail is that of more personalized shopping—with 3D-knittingprogressionsallowingcustomerstoordertailoredclothingondemandanddeliveredinreal-timebydrones or other autonomous delivery vehicles. Natural Language Processing techniques are alsoexpectedtoprovideamoralpersonableinteractionwithAIandimprovetheirabilitiestofactorinmorecomplexconsumerdemandswithouttheinterferenceofahumanconsultant.

Media and entertainment

Present applications. Themedia and entertainment industries are aboundwith applications of AI.Currently, the dominating goals of AI in the media industry are focused on marketing/analytics,personalization, andoptimization.99Whilemachine learning techniques take the lead on themostcommontypeofAIbeingusedinmediaandentertainment,othernotablesubcategoriesbeingutilizedinconjunctionwithmachinelearningarenaturallanguageprocessing,informationretrieval,imagineprocessing,crowdsourcing,andcollaborativefiltering.Furthermore,abstractionandrecommendationalgorithmsarealsobeingusedtofurthercatertowardscreatingamorepersonalizedexperiencefortheuserandincreasingmediaperformativityforcompanies.,100NotableexamplesofAIapplicationsinmedia are in targeted advertising, optimizing and ascertaining user engagement (smart filters andrecommender systems), and discovering new target audiences.101 While AI in the entertainmentindustryemploymanyofthesamesubcategoriesofAI,thegoalsaremoreorientedtowardsnovelty,consumer engagement, adaptability. Example here are the use of AI to make more creative andchallengingvideogameexperiencesand increase replayvaluebydecreasing thepredictabilityandincreasingresponsevarianceinNon-PlayerCharacters(NPCs).102

FutureApplications.WhileitiscertainthatfutureapplicationsofAIinmediaandentertainmentwillcontinue to advance in their goals of optimization, personalization, marketing, novelty, andengagement,itishardtoaccuratelypredictthefuturedirectionsofentertainmentandmedia.Afterall,thecontentandfocusofentertainmentandmediaarearoundthetastesandpreferencesoftheconsumers, which change and flow over time. However, the most likely future will see moreapplicationsofAI as co-creatorsandeven soledevelopersofmediaandentertainment content. ItseemslikelythatthisfieldwillbegintoleantowardsmakingpartnersofAI,ratherthanusingthemastools—pushingforAItousemoreadvancedmachinelearningtonotonlyincreaseoptimizationandperformativity, but to alsomore autonomously and actively contribute to the creation ofmovies,

98Stephen,Andrew,“AIisChangingMarketingasWeKnowIt,andThat’saGoodThing,”ForbesNetwork,October30,2017.http://forbes.com/sites/andrewstephen/2017/10/30/ai-is-changing-marketing-as-we-know-it-and-thats-a-good-thing/#2ea5197cdc4099Sennaar,Kumba,“AIinMovies,Entertainment,andVisualMedia—5CurrentUse-Case,”TechEmergence,November20,2017.http://techemergence.com/ai-in-movies-entertainment-visual-media/100“ApplyingAIandMachineLearningtoMediaandEntertainment,”TowardsDataScience,2017.http://towardsdatascience.com/applying-ai-and-machine-learning-to-media-and-entertainment-8d3ebc4b68f7101Foster,Alana,“ApplyingArtificialIntelligencefromProductiontoDelivery,”IBCTechAdvances,January23,2018.http://ibc.org/tech-advances/ai-transforming-media-production-/2638.article102Lou,Harbring,“AIinVideoGames:TowardaMoreIntelligentGame,”HarvardUniversity,August28,2017.http://sitn.hms.harvard.edu/flash/2017/ai-video-games-toward-intelligent-game/


40

music,art,andstories.,103Other futuredirectionsmaytakeentertainmentandmediaconsumptionoutsideofthedigitalworld,andintoimprovementsofvirtualassistantsinthehome,especiallywithfuture developments in speech recognition and emotional analysis. Enhancements to hardwareinvolving virtual reality and haptic feedbackmay also lead to a future of virtual reality being theentertainmentofchoice,withAIassistinginimmersionandnovelexperiences.

Science

Presentapplications.TheapplicationsofAIwithinsciencearenumerous.Especiallywhenworkingwithlarge (i.e., in the order of petabytes) or convoluted data, AI is able to help scientists discoverconnectionsorcombinationsthatwouldotherwisebeimpossibletodetect.ThedominatingtypeofAIappliedwithinthescientific fields thatutilizethemismachine learning,andmorespecificallydeeplearning.104Oneexampleofthisisinparticlephysics,wheremachinelearningalgorithmscanbeusedtosiftthroughdataonparticleshowersincollisionexperimentstofindtracesofhard-to-detectnever-before-seen participles—finding a needle in a haystack, essentially. The use of suchmethods hashelpedwiththereconstructionanddiscoveryoftheHiggsbosonintheLargeHadronCollider.105Largedataanalysisandpredictionshavealsoprovenquiteusefulinsocialscience—analysinguserdataanddecisionstodrawconclusionsaboutconsumerhabits.106

FutureApplications.TwofieldsthatareduetoseefarmoreAI involvementinthefuturearespaceexplorationandgenomics.Spaceexploration is takingaparticular interest in futureapplicationsofdistributed AI in (robotic) swarms and computing distribution, as well as in increasing situationalawarenessanddecisionsupportforspacecraft.107Furthermore,theuseofmachinelearningtopredictastronomicalphenomenalikesolarflaresormeteorshowerswillalsobeadesirablefuturedirectioninordertosavetime,moneyandeffort.108Additionally,AIcanbecombinedwithgenomicsinmedicalandagriculturalscience.Here,combiningmachinelearningandgeneticinformationto,forexample,examinehowpatientsrespondtomedicinesonagenetic level,orcreatediagnostic techniques formonitoringandimprovingsoilqualityandmaintaininghealthyandsustainableagriculturepractices.109Moregenerally, futureAI applications seem to focuson increasing thedataprocessingpowerandgrowingthelayersofdatathatdeeplearningcanprocess.

103Chu,Eric,JonathanDunn,andDebRoy,etal.,“AIinStorytelling:MachinesasCocreators,”McKinsey&CompanyMedia&Entertainment,December2017.http://mckinsey.com/industries/media-and-entertainment/our-insights/ai-in-storytelling104Appenzeller,Tim,“TheAIRevolutioninScience,”ScienceMagazine,July7,2017.http://sciencemag.org/news/2017/07/ai-revolution-science105ScienceNewsStaff,“AIisChangingHowWedoScience.GetaGlimpse,”ScienceMagazine,July5,2017.http://sciencemag.org/news/2017/07/ai-changing-how-we-do-science-get-glimpse106Faggella,Daniel,“AirbnbMachineLearning—HowDataandSocialScienceMakeitallWork,”TechEmergence,2017.http://techemergence.com/airbnb-machine-learning-how-data-and-social-science-make-it-all-work/107Girimonte,Daniela,andDarioIzzo,ArtificialIntelligenceforSpaceApplications,EuropeanSpaceAgency,2007.PDF.https://www.esa.int/gsp/ACT/doc/AI/pub/ACT-RPR-AI-2007-ArtificialIntelligenceForSpaceApplications.pdf108Bourzac,Katherine,“AIinSpace,”IEEERobotics,August21,2017.http://spectrum.ieee.org/tech-talk/robotics/artificial-intelligence/ai-in-space109Sennaar,Kumba,“MachineLearninginGenomics—CurrentEffortsandFutureApplications,”TechEmergence,2018.http://techemergence.com/machine-learning-in-genomics-applications/


41

Education

Present applications. Although there aremany opportunities for AI to be used in education, highbarrierstoentryintermsofcostandeffectivenesspreventwidespreadadoptionofmanypossibleAIapplications in this field. The goals, however, of AI in education are increasing accessibility,collaboration, and flexibility and combating cost andopportunity restraints.NotableAI techniquesbeingappliedineducationaremachinelearning,crowdsourcing,andnaturallanguageprocessing.TheuseofthesetechniquesinconjunctionallowsforthecreationofIntelligentTutoringSystems(ITS)thatcan supplement face-to-face instruction and provide a more personalized learning experience. AItechniquesarealsousedinadministrativetasksofevaluation,grading,advising,andscheduling.110,111

Future Applications.Given that field of AI as a whole is rife with hype and sometimes large gapsbetweenpromisesandreality,manyofthefutureexpectationsofAIineducationstandoutasbeingmoremodestandpragmatic.SomeoftheexpectedfuturegainsofAIineducationwillresultfromtheuseofAItechniquesinconjunctionwithbigdataanalytictoolsandrecommendationsystemsthatareusedonstudents’performativedata.Thiswillhelpeducatorsoptimizestudents’educationtrajectoriesbybetterunderstandingandfindinglearninggapsandadjustingtothem,thusenhancingthepotentialofthestudentstolearnbymasteryofasubjectmoreeasilyandbuilduponcumulativeinformationmoreeffectively.Furthermore,AImay increaseexperiential learning in theclassroom—learningbydoing—bycreatingamoreinteractivelearningenvironmentthroughvirtualrealityorotherimmersivetechniques.112 Rather than focusing on replacing or removing educators from their interpersonalduties,AIinthefuturewillfocusonalleviatingsomeofthemoretediousandtime-consumingtasksteachers face (lessonplanning,grading,scheduling,etc.)sothat teachersare freer to interactwithstudentspersonally.113,114

Manufacturing

AIuseinmanufacturingrobotsisdiscussedinsubsection3.2

MuchoftheuseofAItechniquesinthemanufacturingsectorcentresonapplicationsinrobotics(seesubsection 3.2 on robotics in industry). Outside of robotics, however, a number of trends can bedistinguishedinthedevelopmentanduseofAIsystemsinmanufacturing.ThefirstistheuseofAIinsystems for predictive maintenance of industrial equipment. Pieces large-scale industrialmanufacturing equipment are beginning to be instrumented with sensors, networked with oneanother, andvirtuallymodelledwithaccuratephysics (creating so-called “digital twins”), such thattheycanbeconstantlymonitoredandpredictiveanalysescanbeperformedontheirperformance.115

110Faggella,Daniel,“ExamplesofArtificialIntelligenceinEducation,”TechEmergence,2017.http://techemergence.com/examples-of-artificial-intelligence-in-education/111Schmidt,Adrien,“HowAIImpactsEducation,”Forbes,December27,2017.http://forbes.com/sites/theyec/2017/12/27/how-ai-impacts-education/#5e93879c792e112Rizzotto,Lucas,“TheFutureofEducation:HowAIandImmersiveTechWillReshapeLearningForever,”MediumFuturePi,June23,2017.http://medium.com/futurepi/a-vision-for-education-and-its-immersive-a-i-driven-future-b5a9d34ce26d113Dickson,Ben,“HowArtificialIntelligenceisShapingtheFutureofEducation,”PCMag,November20,2017.http://pcmag.com/article/357483/how-artificial-intelligence-is-shaping-the-future-of-educati114Houser,Kristin,“TheSolutiontoOurEducationCrisisMightbeAI,”Futurism,December11,2017.http://futurism.com/ai-teachers-education-crisis/115Yao,Mariya,“FutureFactories:HowAIEnablesSmartManufacturing,”TopBots,October27,2017.http://medium.com/topbots/future-factories-how-ai-enables-smart-manufacturing-c1405f4ec0e6


42

Thisdevelopmentprovidesthemanufacturingindustrywithimportantbenefitsintermsofreductionsinwastedlabourandtheriskofunexpectedandundiagnosedequipmentfailures.AsecondtrendistheuseofAIinautomatedqualitycontrol.High-definitioncamerasarecombinedwithcomputervisionanalysis to immediately spot defects and identify root causes of failure in production lines, thusensuringthatexpensivedelaysinproductionarekepttoaminimum.AthirdandfinaltrendistheuseofAIindemand-drivenproduction.116Consumertrendsandbehaviouraldataareincreasinglyfedintodirectfeedbacklooponsupplychainandmanufacturingactivities(e.g.,throughtheuseofsmarthomedevicessuchasAmazonEcho,whichproducesdatathatfeedsbackintoAmazon’ssupplychain),thusreducingtheriskofoverandunderproduction.117

Agriculture

AIuseinagriculturalrobotsisdiscussedinsubsection3.2

MuchoftheuseofAItechniquesintheagriculturalsectorisfocusedonapplicationsinrobotics(seesubsection3.2onrobotics inagriculture).Outsideofrobotics,however,anumberoftrendscanbeidentifiedinthedevelopmentanduseofAIsystemsinagriculture.ThefirstistheuseofAIinsystemsforcropandsoilhealthmonitoring.Deeplearningsystemsarebeingdevelopedthatcanidentify(andhelp mitigate) potential defects and nutrient deficiencies in soil through the camera of a user’ssmartphone.Ananalysisofacquiredimagedataisconductedbythedeeplearningsystem,whichlinkspatternsinthefoliagewithspecificplantdiseases,pestsandsoildefects.118AsecondtrendistheuseofAIinpredictiveanalytics.Here,machinelearningalgorithmsarebeingdevelopedthatcanworkwithsatellitedatatopredictweather,analysecropsustainabilityandevaluatefarmlandsforthepresenceofdiseasesandpests.Thesepredictionscanbeverylocalisedandtailoredtotheneedsofindividualclients.119 Somewhat farther into the future, it is expected that genomics is more widely used inagricultural applications, combiningmachine learning and genetic information to create diagnostictechniques to monitor and improve soil quality and maintain healthy and sustainable agriculturepractices.120

3.2 Robotics

Thissubsectionoffersanoverviewofimportantpresentandexpectedfutureapplicationsofrobotics.Applications are discussed for the following application domains: transportation, security,infrastructure,healthcare,companionship,industry,discovery,service,environmentandagriculture.

Transportation

Presentapplications.Automobilesandtrucksare increasinglybeingequippedwithmore intelligenttechnologies,includingintelligentbrakingandparkingsystems,parkingsystems,proximityandforcesensors,navigationsystemsandadvancedmotioncontrols.Theemergenceofautonomouscars(alsocalleddriverlesscarsorroboticcars)andautonomoustrucksgoesastepfurther—Ityieldsvehiclesthatarebothintelligentandautonomous.Autonomouscarsandtrucksarecurrentlystillinprototyping116Ibid.117Synchrono,By,“2017TopTenTrendsinModernDemand-DrivenManufacturing,”SupplyChain247,June14,2017.http://supplychain247.com/paper/2017_top_ten_trends_in_modern_demand_driven_manufacturing118Sennaar,Kumba,“AIinAgriculture—PresentApplicationsandImpact,”TechEmergence,November16,2017.http://techemergence.com/ai-agriculture-present-applications-impact/119Ibid.120Sennaar,Kumba,“MachineLearninginGenomics—CurrentEffortsandFutureApplications,”TechEmergence,2018.http://techemergence.com/machine-learning-in-genomics-applications/


43

but are expected to become available for the consumer and business markets soon. In publictransportation,autonomoustrainsandsubwayssystemsarealreadyareality,andhavebeenforquitesometime,althoughonlyonasmallscale.Suchsystemsareeasiertoimplementintrainsandsubwaysthanincarsandtrucksbecausetheirnavigationcomplexityislower(astheyareboundtofixedtracks).

Wheeled and legged cargo robots provide solutions for transportation of small andmedium-sizedcargoloadsoversmallerdistances,uptoafewkilometres.Small-sizedtransportrobotscanmoveloadsofupto200kilograms.121Differenttypesofwarehouserobotsarecapableofshelfstockingandmovingheavyitemsandhavefoundincreasinguseinwarehouses.Exoskeletalrobots,finally,arebeginningtobeusedtoassisthumanswithliftingandtransportingheavycargoandequipment.

Inaviation,UnmannedAerialVehicles (UAVs)are increasinglybeingused for the transportationofsmallshipments.Inaddition,UAVsareusedintheanalysisofurbantransportationflowsbecauseoftheir aerial visualisation and gathering information capabilities. The increasingly pressing needs indealing effectively and directlywith road congestion can bemetwell byUAVs,which have foundsignificantuseinthefieldoftrafficengineering.122

Future Applications. Current autonomous vehicle developers have made bold claims about thedevelopment of autonomous vehicles. Tesla, Inc., NissanMotor Co. and BMWhave promised thedelivery of a completely autonomous production cars by the end of 2018, 2022 and 2021,respectively.123,124,125Autonomousvehiclesareexpectedtobeamajorgrowthmarket.Whenreadyfortheconsumerandbusinessmarkets,theyareexpectedtorevolutionizegroundtransportation.Infullyautonomouscars,usersmaybeable toengage inworkor leisureactivities rather than focus theirattentionontheroad.Peoplewillalsobeabletocallupdriverlesstaxiseasilyorsharecarswithgroupsofowners.126JustasUberserviceshavingincreasedvehicleutilizationanddiminishedthenecessityofvehicle ownership, self-driving cars will decrease the need for individuals to own a car. Theintroductionofsignificantnumbersof fullyautonomouscarsthereforehasthepotentialdrasticallydecreasethenumberofvehiclesontheroads.

Furthermore,driverless trucks, smaller cargo robots andUAVsmay revolutionize transportationofgoods.Experimentswithsurfacedeliveryrobotsandwithdrone-deliveryservicesforretailstoressuggestthatthesemaystarttobecomeavailableinthenext5-10years.127Suchsystemswillenablerapiddeliveryofgoodsondemand.Inaddition,autonomoustruckswillallowforfasterandcheaperdeliveryandmay be broken up into different smaller autonomous vehicles along the route so as to optimize

121ForexamplesseeSMPRobotics:http://smprobotics.com/application_autonomus_mobile_robots/transport-robots/122Barmpounakis,EmmanouilN.,EleniI.Vlahogianni,andJohnC.Golias,“UnmannedAerialAircraftSystemsforTransportationEngineering:CurrentPracticeandFutureChallenges,”InternationalJournalofTransportationScienceandTechnology,Vol.5,No.3,October2016,pp.111-122.PDF123Wales,Jimmy,“ElonMuskWillMakeDriverlessCarsaRealitySoonerthanyouThink,”WIRED,January18,2018.http://wired.co.uk/article/fully-autonomous-cars-are-almost-here124Ma,Jie,“NissanPlanstoIntroduceFullyAutonomousDrivingCarsin2022,”BloombergTechnology,December6,2017.http://bloomberg.com/news/articles/2017-12-06/nissan-plans-to-introduce-fully-autonomous-driving-cars-in-2022125Muoio,Danielle,“BMWPlanstoTakeonMercedesbyReleasingaFullyDriverlessCarby2021,”BusinessInsider,March17,2017.http://businessinsider.nl/bmw-to-rival-mercedes-with-level-5-driverless-car-in-2021-2017-3/?international=true&r=US126Davis,Alex,“MercedesPromisesSelf-DrivingTaxisinJustThreeYears,”WIRED,April2,2017.http://wired.com/2017/04/mercedes-promises-self-driving-taxis-just-three-years/127Cohn,Pamela,AlastairGreen,MeredithLangstaff,andMelanieRoller,“CommercialDronesareHere:TheFutureofUnmannedAerialSystems,”McKinsey&Co.,December2017.http://mckinsey.com/industries/capital-projects-and-infrastructure/our-insights/commercial-drones-are-here-the-future-of-unmanned-aerial-systems


44

distributionprocesses.Inthemoredistantfuture(20yearsormorefromnow),thefirstautonomousairtaxisforcommutersandpilotlesspassengerjetscanbeexpected.

Security

Presentapplications.Militaryapplicationsofroboticsarewide-reaching.Roboticsystemsarecurrentlybeing used for reconnaissance and surveillance, combat and engagement of military targets, thelocation, handling and destruction of hazardous objects, search and recovery, and breaching.Reconnaissance and surveillance ismainly tasked to UAVs, but there are also ground robotswithsimilar capabilities. For combat,many robots are equippedwith remotely controlledweapons butsomecanoperateautonomously.TheSamsungSGR-A1isasentryrobotthatrecognizesandtargetspersons and fires at them unless a proper access code is supplied through voice recognition. TheGoalkeeperisaclose-inweaponsystemforuseonshipsthatcanautonomouslydetectanddestroymissiles,aircraftandsurfacevesselsthatareincloseproximity.Inaddition,therearemanysystemsonthemarketthatarecapableofhandlinganddestroyingbombsandofsearchingandrecovering.

Inpolicingandlawenforcement,UAVshavebeguntobeintroducedforsurveillanceandoccasionallyfordispersingcrowdsthroughtheuseofteargas,paintballsorothermeans.Groundrobotsarealsobeingusedforsurveillanceandreconnaissance,guarddutyandpatrol inprisonsandpublicplaces.Sincesecurityrobotsarecapableofcontinuouslypatrollingwithoutgettingtiredorneedingtorest,theuninterrupted,round-the-clockroboticsecuritydutyoperationhasbecomeaneffectivewaytocutsecurity service costs.Cooperationbetweenmobile security robotsand securityofficersallows forefficientallocationofhumanresourcesforincriticalsituations,ratherthanincurringtheexpenseandburdenofmanagingsecuritypersonneltoconductroutinepatrols.

Robots equippedwith non-lethalweapons are capable of physically countering threats to securitywithoutthepresenceofasecurityofficeronsite.Bombdisposalrobotshavebeenusedtosafelydetectanddisposeexplosivesordnanceforover40years,includinganythingfromlandminestounexplodedmunitions.128Theserobotsarecontrolledbyoperatorsfromasafedistanceandoperateasaremotepresenceforthebombdisposalexperts.DronesthatcancatchdroneswithnetsareoneofthelatestroboticinnovationsforlawenforcementannouncedbytheOrganizersofthe2018WinterOlympicsinSouthKorea.Theyareasafetyprecautiontodefeatunauthorizeddronespotentiallyusedtodeliverabomb.129

Search&Rescue/Recoveryrobotsplayan increasingly importantrole inthemilitaryandinpolicing(includingdisasterreliefandpost-disasterreconstructionefforts).Theserobotsarebeingdeployedtocarryouttasksnormallyregardedasthe“threeDs”—dull,dirtyanddangerous.Forexample,searchrobotscouldstartsearchinginplaceswherefireorsmokeinstillpresentorothersituationswherehumanrescuersarehelpless.Small,unmannedgroundvehiclescouldpenetrate the rubbledeeperthan humans or canine rescuers, enabling the rescuers to find victims at a faster rate. Larger,unmannedgroundvehiclescouldalsobeusedforlogisticspurposes—alleviatingdisastermanagementburdens.

FutureApplications.Withadvancementsinroboticsandremote-controlsystems,futurerobotsinthesecurityfieldwillbeincreasinglyadaptabletotheirenvironments.Forexample,thereareprototypesbombdisposalrobotsbeingdevelopedthatareabletojumpoverwallsandlandontheotherside.

128Allison,PeterRay,“WhatDoesaBombDisposalRobotActuallyDo?”BBC,July15,2016.http://bbc.com/future/story/20160714-what-does-a-bomb-disposal-robot-actually-do129Pettit,Harry,“Drone-CatchingDronesArmedwithNetscouldbeUsedtoBolsterSecurityatthisWeek’sPyeongchangWinterOlympics,”Dailymail,February6,2018.http://dailymail.co.uk/sciencetech/article-5357575/Drone-catching-drones-armed-nets-hit-Winter-Olympics.html


45

Robotsarealsobeingdevelopedthathavetwoarms,enablinggreatmanualdexterity,suchasallowingthemtoopencarbootsandlookinside.Ratherthanusingasinglerobot,function-specificrobotsareemerging,andwilloperatetogetherinteams.

Military exoskeletons of various sorts are currently being tested to help military personnel carrygreaterweights,runfaster,andincreaseendurance.Thedefenceindustrydemandsever-lightergears,withever-longerbatterylife,andclearadvantagestothetroops.130Askeletonsimilartothepopularlyknown“IronManSuit”,designedtoprotectcommandosfrombulletsandblastsastheybustdowndoors incounterterrorismmissions, iscurrentlyunderdevelopment.131 In lawenforcement,groundrobotsarebeingconsideredthatcanfulfilfunctionsofarrestanddetainmentandshowofforce.Onecanreasonablyexpectthat,inthesoon-to-comefuture,atleastsomerobotsusedbythepolicewillbeintelligentmachinescapableofusingcoerciveforceagainsthumanbeings.Suchpolicerobotsmaydecrease dangers to police officers by removing them from potentially volatile situations. Thosesuspected of crimes may also risk less injury if robots can assist the police in conducting saferdetentions,arrestsandsearches.132Theyareunderstandablycontroversial,however.

Infrastructure

Presentapplications.Themostpopularrobotscurrentlyinuseforinfrastructuralapplicationsaremicroaerial vehicles (MAVs).MAVs are a type of unmanned aerial vehicles (UAV) that can be used forentertainment (consumer-grade drones), combat (military-grade drones), or, in this case,infrastructureinspection.133CurrentapplicationsofnoteforMAVsarethecommunicationsectortomaintaintelephonelinesandcelltowers,aswellasprovidetemporarycellular“hotspots”toreducetheloadonexisting infrastructureorprovidecellcoveragetoareaswithout.Energycompaniesareusing MAVs to inspect power lines, wind turbines, and assess storm damage.134 Additionally,municipalitiesareusingMAVstoinspecttheinfrastructuralintegrityofbridges,roads,andotherlargestructuresthatpresentmanycostsandrisksforhumaninspectors.135

Anotherareaofnoteisrobotsforcollaborativeconstruction.Robotsinthiscategorycanaidhumansin difficult or tedious building tasks. The most popular robots of this type are those created byConstructionRobotics.TheSAM100(Semi-AutomatedMason)workswithhumanmasonsby layinglaysthebricksforthemasonsonaconstructionproject,whilethemasonsworkbehindthemachinetofinishandsealthebrickwork.Thisincreasesthespeedandconsistencyoflayingbrickswhileallowingthe humanmason to perform themore skilful workwithout fatiguing. Construction Robotics alsocreates another collaborative construction robot called theMule. TheMulehelpswithperformingheavyliftingonconstructionsites—preventingwastefrombreakage,injuries,andspeed.136Thefocusfor robotsof thisnature is to remove the tediousnessof repetitive tasksand thephysicalharmofrepetitiveorheavywork,withoutreplacingtheskillofhumanconstructionworkersand increasingproductivity.

130Marinov,Bobby,“MilitaryWearableExoskeletonResearch,”ExoskeletonReport,October3,2016.http://exoskeletonreport.com/2016/10/military-wearable-exoskeleton-research/131Machi,Vivienne,“’IronMan’SuitonTrack,butHurdlesRemain,”NationalDefense,May22,2017.http://nationaldefensemagazine.org/articles/2017/5/22/iron-man-suit-on-track-but-hurdles-remain132Joh,ElizabethE.,“PolicingPoliceRobots,”UCLALawReview,2016.133Dijkshoorn,Nick,“SimultaneousLocalizationandMappingwiththeAR.Drone,”UniversiteitvanAmsterdam,July14,2012.PDF.134Walker,Jon,“IndustrialUsesofDrones—5CurrentBusinessApplications,”TechEmergence,July6,2017.http://techemergence.com/industrial-uses-of-drones-applications/135UK-RAS,RoboticandAutonomousSystemsforResilientInfrastructure,UK-RASNetwork,2017.PDF.136LearnmoreaboutConstructionRoboticshereunder“Products”:http://construction-robotics.com/mule/#


46

FutureApplications.Thefutureofinfrastructuralroboticsisripewithopportunities.Whilethepushfor“smart-cities”providesincreasingdemandsforrobotsthatcanperformmoreparticularmonitoringtasks:reportingonwatermains,inspectingoilpipelines,androadintegrity.137Alongwiththis,therearealsocallsforusinginfrastructuralroboticstobuild“resilientcities”or“self-repairingcities.”Thisinvolves increasing the capabilities ofMAVs towork collaborativelywith one another to not onlyimprovetheaccuracyofthedatacollectedoninfrastructural integritybutalsoincreasethequality.These improvedMAVswoulduse3Dmodelling to assess problemsautonomously (rather than viacamera)and increase thecomplexityof theassessment, likeCarnegieMellonUniversity’sRoboticsInstituteseekstodowiththeirARIAProject.138

Additionally, “PerceiveandPatch”and “PerchandRepair”designs areexploring thepossibility forfutureinfrastructuralrobotstonotonlyfindandmapbreakdownsof infrastructurebutalsofixthebreakdownspreventativelyorimmediatelyaftertheyoccur.Potentialapplicationsforthesetypesofrobots are projected to be in pothole repair and road maintenance, pipe-repair, and bridgemaintenanceandrepair.139Moreover,“ConstructandConfirm”robotsareenvisionedtosurpassthecollaborative construction robots to be able to construct large and small structures autonomouslyusingCADmodelsand3Dprinting.OnesuchexampleisMITMediaLab’sDigitalConstructionPlatform(DCP)thatusessprayfoamtoconstructstructureson-demand,atanylocation.140Theinfrastructuralrobotsofthefuturetonotonlymaintainandmonitorexistinginfrastructurebuttocreateanddesignnewinfrastructurequicklyandcheaply.

Italsoseemsentirelypossiblethatexosuitswillalsocomeintouseininfrastructuralmaintenancewiththeir increasing popularity in industrial and healthcare applications. These exosuitswill be able toreduceworkerstrain,improvegripstrength,andallowworkerstoliftheavierloadsandperformmorestrenuoustasksthanbefore.141Accordingly,thegoalsoftheseexosuitsalsoalignwellwiththegoalsofinfrastructuralroboticsaswell:increasesafetyandinfrastructuralconditionswithasmuchspeedandlittlewasteaspossible.Forthejobsininfrastructurethatcannotbecompletedentirelybyrobots,itstillseemsthatthereisasubstantialchanceofroboticassistancethroughanexoskeleton.

Healthcare

Present applications. As oneof the top two fastest growing industries for robotics, thehealthcareindustryhasnoshortageofopportunitiesforrobotassistantsandcollaboration142.Oneofthemostwell-knownrobotscurrentlyinuseisthedaVinciSurgicalSystemfromIntuitiveSurgicalthatbelongstoaclassofrobotscalledteleroboticassistants.Theserobotsareusedtoenhanceahumansurgeon’scapabilities in the operating room by allowing the surgeon to complete certain operations more137Matthews,Kayla,“ThreeWaysRobotsCouldImproveInfrastructureinDevelopingCountries,”EngineeringforChange,September10,2017.http://engineeringforchange.org/news/robots-could-improve-infrastructure-developing-countries/138MoreinformationaboutCarnegieMellon’sARIAProjecthere:http://aria.ri.cmu.edu/about-aria139HighwaysReporters,“RobotPotholeFillersBeingDeveloped,”TransportNetwork,February11,2015.http://highwaysmagazine.co.uk/robot-pothole-fillers-being-developed/140Ackerman,Evan,“RoboticConstructionPlatformCreatesLargeBuildingsonDemand,”IEEESpectrumRobotics,April26,2017.http://spectrum.ieee.org/automaton/robotics/industrial-robots/robotic-construction-platform-creates-large-buildings-on-demand141Zingman,Alissa,ScottEarnest,BrianLowe,andChristineBranche,“ExoskeletonsinConstruction:WillTheyReduceorCreateHazards?”CentersforDiseaseControlandPrevention,June15,2017.http://blogs.cdc.gov/niosh-science-blog/2017/06/15/exoskeletons-in-construction/142Goepfert,JessicaandMichaelShirer,“IDCForecastsWorldwideSpendingonRoboticstoReach$135Billionin2019DrivenbyStrongSpendingGrowthinManufacturingandHealthcare,”BusinessWire,February24,2016.http://businesswire.com/news/home/20160224005169/en/IDC-Forecasts-Worldwide-Spending-Robotics-Reach-135


47

quickly,withgreaterprecision,andminimalinvasion.Theserobotsdonotoperateautonomously,butinstead,theirmovementsarecontrolledbyasurgeonoperatorthroughtheuseofajoystickorothercontroller.143Asimilar typeofrobotcalledasurgicalcobot (short forcollaborativerobot),whereasurgicalrobotdoessmalltasksautonomously,butcloselydirectedandalongsideahumansurgeon.Forinstance,suturingupacutasthesurgeoncontrolstheplacementanddirectionintermittently144.

Robotassistantsareanotherprominentapplicationforroboticsinthehealthcarefieldatthemoment.Theserobotsassistwithtaskslikewelcominganddirectingpatients,deliveringmaterials,andmovingpatients.TheTUGrobotbyAethon,forexample,canautonomouslytransportlinens,sterilematerials,food, or prescriptions to hospital rooms without interference from nurses or other hospitalpersonnel—reducing staff needed and increasing the speed of delivery.145 To add, RIKEN andSumitomo Riko Company Limited have designed a robot named “Robear” who can lift andmovepatientstosavenursesfromrepeatingthisarduoustaskconstantly.146Foramorepersonalroboticexperience,onecanlooktosocialroboticassistants(SARs).OnecouldtakeSoftBank’sPepperasanexample.Pepperisahumanoidrobotichospitalreceptionistthatcangreetpatients,showpatientstotheirrightlocationinthehospital,andspeaktwentydifferentlanguages147.SomeofthenotabletypesofAIusedwithinSARsandothercarerobotsare:NaturalLanguageProcessing,reasoninganddiagnosisalgorithms,imagerecognition,andmachinelearningtoadapttoinconsistentenvironments.148

Otherongoingapplicationsofhealthcarerobotshavebeenincarefacilitiesandhomesoftheelderly.Theserobotscanplaygames,distributepills,andconnectelderlyresidentstofriendsandfamilyviaascreenandcamera.Thiscategoryofrobotswouldbecalledcarerobots,astheyaresupplementaltoapatient’scaregivingroutine.Theserobotsareabletoadaptandlearntopatientbehaviourtoprovidethem a source of care and company when nurses and family are unable to.149 Additionally,exoskeletonshavealsoofferedarobotichandtothosewhoaredisabledorunabletowalk.Bysensingmicromovementsontheskin,someoftheexosuits,likeRobotSuitHALfromtheDaiwaHouseIndustry,can fluidly adapt to the patients desired movements to help them regain some mobility andautonomy.150Theseexoskeletonscanalsobeusedforrehabilitation—retrainingindividualstoregainmobilityafterstrokes.151

FutureApplications.Lookingtowardsthefuture,therearemanyexpansionsforteleroboticassistantsand surgical cobotsprojected.Oneof theseprojections is theability toundertakeandcreatenewprocedures,ratherthanenhanceandhelpwithexistingprocesses.Anotherisincreasingthedistance

143Anandan,TanyaM.,“RobotsandHealthcareSavingLivesTogether,”RoboticIndustriesAssociation,November23,2015.http://robotics.org/content-detail.cfm/Industrial-Robotics-Industry-Insights/Robots-and-Healthcare-Saving-Lives-Together/content_id/5819144Shademan,Azad,RyanS.Decker,andJustinD.Opfermann,etal.,“SupervisedAutonomousRoboticSoftTissueSurgery,ScienceTranslationalMedicine,Vol.8,No.337,May4,2016,pp.337-364.145MoreinformationaboutAethon’sTUGhere:http://aethon.com/tug/tughealthcare/146Szondy,David,“RobearRobotCareBearDesigntoServeJapan’sAgingPopulation,”NewAtlas:Robotics,February27,2015.http://newatlas.com/robear-riken/36219/147BBCTechnology,“PepperRobottoWorkinBelgianHospitals,”BBCNews:Technology,June14,2016.http://bbc.com/news/technology-36528253148Ashok,Asokan,“TheImpactofArtificialIntelligenceinHealthcare,”Medium,August24,2017.http://medium.com/@Unfoldlabs/the-impact-of-artificial-intelligence-in-healthcare-4bc657f129f5149Tarantola,Andrew,“RobotCaregiversareSavingtheElderlyfromLivesofLoneliness,”Engadget,August29,2017.http://engadget.com/2017/08/29/robot-caregivers-are-saving-the-elderly-from-lives-of-loneliness/150Bock,T.,T.Linner,andW.Ikeda,“ExoskeletonandHumanoidRoboticTechnologyinConstructionandBuiltEnvironment,”Intech,2012.http://cdn.intechopen.com/pdfs-wm/25773.pdf151Louie,DennisR.andJaniceJ.Eng,“PoweredRoboticExoskeletonsinPost-StrokeRehabilitationofGait:AScopingReview,”JournalofNeuroEngineeringandRehabilitation,June8,2016.http://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-016-0162-5


48

inwhichoperationscanbeperformed—teleroboticsurgeriesarecurrentlyperformedincloserangebutexpandingthescopeofthesetechnologiescouldallowhighlyspecializedsurgeonstooperateonindividualsaroundtheworld.Additionally,SARsandotherrobotassistantsareprojectedtocontinuegainingmomentumasthepushforfullydigitalhospitalstocombatcostandstafflimitationsincrease.In these digital hospitals, it is projected that robotswill take care of all delivery tasks, alongwithmonotonous or repetitive tasks like administering IVs or taking blood samples. Furthermore,integrating AI machine learning algorithms and diagnostics into surgical robots and teleroboticassistants will improve their ability to work alongside humans in major cases, but also moreindependentlyinsmalleroperations.152

Another potential area of growth in the future is that of rehabilitation and prosthesis. Roboticexoskeletonsuitsarecurrentlyunderdevelopmenttoenablethosewithlowerbodyparalysistowalksmoothlyandtohelpthosewhohavesufferedastrokerecoverfull functionality.Theseare indeedimprovementsofperformativityforexistingmodelsandprototypesofexoskeletonsinhealthcare.IftheseperformativeimprovementstoexoskeletonsarecombinedwithotheradvancementsinAIlikemachinelearningandmotionplanning,aswellastheintegrationofcompaniontechnologies(virtualreality,forexample),itseemsquitelikelyexosuitswillbemorenaturalandpreciseaswell.Moreover,researchinroboticprostheticsiscurrentlyworkingondevelopmentsusingneuralinterfacingtohelpgiveusersmorenaturaloperationandmotionoftheirprostheticlimbs.153,154

Anotherofthemostprominentdesiresforthefutureofhealthcarecompanionrobotsistheincreasingrangeofexpression,understanding,andconnectionwiththeirhumancounterparts.Especiallywithcarerobots,thereisanincreasingdemandforarobotcaretakerthatcantakecareofhousekeepingtasks,beintegratedwithinasmarthome,andrecognizepatientproblemstophonethepolice.Itseemslikely in the future care robots will be moving towards becoming live-in assistants to help withincreasingcostofhealthcareandqualifiedpersonnellimitations.155Additionally,theroboticassistantsofthefuturearelookingtohavemanyupgradesinAIthatwillhelpthemfillmanymorerolesthanwhattheycurrentlydoincludingconsultations,designingtreatmentplans,physicalassessment,andsentimentanalysis/facialrecognitionimprovements.

Companionship

Present applications. One of themost notorious applications of companion robots are sex robots.Further advanced beyond their blow-up doll predecessors, the sex robots of today can haveprogrammablepersonalities,lifelikeexpressions,andreactions.Mostnotably,sexrobotsaredesignedtofeelincreasinglymorelifelikewithsimulatedorgasms,automatedsexualpositions,andaphysicalbodythatfeelsrealtothetouch.Whilethefocusofthissubsetoftechnologyistoexpandtherangeofoptionsforsexualpartners,sexrobotsalsoprovideanoutletforpracticingfacetsofnichesexual

152FormorefutureprojectionsaboutAIinmedicalrobots,pleaseseethislinkatTheMedicalFuturist:http://medicalfuturist.com/the-technological-future-of-surgery/153Bogue,Robert,“RobotsinHealthcare,”IndustrialRobot:AnInternationalJournal,Vol.38,No.3.2011,pp.218-223.154Tolearnmoreaboutadvancedprosthetics,seeJohnsHopkinsAppliedPhysicsLaboratory’spage“Prosthetics”:http://jhuapl.edu/prosthetics/program/details.asp155Outing,Steve,“YourFutureCompanioninYourOldAgeCouldbeaRobot,”MarketWatch,October13,2017.http://marketwatch.com/story/your-future-companion-cool-socially-adeptand-a-robot-2017-10-13


49

desiressuchasrape,paedophilia,andsexwithcelebrities.156,157TheAIinvolvedinsexrobotscurrentlyis focused on machine learning and perception, with most of the dolls being controlled or givencommandsviaphoneapplication.158

Beyondtheuseofelderlycare,roboticpetsarebeingusedfortherapytotreatdepressionorotherwisecomfort individualsofallages.Onesuchexampleofthis isthePAROSeal,whoencouragespatientengagementandfocusesonrelaxingandmotivatingpatientsthroughhapticandaudiblefeedback.159Theserobotsareprimarilytheretoprovidereassuranceandmitigatefeelingsofanxietyorinsecurity.While not distributingmedicalmaterials or reminders, robots like PARO still fulfil a crucial role ofemotionalsupportduringapatient’srecovery.Furthermore,inapplicationsinwhichindividualsareunabletotakecareofanorganicpet,suchasincasesofdementiaoratanelderlycarefacility,roboticpets have been shown to decrease stress and anxiety in patients and reduce the need formedications160.As such, even roboticpetsprovide similar effects to a support animal in situationswhichtherapyanimalscannotbefeasiblyused.

Future Applications. Advancements in the field of animal-like robots may indeed lead to thereplacementoftherapydogsandguidedogsinfavourofroboticones.Thisalsohelpstooffsetthecostoflivesupportanimals,aswellasaccompanytheirusersintomoreplacesthansupportanimalscango.161Giventhis,thecurrenttrendsforthiscategoryseemtobepushingtechnologytobecomemorerelatable,helpful,andemotionallyandserviceably integratedwithin thehumancommunity.Whileroboticpetsmaystillbeusedastoysinthefuture,itseemsthatregardingcompanionship,therewillbeapush tomake roboticpetsmoreeasilyable to transitionbetweencomfortingandaiding. Sexrobots, particularly, and other companion robots, in general, will likely be looking at continuedimprovements towards becoming more “realistic” corporeally, with self-lubrication and increased“real-feel.” But it seems the future of companion robots will also be a bit more focused onimplementing improvements in the software—creating sex and social robots that utilize naturallanguageprocessing,increasedrangeofautonomousmotion,socialintelligence(affectivecomputing),andcreativityinresponsesandactions.WhilethesespecifictypesofAIimprovementsarenotdirectlystated,thecommongoalsofgeneratingamore‘natural,’‘sensual,’or‘enjoyable’experiencewitharobotseemtorelyuponimprovementsinthesecategories.

Industry

Presentapplications.Asoneoftheprojectedfastestgrowingsectorsofroboticsapplications,industrialrobotshaveearnedastableandsteadyplacewithinongoingandfutureroboticsdevelopments.162Theprimarygoalsofrobotsintheindustrialsectoraretoincreaseefficiencyandproductivitywhilesavingoncostsandreducingtoriskemployees.Furthermore,thereisanactivecomponentofperformativity

156Sharkey,Noel,AimeevanWynsberghe,ScottRobbins,andEleanorHancock,OurSexualFuturewithRobots:AFoundationforResponsibleRoboticsConsultationReport,2017.http://responsible-robotics-myxf6pn3xr.netdna-ssl.com/wp-content/uploads/2017/11/FRR-Consultation-Report-Our-Sexual-Future-with-robots-1-1.pdf157Viceinterviewandimpressionsofamalesexdollhere:http://video.vice.com/en_us/video/male-dolls/57f41d3556a0a80f54726060158FormoreinformationabouttheAIinsexrobots,pleaseseethislink:http://realbotix.com/Harmony159MoreinformationaboutPAROcanbeacquiredhere:http://parorobots.com/160Petersen,Sandra,SusanHouston,andHuanyingQin,etal.,“TheUtilizationofRoboticPetsinDementiaCare,”JournalofAlzheimersDisease,Vol.55,No.2,November19,2017,pp.596-574.PDF.161Jung,Scott,“RoboticDogsGuidetheBlind:TheFutureofFetching,”TechInnovationIntelIQ,August5,2014.http://iq.intel.com/the-future-of-fetching-robotic-dogs-guide-the-blind/162Glaser,April,“TheIndustrialRoboticsMarketWillNearlyTripleinLessthan10Years,”Recode,June22,2017.http://recode.net/2017/6/22/15763106/industrial-robotics-market-triple-ten-years-collaborative-robots


50

thataccompaniestheuseoftheserobots,astheycanproducemoreconsistentproductsandreducewaste. Given these goals, some of themost common robot applications are inmaterial handling,welding,finishing(sanding,sealing,milling,etc.),andassembly.163Onesuchexampleofsuchamachinecan be Trumpf’s TruMatic 600 Fiber: a robot that is able to intelligent place parts onmaterial forcutting, cut the parts, and process them without the intervention of a human operator.164Furthermore, intelligent assist devices (IADs) are being used in conjunction with fully-automatedindustrial robots to not only increase productivity and performance but also to reduce strain onworkers.Thesesemi-automaticrobotsandIADscandotaskssuchasliftingandmovingofheavytasksorheavymaterialmanipulationthatpresentadversehealthconsequencesonworkerswhoperformthese tasks repetitively.165One such exampleof these types of robots is the KUKARobotics liftingrobots.Withstrong,flexiblearms,theserobotscanhelphumanworkersliftandmoveheavymaterialseasily.TheKUKAL750TitanFcanliftandmoveupto750kgandhasareachof3600mm.Thisallowsamuchwiderrangeofproductmanipulationandplacementthancouldbesafelyorpossiblyachievebyhumanoperatorsalone.166

FutureApplications.Thefutureofindustrialrobotsseemstobesomewhatlinear.Thesefutureprojectsarestablyfocusedonincreasingefficiency,longevity,standardizationandloweringcost,errormargins,and user-friendliness.167 Another notable push for the future of industrial robotics, similarly toinfrastructuralandhealthcarerobotics,isthatofcollaborativerobots(cobots).Someofthesecobotsfall into thecategoryof IADs, lookingmore towardsautonomouslyorsemi-autonomouslyassistinghumanoperatorsintaskssuchaslifting,machinecutting,orplanning.OnesuchexamplestillbeingtestedandrefinedisAudiBrussels’“Walt”cobot.Waltcanrecognizegestures,beawareofworkersaroundittoincreasesafety,andcommunicatethroughadigitalfacewithitsoperators.168Thefutureforthesetypesofcobotsistoincreasetheirabilitiestosenseandcommunicatewiththeirhumanco-workers, to supplement the labour market, not replace human workers or present additional,unnecessaryrisks.169Otherareasofprojectedadvancementarethoseofroboticexoskeletonsuitsthathelp reduce strain and increase strength on operators. While some of the exosuits for industrialapplicationsdonotutilizerobotics,ithasbeennotedthatfuturemodelswillbedoingjustthat.170

Discovery

Presentapplications.Oneofthemostnotableusesofrobotsfordiscoveryisthatofspaceexploration.Theuseof robots for long-termordeep space travel is paramount for spaceagencies, as it is too

163PleaseseeRobotWorx’slistonIndustrialRobotIntegrationformoreinformationonthetypesofrobotsusedinindustrialapplications.AccessedMarch,2018:http://robots.com/applications164MoreinformationontheTruMatic600FibercanbefoundatTrumpf’swebsitehere:http://trumpf.com/en_INT/products/machines-systems/punch-laser-machines/trumatic-6000-fiber/165AmericanGrippersInc.,“RoboticsandAutomatedAssemblyRelieveRepetitiveStrainInjuryinWorkers,”RoboticIndustriesAssociation:RoboticsOnline,February12,2016.http://agi-automation.com/2016/02/robotics-and-automated-assembly-relieve-repetitive-strain-injury-in-workers/166FormoreinformationontheKUKAL750TitanFpleaseseethispage(accessedMarch,2018):http://robots.com/kuka/kr-1000-l750-titan-f167Formoreinformationonfuturegoalsofindustrialrobotics,pleaseseethislinkatRobotWorxentitled“IndustrialRobotsandtheFuture”:http://robots.com/blog/viewing/industrial-robots-and-the-future168MoreinformationaboutWaltcanbefoundhere:http://flanderstoday.eu/innovation/robot-human-interaction-breakthrough-audi-brussels169InternationalFederationofRobotics,TheImpactofRobotsonProductivity,EmploymentandJobs,April2017.http://ifr.org/img/office/IFR_The_Impact_of_Robots_on_Employment.pdf170Strickland,Eliza,“FordAssemblyLineWorkersTryOutExoskeletonTechtoBoostPerformance,”IEEESpectrum,November10,2017.http://spectrum.ieee.org/the-human-os/biomedical/bionics/ford-assembly-line-workers-try-out-exoskeleton-tech-to-boost-performance


51

dangerous and expensive to send humans to achieve the same tasks. Furthermore,when sendinghumansintospace,enoughequipmentandsuppliesneedtobeincludedfortheirsurvivalaswell—increasingtheweightandspacerequirementsofthesevessels.SomeofthemostnotableofspacerobotsareSpiritandOpportunity,bothroversonMars,usingtheAIdrivingsystem“Autonav”thatallowsthemtolearnandadjusttoobstaclesandterrains.TheCuriosityrover,alsoonMars,usesanAIprogram called “Autonomous Exploration for Gathering Increased Science” (AEGIS) that allowsCuriositytodecidewhichpicturestotakeandsamplestocollectforhumansbackonearth.171Anotherongoingapplicationofroboticsisthatofdeepseaexploration.Divingtodifficult-to-reachareas,suchas Mariana Trench, proves too dangerous for human explorers; however, robots can be built towithstandhigherpressuresandcompletedivesthehumanbodycannottolerate.Onesuchrobotofthiskind is theNereus, ahybridvehicle that combinesaspectsofboth remotelyoperatedvehicles(ROVs)andautonomousunderwatervehicles(AUVs).WhiletheNereusiscapableofautonomouslysurveying its area through sonar and camera, it also can be controlled by a human operatorremotely.172 OceanOne is also an example of a hybrid vehicle that dives to shipwrecks to collectmaterialsandbringthembacktothesurface.OceanOnealsoemploysAI-assistedcollisionavoidanceandnavigationsoftwaredesignedbystudentsintheStandfordRoboticsLab.173Outsideofexploration,anotheruseofrobotics fordiscovery is therobotsbeingusedtogatherdataandspotproblems inuninhabitableenvironments.One suchenvironment is thatof nucleardisaster sites. The robots atFukushima had been used for five years after the disaster to provide humanswith the invaluableinformationneededtosafelycontaintheradiationandlaterclean-uptheradiation.Assuch,therobotshave primarily been used to survey damage, radiation levels, and monitor humidity andtemperature.174

FutureApplications.Especiallyforthefutureofspaceexploration,thefurtherautomationofrobotsandwideruseofAIseemtobethemostlikelydirectionsofadvancementinthissector.Bymakingrobotsmoreautonomous,theycanconductmoreresearchandsendbackmoredatawithouthumanguidance.Furthermore,withmoreautonomous robots, theycanoperate inconditions thatdonotallowfordirecthumancontrolorguidance:openingupthepossibilityfortherobotstogaindatafromexoplanets, in oceanic trenches, or in other inhabitable areas. In contrast to industrial andmanufacturing fields, the future goals for discovery robots seem tobe generatingmore andmoreautonomousrobotsthatcangathermoredata,moreeffectively,andfurtherawayfromhuman’sdirectsphereofoperation.

Service

Presentapplications.Inthefoodserviceindustry,therearemanyapplicationsofbothrobotsandAIinresponsetoincreasingwagesandproductioncosts.TwooftheprimaryapplicationsofbothrobotsandAIareinKiosks,delivery,andpreparation.175KiosksutilizeAItoproviderecommendations,makemenuadjustmentsbasedonthetimeorweather,andevenprovideacustomer-friendlyexperience

171Ward,Tom,“NASA:AIwillLeadtheFutureofSpaceExploration,”Futurism,June27,2017.http://futurism.com/nasa-ai-will-lead-the-future-of-space-exploration/172Bowen,AndrewD.,DanaR.Yoerger,andChristTaylor,etal.,“TheNereusHybridUnderwaterRoboticVehicleforGlobalOceanScienceOperationsto11,000mDepth,”IEEEXplore,June30,2009.173Carey,Bjorn,“MaidenVoyageofStanford’sHumanoidRoboticDiverRecoversTreasuresfromKingLouisXIV’sWreckedFlagship,”StanfordNews,April27,2016.http://news.stanford.edu/2016/04/27/robotic-diver-recovers-treasures/174Hornyak,Timothy,“HowRobotsareBecomingCriticalPlayersinNuclearDisasterCleanup,”Science,March3,2016.sciencemag.org/news/2016/03/how-robots-are-becoming-critical-players-nuclear-disaster-cleanup175Sennaar,Kumba,“ExamplesofAIinRestaurantsandFoodServices,”TechEmergence,2018.http://techemergence.com/ai-in-restaurants-food-services/


52

throughchat-botstylecommunicationalgorithms.176Robotsarecurrentlyalsobeingemployedtodotediouscookingtasks,likeflippingburgers,withthehelpofsensorsandAI-decisionmaking.Onesuchexample of this is Flippy the burger chef employed at CaliBurgers Pasadena. Robotic applicationsrelatedtothefoodindustryisthatofdelivery.Autonomousdeliveryrobots(ADRs)areusedforbothfood delivery and delivery of other goods, delivery robots can use 3G technology paired withapplicationstosafelybringobjectstothecustomer’sdoorstep.177NotableusesofthistechnologyareDomino’spizzabotsandAmazon’sdeliverydrones.

Anotherareaofroboticsintheservicesectoristhatofdomesticorpersonalrobots.Thissubsectionofrobotstakescareofhouseholdchoresandassistwithcleaningtasks.Themostnotableofthesetaskscurrentlybeingperformedissweeping/mopping,lawnmowing,andcleaningguttersandswimmingpools.Theserobotsarecreatedtohelpsavehumanstimefromdoingmonotonouschoresroutinelytobothincreaseaestheticappealandcleanliness.Municipalservicerobotsarealsoratherpopular:takingovertasksthataretoorepetitive,dirty,dangerous,orinefficientforhumanoperators.178Oneoftheareasofpublicservicerobotstohighlightarerobotfirefighters:robotsthatcanperformsearchandrescue,fireassessment,andfiresuppressioninsituationsthatitistooriskyorimpossibleforhumanfirefighters.Allof therobotscurrently inusefor firefightingoperationsarecontrolledviaahumanoperator, despite automatically collecting data via cameras, sensors, and rangefinders to help thedriver.179

Theuseofroboticsforentertainmentisalsoasubcategorydeservingofnoteintheservicecategory.WaltDisneyWorldandDisneylandleadingthewaywithroboticsusedforcommercialentertainment:makinguseofroboticsforliveshows,amusementparkrides,andevenasrealisticadditionstoparkscenery. These types of robots provide a source of realism, interactivity, immersion, and elicitingemotional responses from those who see and interact with them. These types of robots areautonomous and have programmed “personalities”, in a similar technique used in other socialrobots.180Anotherareaofentertainment inrobotics is thatofpersonalentertainment: toys.Robottoysarerobotsdesignedforpersonalamusementandemployeeaprettydiverserangeoffeatures.Robottypescanbestaticormobile.Someofthesmarteronesemploygestureandfacialrecognitionandlearningalgorithms(so,notdirectlyemployingAIperse,butemployingsimilartechniques).Someoftheserobotsalsohaveapersonalityprogrammedandcanchangepersonalitybasedoninteractionswiththeirowners181,,likeSphero’sBB8orR2D2.182

FutureApplications.Theoverallfuturegoalsfortheserviceindustryseemtobequitesimilaracrossthe board: increasing efficiency, expanding ways of interacting and communicating with humans,improvingsensorsandcapabilities.Inthedomesticsphere,creatingrobotsthatcanachievesomeof176Walker,Jon,“FastFoodRobots,Kiosks,andAI:UseCasesfrom6RestaurantChainGiants,”TechEmergence,January27,2018.http://techemergence.com/fast-food-robots-kiosks-and-ai-use-cases/177Espinoza,Javier,“DeliveryRobotsHittheStreets,butSomeCitiesOptOut,”FinancialTimes:ArtificialIntelligenceandRobotics,January31,2018.http://ft.com/content/0a2a5a76-e0ea-11e7-a0d4-0944c5f49e46178OpenAccess:Government,“RiseoftheRobotsinthePublicSector,”AdjacentDigitalPoliticsLTD,December15,2017.http://openaccessgovernment.org/rise-robots-public-sector/40598/179Lattimer,BrianY.,“RoboticsinFirefighting,”SFPE:EmergingTrendsinEngineering,2015.http://sfpe.org/?page=fpe_et_issue_100180Panzarino,Matthew,“DisneyhasBegunPopulationitsParkswithAutonomous,Personality-DrivenRobots,”TechCrunch,February8,2018.http://techcrunch.com/2018/02/08/disney-has-begun-populating-its-parks-with-autonomous-personality-driven-robots/181Formoreexamplesoftoyrobots,pleasesee:http://techadvisor.co.uk/feature/digital-home/best-robots-2018-3652561/182Asasidenote,sexrobotsandpetrobotsmayverywellfallintotheentertainmentcategoryaswell.However,sincetheprimaryreasonbehindpurchasingsexorpetrobotsisbeyond“havingfun”or“play”,wewilllimitthediscussionheretorobotswhoseprimaryroleisentertainmentoramusement.


53

themorecomplicatedtasks,likewashingdishesordoinglaundry,seemstobethenextlikelydirection.Most of the difficulties of this come from being able to make judgments about differently sized,shaped,orcolouredobjectsmoreaccurately.Furthermore,inthefuture,therewillmorethanlikelybemore integration and cooperationbetween the varioushousehold robots themselves, as smarthomes are also projected to bemore popular in the future.183 Other improvements to bemade,especiallytomobilerobotsintheserviceindustryisthecontinuedimprovementofsensorsand“mapawareness”tobeabletointegrateabitmorefluidlywithhumans.Especiallywithdeliveryrobotsorrobotic chefs, being able to senseotherpeople andevenanimals accurately, andadjust basedonhumanaction,iscrucialtocarryingouttheirtaskssafelyandeffectively.

RobotsinthepublicservicesectorarealsoprojectedtobecomemorecomplexandintegrateAIforenhancedcommunicationandmachinelearning.Themainreasonfordoingsoistouserobotswithinmorehumanresourcepositions.Inthefuture,itseemsthatintelligentpersonalassistantswillbeinhighdemand,especiallyinpublicsectorssuchastaxmanagement,socialsecurity,orotherpositionsthathavestrongpaperworkandfilingcomponents.Assuch,futureroboticsapplicationswithintheservicesectorwillalsobetryingtoemploymoreAIdeeplearningtoincreasetherangetasksrobotscan manage.184 As such, potential future jobs for these robots are in libraries, courts, and trainstations.185Thefutureprojectionsforthetoyandentertainmentindustryseemtobetheonerifewithuncertainty. In some aspects, it seems that the “just a toy” industry is falling out of favour forcompanionandcarerobots:robotsthatcanbe“played”with,butalsoperformotherhelpfulorutility-drivenfunctions.186OneotherareaofgrowthforentertainmentcouldbetheadvancinguseofAIinvideogamestodevelopamorechallenging,robust,orunpredictableexperience.Roboticsinvirtualrealityapplications,perhapsevenforgaminginvirtualreality,mayalsobeanapplicationseeninthefuture.187

Environment

Presentapplications.Environmentalrobotsarerobotsthatfilltheneedsofprotectingtheenvironmentorassistingwithimprovingoreliminatinghumanactivitiesthatareharmfultotheenvironment.Onesuchexampleofthispracticeisroboticsinhorticulture.Thehorticulturalindustryisfacedwithhighlabourcosts,changingtreatmentvariables,delicateplants,andhighdemandformoreproductsgrownmoreefficiently.188SomeofthenotablethingsthathorticulturalrobotscandothroughtheuseofAIisplantandweedidentification,collisionavoidance,foreignsubstancedetection,andmappingtreeandplantlocations.Therearealsorobots189thatarecapableofsortinganddifferentiatingbetweenflowers

183Formoreonfutureapplicationsofdomesticrobots,pleaseseeSwanRobotics’discussiononDomesticRobots:http://swanrobotics.com/robot-applications/domestic_robot/184GuardianCities,“TheAutomatedCity:DoweStillNeedHumanstoRunPublicServices?”TheGuardian,2016.http://theguardian.com/cities/2016/sep/20/automated-city-robots-run-public-services-councils185Formoreinformationonfuture-projectionsofservicerobots,pleaseseeSanbot’spageon“PublicService”:http://en.sanbot.com/industrial/public-service186vanHooijdonk,Richard,“WillRobotsbeourKids’FutureToys,Teachers,andCompanions?”August3,2016.http://richardvanhooijdonk.com/en/will-robots-kids-future-toys-teachers-companions/187Lou,Harbing,“AIinVideoGames:TowardaMoreIntelligentGame,”HarvardUniversity,August28,2017.http://sitn.hms.harvard.edu/flash/2017/ai-video-games-toward-intelligent-game/188Pekkeriet,E.J.andE.J.vanHenten,E.J.,“CurrentDevelopmentsofHigh-TechRoboticandMechatronicSystemsinHorticultureandChallengesfortheFuture,”GreenhouseTechnologyGroup:WageningenUniversity,2011.http://edepot.wur.nl/169021189Hollick,Victoria,“LeadingtheWayinHorticulturalRobotics,”UniversityofSydney,October6,2016.http://sydney.edu.au/news-opinion/news/2016/10/06/horticultural-robotics-centre-will-revolutionise-aussie-farming.html


54

andpullingweeds.,190Otherimportantusesofroboticsinhorticultureisthatofpollination:robotsliketheQuadDustercandistributepollenmorequicklyandeffectivelythanhumancounterpartsandactascollaborativerobotswithinthisindustry.

Anotherapplicationofrobotsintheenvironmentisthatofwastemanagementandremoval.Similarto thedomestic robots, someof theenvironmental robotscandetectandremovewaste.AWasteShark,developedbyRanMarine,iscalledaseadrone,aclassofunmannedsurfacevehicles(USVs)thatisdesignedtohelpremovehumanwaste(plastic,strings,othersmalltrash)fromtheocean191.Whilefindingtheresourcestofindandemployhumanclean-upcrewsforthemenialtaskoftakingwrappersoutoftheoceanprovestobeabitchallenging,theserobotscanhelpresolvethisissuewithlowcostsandhighefficiency.Therearealsorobotsthathelptomonitoroilspills.ArobotcalledtheSeaGlidermonitorsthewaterconditions,toxicity,andsalinity,toinformhumanworkersaboutthenatureandscopeoftheoilspill.ThistypeofrobotisconsideredanAutonomousUnderwaterVehicle(AUV).Otherapplications of robotics in this sense also include: maintaining reef integrity, planting trees, andmonitoringglobalwarmingconditions(airquality,toxicity,etc.).192,193

FutureApplications.Theincreaseofautonomyandthecreationofautomatableenvironmentaltasksseemstobethetwomostprominentgoalsinthefutureofenvironmentalrobotics.Asmanyoftheproblemscurrentlyfacedenvironmentally–pollination,soilintegrity,andwastemanagement/control–haveeitherbeenprovidedbytheenvironmentorignoredbyhumanslongenoughtocauseproblems.Assuch,humanlabourforsuchtaskshasnotbeenneededordesireduntilthispoint,sothefutureofenvironmentalrobotsisnottotake-overorcollaborateonhumanjobs,buttoreplaceenvironmentjobsorcreatenewjobs.Oneofwhichprominentlyseemstobepollination—withthediscussionofpollinationbeing carriedoutbyautonomousdrones in the future.194 Especially inhorticulture, therobotsofthefuturewillalsoemploymoreAItohelprobotsdothemoredelicateorcomplicatedtasksof assessing and functioning within a complex environment, handling and buffering very delicateproducts,sortingproductsaccurately.

Agriculture

Presentapplications.Themajorityofthecurrentapplicationsofroboticsinagriculturalapplicationsarefocusedaroundtheoptimizationandautomationofcommercialagriculture.Thatis,focusingonresolving constraints on labour availability, cost of production, and time. Current applications ofroboticsareforweedcontrol,fertilization,harvesting,maintenance(seeding,mowing,pruning,etc.),herding,andmilking.195,196Forprivateuse,someoftherobotsavailable,at leastona limitedscale,includerobotsforlawnmaintenance(sprinklers)andlawnmowing(asdiscussedindomesticrobots),aswellasan“all-in-one”gardeningassistantrobotthatdoesplanning,planting,andmaintenanceto

190Formoreinformationaboutongoingusesofrobotsinhorticulture,pleasevisit:http://sciencelearn.org.nz/resources/2066-robots-for-horticulture191Atherton,KelseyD.,“Meet‘WasteShark’,theDroneThat’sPickingupGarbagefromtheOceans,”PopularScience,September12,2016.http://popsci.com/waste-shark-is-garbage-collecting-sea-drone192Formoreinformationoneco-robots,pleaseseeRobotsinServiceoftheEnvironment’spagehere:http://robotsise.com/todays-eco-robots/193Walker,Kris,“EnvironmentallyFriendlyRobots,”AZOCleantech,February15,2013.http://azocleantech.com/article.aspx?ArticleID=368194Moreinformationaboutbee-droneshere:http://npr.org/sections/thesalt/2017/03/03/517785082/rise-of-the-robot-bees-tiny-drones-turned-into-artificial-pollinators195Formoreinformationoncurrentagriculturalrobots,pleaseseeIntoRobotic’sarticleon“RobotsinAgriculture”.http://intorobotics.com/35-robots-in-agriculture/196Owen-Hill,Alex,“Top10RoboticApplicationsintheAgriculturalIndustry,”RobotIQ,August1,2017.http://blog.robotiq.com/top-10-robotic-applications-in-the-agricultural-industry


55

becontrolledbyanapp.Thissmall-scalerobotisapartofaDIYkitthatallowsuserstoadjustittothesizeoftheirpersonalgarden.197,198

Future Applications. Some of the future improvements to agricultural robots are geared towardsoptimization and speed. As such, there is a continued push to advance and implement more AItechniquessuchasmachine learningandpredictiveanalytics.199ByemployingmoreAI software inthesesystems,itbecomespossiblefortheagriculturalrobotsofthefuturetohandlemoredelicatecrops, inmorediverseenvironments,allwhile reducing theneed forhuman intervention.Anotherimprovementtotheserobotsthatwillbeseeninthefutureisalsoincreasedcollaborationfromroboticassistants.Datagatheredfromaerialdronescanbecombinedwithrobotsmonitoringthesoiltodirectautomatedseeding,maintenance,andharvestingrobotshowtotendtothecrops.Thisisprojectedtoincreasetheyield,reducewaste,andpreventexcessivestrainontheenvironmentthroughwastingwater or overusing pesticides and fertilizer.200 Another area that may see changes within animalprocessing,suchasroboticbutchersthatautomatethemeatmanagementprocessfromslaughtertopackaging.201

197Vaqar,Ali,“GardenSpaceisaNewPersonalGardeningRobotthatWatersyourPlantsandScaresAwayAnimals,”WonderfulEngineering,November,2017.http://wonderfulengineering.com/gardenspace-personal-gardening-robot-waters-plants-well-protects/198Kim,Gene,“ThisRobotWillGrowalltheFoodYouNeedinyourBackyard,”BusinessInsider,July21,2016.http://businessinsider.com/farming-robot-farmbot-automatically-grow-vegetables-backyard-garden-2016-7?international=true&r=US&IR=T199Sennaar,Kumba,“AIinAgriculture—PresentApplicationsandImpact,”TechEmergence,November16,2017.http://techemergence.com/ai-agriculture-present-applications-impact/200Thomas,Jim,“HowCorporateGiantsareAutomatingtheFarm,”NewInternationalist,November1,2017.http://newint.org/features/2017/11/01/agriculture-robots201Garfield,Leanna,“TheWorld’sBiggestMeatProducerisPlanningtoTestoutRobotButchers,”BusinessInsider,October26,2016.http://businessinsider.com/jbs-meatpacking-testing-robot-butchers-2016-10?international=true&r=US&IR=T


56

4. Social and economic impacts of AI and robotics Ashighlightedinthepreceding,AIandroboticsholdpromiseforaidinghumanbeingsinawiderangeofendeavours.ThecurrentandexpectedconsequencesofAIandroboticsneedadditionalclarificationas a step towards amore complete picture of the field. To this end, this section supplies a socio-economicimpactassessment(SEIA),includingenvironmentalimpacts,ofartificialintelligence(AI)androbotics.ThissectionendeavourstoorientateandinspireforthcomingworkinSIENNA,aswellasassistotherswhenstudyingthesocial,economicandenvironmentalimpacts.Wedefine SEIAas ananalysis used to identify andassess the social, economic andenvironmentalimpactsofAIandroboticsonsociety. Impactsrefertothepotentialchanges–whetherpositiveornegative,directorindirect,inwholeorinpart–causedbyorassociatedwiththetechnologicalfieldunder consideration. Following the European Commission’s Impact Assessment Guidelines,202examplesofsocialimpactsincludethoseaffectingtheprotectionofparticulargroups,equaltreatmentandopportunities,non-discrimination,individuals’privateandfamilylife,personaldata,governance.Economicimpactsincludethoseonemployment,theconductofbusiness,operatingcosts,smallandmedium enterprises, and financial burdens on businesses, workers, consumers and households.Environmentalimpactsincludethoseaffectingtheclimate,energyuse,airquality,waterqualityandresources, land use, environmental consequences of firms and consumers, waste production,generationandrecycling.ThecentralresearchquestionofthisSEIA is:Whataresomeofthemaincurrentandfuturesocial,economicandenvironmentalimpactsofAIandroboticsasreportedbyrecentliteratureandexpert(interviewees)inthetopic?Asdetailedbelow,weanswerthisquestionbyidentifyingandassessingcurrentandexpectedimpactsbasedoninformationgatheredfrompublications.Wealsoconsultedwithexperts inAIandrobotics inordertoanswerthisquestionand increaseourunderstandingofcurrentopinionsaboutimpactsassociatedwiththetechnologicalfield.Onthesebases,thisSEIAoffersconclusionsabouttheways inwhich itmaybepossibletomitigatenegative impactsandmaximisepositive impacts. This SEIA should not be interpreted as a comprehensive empirical examination(neither in itsaim,designorcontent)ofthesocial,economicandenvironmental impactsofAIandrobotics,butratheraqualitativestudythatcanhelptogroundfurtherresearch.Thefollowingsub-sectionscomprisethisSEIA.Sub-section4.1detailsthisSEIA’sapproach.Sub-section4.2identifiesthesocial,economicandenvironmentalimpactsofAIandrobotics.Sub-section4.3offersthe assessments of the impacts identified. Sub-section 4.4 sets out our conclusions regarding themitigationofnegativeimpacts(andmaximisationofpositiveones).4.1 Approach OurapproachtothisAIandroboticsSEIAinvolvedthefollowingkeysteps.

Figure1.StepsintheSIENNAAIandroboticsSEIA202Fordetailedinformationandprecisequestions,see:EuropeanCommission,ImpactAssessmentGuidelines,15January2009.http://ec.europa.eu/smart-regulation/impact/commission_guidelines/docs/iag_2009_en.pdf

PrepareSEIAplan

Identifysocial,economicandenvironmental

impacts

Consultstakeholders

Assessimpacts

Formulateconclusions

Reviewandpublishresults


57

Preparation ScopingSIENNA’sDescriptionofAction(DoA)delimitsthisSEIA’stopicandcoverage.Consequently,itbroadlycoverscurrentandexpectedsocial,economicandenvironmentalimpactsofAIandrobotics.Scopingoccurred during SIENNA’s proposal-writing phase. Our team relied on the DoA and internal teamdiscussionsinoutliningthisSEIA.DesignPost-proposalpreparationfirstlyinvolveddesigningourapproachforthisactivity,includingtheSEIA’sbasicoutline, specificmethodologiesandkeydefinitions.Preparationalso included thecreationofmaterialsenablingharmoniseddatarecordingandorganisationamongstdifferentresearchers.Impact identification SearchtermsandsearchesOurresearchinvolvedaliteraturereviewontheimpactsofAIandrobotics.Forourreview,wereliedonmethodsdescribedbyCreswell203.Weaimedtoidentifycontemporaryresearchandstudiesthatanalysed social, economic and environmental impacts of AI and robotics. We carried out theidentificationofimpactsfromDecember2017toJanuary2018.Given the scopeof thisSEIA (i.e., currentand future impactsofAIand robotics),our search stringqueries included impact (social, economic, environmental) AND technology (“AI and robotics”,“human-machine-interactions”),wheretheBooleanoperatorANDindicatesarequiredconjunctionofsuchterms.ThecompletelistofsearchstringsappearsbelowinTable3.Thepurposeandeffectofsuch search string queries was to limit the number of results returned to thosemost apparentlyrelevant.Wedidnotrelyonadditionalfilterssuchasdateorsubject,becausewefoundnoprincipledreasonstoexcludeolderstudiesfromanyfieldsolongastheypertainedtotheSEIA’stopic.

SearchtermexampleseconomicimpactAIandrobotics economicimplicationsAIandroboticseconomicimpacthuman-machineinteractions economicimplicationshuman-machineinteractionssocialimpactAIandrobotics socialimplicationsAIandroboticssocialimpacthuman-machineinteractions socialimplicationshuman-machineinteractionsenvironmentalimpactAIandrobotics environmentalimplicationsAIandroboticsenvironmentalimpacthuman-machineinteractions environmentalimplicationshuman-machine

interactionsimpactassessmentAIandrobotics impactassessmenthuman-machineinteractionsimpactofAIandrobotics impactofhuman-machineinteractionsAIandroboticshuman-machineinteractionsimpactclimate/biodiversity

human-machineinteractionsimpactclimate/biodiversity

negative impacts/implications/dangers/adverseconsequences/risksAIandrobotics

negativeimpacts/implications/dangers/adverseconsequences/riskshuman-machineinteractions

benefits/advantagesAIandrobotics benefits/advantageshuman-machineinteractionsTable3.SearchtermexamplesWe chose to search for and include peer-reviewed journals, industry reports, conferencepresentations,workingpapersandmediapublications.Weconducted searchesviaSSRN,Springer,ScienceDirect andGoogle. The choice of the search tools arose in response to the need formore203Creswell,JohnW.,ResearchDesign:Qualitative,Quantitative,andMixedMethodApproaches,4thEd.,Sage,LosAngeles,2014,Ch.2.


58

literatureonthetopicthaninitiallyfoundthroughacademicrepositories/searchenginesalone.Here,the choice of search tools tracks the choice to include non-peer-reviewed publications: a moreexpansiveaccountof impacts.Weneverthelessacknowledgethatmediaaccounts, forexample,donot pass the same rigorous evaluation process as, for example, peer-reviewed publications. Wescrutinised non-peer-reviewed literature to a greater extent (e.g., ensuring that the authorswererecognisedexperts,claimsmade,ifany,werebasedoncitedsources).SelectionBeforeselectinganypieceofliteratureforfullreviewandanalysis,weensureditspertinencetotheSEIA’stopicbyadoptinganinclusion/exclusioncriterionthat,inpractice,tooktheformofthefollowingquestion:(1)Doesthispieceofliteraturepertaintotherelevanttechnology’s(x)socio-economic[orsocialoreconomic]orenvironmentalimpact(y)?Thatis,variablexandyarenecessaryandsufficientonlyinconjunction.Weposedthisquestiononlyafterdownloadingandreadingeachresult’s(a)title,(b)abstract,ifany,(c)introductionand(d)executivesummary,ifany.Weexcludedthosepiecesofliteraturethatlackedpertinencebasedonourcriterion.Conversely,weincludedliteraturethatappearedtobepertinent.Again,weusedthiscriterionaftersearchqueriesusingsearchstringsdesignedtofilteroutirrelevantresults.Wecollectedandorganisedliteratureselectedforreviewandanalysisandthendividedthetaskofreadingpublicationsbetweenourteammembers.AnalysisandrecordingWeanalysedandrecordedeachpublication’sdescriptionofimpactsusingthefollowingprocess:(1)ateam member read a publication; (2) the team member recorded a publication’s citation in astandardisedmannerwiththesamecategoriesofinformationnotedbyrespectiveresearchers;(3)theteammemberrecordedapublication’sidentificationanddescriptionofcurrentand/orfuturesocial,economicand/orenvironmentimpactsofAIandrobotics.Althoughwedesignedtheidentificationanddatarecordingprocessduringthepreparationstep,weadjusted and updated aspects of it during its implementation. Such adjustmentswere isolated toterminologyusedduringcollectionandrecordingtopresentthefindingsmoreaccurately.Werefinedthescaleandterminologyofvariousheadingsonthebasisof the literatureandconsultationswithexperts.Forexample,the“likelihoodofimpactoccurring”columnrequiredanadditionaldatapoint–“currentlyoccurring”.Again,ourdesignaimwastoincludefindingsinaharmonisedandconsistentway,recognisingtheprocesstobeaniterativeonethatcouldchangebasedonliteraturereviewedandexpertsconsulted.Consultations Experts,designandaimsOur team supplemented the literature reviewwith expert consultations – i.e., 11 semi-structuredinterviews.We identified intervieweesusingcontact informationcollected for theSIENNATask1.4Stakeholderanalysisandcontactlist.Ourgoalattheoutsetwasforatleast10interviewswithvarioustypesofexpertsinvariousgeographiclocationsandacademicdisciplines.Weendeavouredtoconsultasdiverseagroupof intervieweesaspossible,butweacknowledgethat thenumberof interviewsconductedonlyallowsforanimpressionofwhatcertainexpertsthinkoftheimpactsofAIandrobotics.Our interviews do not provide a representative sample. Ultimately, we aimed to gain additionalsupplementalinsightaboutthesocio-economicandenvironmentalimpactsofAIandroboticsthroughtheviewsofindividualsactivelyengagedinresearchandconsiderationofsuchtopics.


59

We selected interviewees with knowledge and expertise in AI and robotics and their (potential)impacts.Thenumbersofexpertsandacademicdisciplineswereasfollows:1economist,2academicresearchers in AI (1 assistant professor in natural language processing and 1 doctor inneuroenhancement),2regulators(privacyanddataprotection),1scienceandtechnologyjournalist,1senior academic researcher/professor in robotics from a related project in robotics (autonomousvehicles’ systems), 2 legal experts, 1 representative from a European consumer-protectionorganisationand1academicresearcher(assistantprofessor)intheethicsofrisk.StructureanddetailsGiventhegeneralaimofourconsultations,weadoptedasemi-structuredinterviewmethodology204.Thismethodologyprovidesaguidedstructureforinterviewersandintervieweestofollow205.Italsoallowedfortheclarificationofquestions,aswellasfollow-upandextemporaneousdiscussion.Theopportunity for additional insights led our team away from the rigid structure of standardisedinterviews.Sincethecentral researchquestionandtopicof thisSEIAareclearlydefined,weoptedagainstanunstructuredapproachtotheinterviews.Before beginning the interviews, we drafted questions (see Annex 1) that drew from EC impactguidelines. The questions focused on impacts, affected parties, costs andmitigationmeasures. Toensure that the questions were answerable by all experts regardless of their respective fields ofexpertise,we kept thembrief and general. All interviews included the samequestions. Again, thepurposeoftheseinterviewswastosupplementourliteraturereview.We interviewed experts via Skype and/or phone. The average duration of each interviewwas 40minutes. In accordance with SIENNA ethics requirements, we provided interviewees with projectInformationSheetsandobtainedsignedconsentforms. Interviewresponsesweresummarised.Weaudio-recordedtheinterviewswiththeconsentofinterviewees.Neitherinterviewsummariesnoranyinterviewdatapresentedbelowincludepersonaldata.Wedeletedaudiorecordingsaftercompletingthe interview summaries and, upon request, gave interviewees an opportunity to review thesummaries.Assessment OverallaimBasedontheliteraturereview,interviewsandourinterpretation,thisportionoftheSEIAsummarisesandsynthesisestheappraisalsofimpactsandtheirvariousqualities.Theoverallaimistoelucidatetheidentifiedimpactsandprovideamorecompletepictureofthewaysinwhichtheyaresignificant.Inbrief,thissectionexplicatesmattersconnectedwiththewhen,who,whereandhowofimpacts,aswellasthechangestheywillbringandthewaystoreducenegativeimpacts.GeneralapproachtoassessmentsThedifferentpartsofsub-section4.3belowprovidespecificassessmentsanddetailsoftheirrespectiveapproaches;however,ourgeneral approach toassessment consistedof the following steps206.Wequalitatively assessed the significance of impacts identified in the literature and interviews(respectively).Forexample,wefirstcategorisedthefieldfromwhichimpactswillemanate(AI,roboticsorAI and robotics) aswell as thenatureof impacts (i.e.,whether the impactswill havepositive–beneficial–ornegative–adverse–effects).Wedescribedandinterpretedthequalitiesoftheimpacts204Lune,Howard,andBruceL.Berg,QualitativeResearchMethodsfortheSocialSciences,9thEdition,Essex,Pearson,2017,p.70.205Ibid,p.67.206EuropeanCommission,opcit.,2009,p.32-42.


60

andtheirsignificancealongadditionaldimensions(i.e.,theapplicationandsectorsfromwhichimpactswill arise, their timing, geographical reach, effects on values, parties affected, resilience andvulnerabilityofaffectedpartiesandcosts).Wethensynthesisedmeasuresproposedinpublicationsandinterviewsformitigatingnegativeimpacts.Wedidnotformalisetheremainingstepsofourapproach–formulateconclusions,reviewandpublish.Theconclusionsthatappearinsub-section4.4ofthisSEIArelyonanddrawfromoursynthesisofviewsculledfromtheliteratureandinterviews,respectively.ThefinalstepofreviewandpublicationofthisSEIAwilloccurafteritscompletion.Thenextsectionbelowpresentsourfindingsduringthefirststepofourapproach–identificationofimpacts.4.2 Social, economic and environmental impacts of AI and robotics:

identification BelowwepresenttheimpactsofAIandrobotics,groupedrespectivelybydomain–social,economicandenvironmental.Wealsodividetheimpactsintoseparatetablesaccordingtowhethertheycamefromourliteraturerevieworfromtheinterviews.Wepresentthoseimpactsintervieweesidentifiedoverthecourseofeachrespective interview(i.e., interviewees identified impactsduringtheentireinterview;thus,wedonot limit identificationtojustonequestion).Weremovedduplicateimpacts(e.g.,whenmultipleauthorsand/orintervieweesmentionedanimpactmorethanonce,weonlylistedit once in the relevant table).We also grouped similar impacts under higher order headings (e.g.,multipleauthorsandintervieweesmentionedAIandrobotics’deleteriousimpactonprivacy–whetherdataprivacy,personalprivacyetc.).Suchgroupingwasdonetoavoidtheproliferationoftermsforequivalent impacts. After the tables, we discuss some of their content; however, more extensiveconsideration and substantive discussion of the impacts appear in the next sub-section and itsconstituentparts.Social impacts Fromtheliteraturereview

SocialImpactsAlterhumaninteractions207Alterlegalandregulatoryframeworks208Altermoralconceptions209Alterunderstandingsofthescopeandlimitsofanaloguepersonhood210

207Byalter, theauthormeansboth improveanddiminish.Bossmann, Julia, “Top9ethical issues in artificialintelligence”, World Economic Forum, 21 October 2016; IEEE Global Initiative for Ethical Considerations inArtificial Intelligence and Autonomous Systems, Ethically Aligned Design: A Vision for Prioritizing HumanWellbeingwithArtificialIntelligenceandAutonomousSystems(AI/AS),version2,2018.208 Ezrachi, Ariel and Maurice E. Stucke, “Artificial Intelligence & Collusion: When Computers InhibitCompetition”,OxfordLegalStudiesResearchPaperNo.18/2015;UniversityofTennesseeLegalStudiesResearchPaperNo.267,2015.209Geraci,RobertM.,ApocalypticAI:VisionsofHeaven inRobotics,Artificial Intelligence,andVirtualReality,Oxford,OxfordUP,2010;Dignum,Virginia,“DesigningAIforhumanvalues”,ICTdiscoveries,SpecialissueNo.1,25Sept.2017.210Burgess,J.Peter,LucianoFloridi,AuréliePolsandJeroenvandenHoven,TowardsaDigitalFuture,EuropeanDataProtectionSupervisor,EthicsAdvisoryGroupreport,25January2018.


61

SocialImpactsDiminishindividuals’controlovertheirdata211Diminishprivacy212Improveeldercare213Improvehealthcare214Improvesafetyofpersonnelininhazardousenvironments215Increasebias216Increasediscrimination217Increaseharmandthreatofharmfromautonomousweapons218Increasejoblosses219Increaserulingclassdominationandwealth220Increasetransportationsafety221Promptunintendedconsequences222

Table4.SocialimpactsidentifiedintheliteratureFromtheinterviews

SocialImpactsDiminishindividuals’controlovertheirdataDiminishpluralisminmediaDiminishprivacyImprovecommunitiesthroughbettermatchingofpeoplewithsimilarinterestsImprovecybersecurityImprovedecision-makingthroughdataanalytics

211Kania,Elsa,“ThePolicyDimensionofLeading inAI”,Lawfare,19October2017;EuropeanDataProtectionSupervisor,"Artificial Intelligence,Robotics,PrivacyandDataProtection",Backgrounddocumentforthe38thInternational Conference of Data Protection and Privacy Commissioners, October 2016; Edwards, Lilian andMichaelVeale,“Enslavingthealgorithm:froma‘righttoanexplanation’toa‘righttobetterdecisions’?”draftpaper,October13,2017.212Altman,Micah,AlexandraWood,DavidR.O’BrienandUrsGasser,“PracticalApproachestoBigDataPrivacyOverTime”,BrusselsPrivacySymposium,draftpaper,6November2016;Pagallo,Ugo,“Robotsinthecloudwithprivacy:Anewthreattodataprotection?”ComputerLaw&SecurityReport,vol.29,no.5,2013;EuropeanDataProtectionSupervisor,opcit.,2016.213Frey,CarlB.,andMichaelA.Osborne,“TechnologyatWorkv2:TheFutureIsNotWhatItUsedtoBe”,CitiGPSReport,2016;Melkas,Helinä,LeaHennala,SatuPekkarinen,andVilleKyrki,"HumanimpactassessmentofrobotimplementationinFinnishelderlycare",InternationalConferenceonServiceology,2016.214PwCGlobal,WhyAIandroboticswilldefineNewHealth,June2017.215EuropeanParliament,EthicalAspectsofCyber-PhysicalSystem,June2016;Pagallo,opcit.,2013.216 Smith, Lauren, "Unfairness By Algorithm: Distilling the Harms of Automated Decision-Making", Future ofPrivacyForum,2017;Bossmann,opcit.,2016.217Smith,Lauren,opcit.,2017.218Gubrud,Mark,“WhyShouldWeBanAutonomousWeapons?”,IEEESpectrum,1June2016.219 Stone, Peter, Rodney Brooks, Erik Brynjolfsson, Calo, Ryan, Oren Etzioni, Greg Hager, Julia Hirschberg,ShivaramKalyanakrishnan,EceKamar,SaritKraus,KevinLeyton-Brown,DavidParkes,WilliamPress,AnnaLeeSaxenian,JulieShah,MilindTambe,AstroTeller,"Artificial IntelligenceandLife in2030,"StanfordUniversity,Stanford,CA,2016.220Sterniczky,Aaron,AdamOstolski,IoanaBanach-Sirbu,JaakkoStenhäll,JanPhilippAlbrecht,JaroslavFiala,LuisEsteban Rubio, Mohssin El Ghabri, Nóra Diószegi-Horváth, Pieter Pekelharing, Tiffany Blandin “GreenObservatory:RoboticsandArtificialIntelligence,”2017.221Hislop,Donald,CrispinCoombs,StanimiraTanevaandSarahBarnard,“Impactofartificialintelligence,roboticsandautomationtechnologiesonwork”,CIPDRapidevidencereview,2017.222Bossmann,opcit.,2016.


62

SocialImpactsImproveeldercareImprovehealthcareImprovelanguagetranslationImproveprotectionsforworkersincountrieswithweakersafeguardsImprovetransportationsafetyImprovevotingsecurityIncreaseavailabletimeIncreasebiasIncreasebigdataminingandanalysisIncreasecyberwarfareIncreasediscriminationIncreaseharmandthreatofharmfromautonomousweaponsIncreasejoblossesIncreasepoliceprofilingIncreaserulingclassdominationandwealthIncreasesurveillancePromptnewformsofco-operationandinclusionReducerepetitivetasks

Table5.SocialimpactsidentifiedintheinterviewsAmongotherthings,theabovetableshighlightstherangeofsocialimpactsofAIandrobotics.Thatis,AIandroboticsimpactmany,variousaspectsofthesociallifeofhumanbeings.Itbearsmentioningattheoutsetthatidentifyingimpactsinthistabularformdoesnotonitsownindicatewhetherspecificimpactsaffectallindividualsandsociety,orwhethertheydifferentiallyaffectcertainindividualsandpartsofsociety.Suchassessmentappearsbelow.However,wesuggestthatevenwithoutassessmentoftheimpacts,thetablesofferausefulsynopsisthatorientatesandintroducestheimpacts(andtheirrange)identifiedinourliteraturereviewandinterviews.Economic impacts Thetablesbelowpresenttheidentifiedeconomicimpacts.Fromtheliteraturereview

EconomicimpactsAccelerateeconomicinequalityaroundtheworld223Alterenterprises’organisationalstructures224Alterindustrialdesignprocesses225Improveworkplacerelationships226Increasecareerdisruptions(i.e.,multiplejobchanges)227

223FreyandOsborne,opcit.,2016.224Dirican,Cüneyt,“TheImpactsofRobotics,ArtificialIntelligenceonBusinessandEconomics”,Procedia-SocialandBehavioralSciences,Vol.195,3July2015.225 IEEEGlobal Initiative forEthicalConsiderations inArtificial IntelligenceandAutonomous Systems,op cit.,2018.226 Preimesberger, Chris, “Experts Predict How AI Will Impact Us in 2018,” eWeek, 2017.4December2017.227FreyandOsborne,opcit.,2016.


63

EconomicimpactsIncreasecompetitiveness228Increaseconsumerdemand229Increasediscrimination230Increaseefficiency231Increasejoblosses232Increaseprofitability233Reducecosts234Reducerisk235Revolutionisedigitalmarketing236

Table6.EconomicimpactsidentifiedintheliteraturereviewFrominterviews

EconomicimpactsAlterindustrialdesignprocessesConcentrateeconomicpowerEnhanceprivacyforconsumerservicesImproveaccesstoretailservicesImprovesupplychainsImproveworkplacerelationshipsIncreasecompetitivenessIncreaseconsumerdemandIncreasediscriminationIncreasefarmingproductivity(yields)IncreasejoblossesPrompttaxlossesReducecompetitivenessandmarketsharethroughover-relianceandinvestmentinsingletechnologylineRevolutionisedigitalmarketingStunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Table7.Economicimpactsidentifiedintheinterviews

228Arntz,M.,T.GregoryandU.Zierahn,“TheRiskofAutomationforJobs inOECDCountries:AComparativeAnalysis”,OECDSocial,EmploymentandMigrationWorkingPapers,No.189,OECDPublishing,Paris,2016.229Rao,AnandS.andGerardVerweij,SizingthePrize:What’stherealvalueofAIforyourbusinessandhowcanyoucapitalise?PwCAIanalysisreport,2017.230Smith,opcit.,2017.231Dirican,opcit.,2015.232Berriman,RichardandJohnHawksworth,“Willrobotsstealourjobs?ThepotentialimpactofautomationontheUKandothermajoreconomies”,PwCglobalreport,opcit.,2017;Arntzetal.,opcit.,2016;Dirican,opcit.,2015.;UNCTAD,BeyondAusterity:TowardsaNewGlobalDeal,TradeandDevelopmentReport,2017.;Hawking,Stephen, "This is themostdangerous time forourplanet",TheGuardian, 1Dec2016;Wisskirchen,Gerlind,BlandineThibaultBiacabe,UlrichBormann,AnnemarieMuntz,GundaNiehaus,GuillermoJiménezSoler,Beatricevon Brauchitsch, “Artificial Intelligence and Robotics and Their Impact on the Workplace,” The IBA GlobalEmploymentInstitute,2017;West,DarrellM.,“Whathappensifrobotstakethejobs?Theimpactofemergingtechnologies on employment and public policy”, Brookings Institution, 2015; Makridakis, Spyros, “TheforthcomingArtificialIntelligence(AI)revolution:Itsimpactonsocietyandfirms”,Review,Futuresvol.90,2017.233Arntzetal.,opcit.,2016.234FreyandOsbourne,opcit.,2016.235Dirican,opcit,2015.236 Orchestrate, “The impact of Artificial Intelligence & Machine Learning on the future of Marketing”, 1December2017.


64

Asthesection’sgeneralintroductionhighlighted,whichbearsrepeatingandconsiderationhere,thereweremultiplementionsofjoblosses.ThisimpactofAIandroboticsismuchdebated(e.g.,whethertheimpactofAIandroboticsintheworkplacewillresultinoveralljoblossesbysubstitutingforhumanlabourorwhetherAIandroboticswillcomplementhumanlabour,notleadingtooveralljoblosses)237.Althoughansweringquestionsaboutnetjoblosses(orgains)depends,atleastinpart,onthetrajectoryandimplementationofAIandrobotics,wefoundintheliteratureandinterviewsastrongindicationthatjoblosseswilloccur.Ifweacceptthisasfact,thenatleastsomeindividualswillloseouttoAIandroboticsbyalteringtheprospect,natureoravailabilityofwork.Insection4,wemorethoroughlydetailthepositiveandnegativeimpactsdescribedintheliteratureandinterviews;however,itbearsnotingnowthattherearebothbeneficialandadverseeconomicimpactsfromAIandrobotics.Environmental impacts Thetablesbelowpresenttheidentifiedenvironmentalimpacts.Fromtheliteraturereview

EnvironmentalimpactsIncreaseharmfromincreasedrobotuse(theircomponentmaterialsgenerallytoxicandnon-biodegradable)238Reducepesticideuse239Reducewasteduringproduction240Reducewaterandenergyfootprints241Reducenegativeeffectsofclimatechange242Reduceresource-informationgaps(esp.helpingtounderstandland-usepatternsanddecisionsupport)243Reducehumanpopulation244

Table8.EnvironmentalimpactsidentifiedintheliteraturereviewFromtheinterviews

EnvironmentalimpactsIncreaseenergyconsumptionIncreasepollutionthroughuseofcars

Table9.EnvironmentalimpactsidentifiedintheinterviewsFromtheliteraturereviewandinterviews,wefoundthatauthorsandintervieweeswerebullishaboutthebeneficialimpactsofAIandroboticsontheenvironment.Whetherthroughapplicationstoreducetheeffectsofclimatechangeorapplications todiminish theneed forpesticides, it seems that thequalityoftheenvironmentwillimprovewithAIandrobotics.Still,thisisafacileviewif,forexample,thenegativeimpactofincreasedenergyconsumptionunderminesoroutweighsthepositiveimpacts.

237Foranoverviewofthisdebate,cf.Autor,DavidH.,“WhyAreThereStillSoManyJobs?TheHistoryandFutureofWorkplaceAutomation”JournalofEconomicPerspectives,vol.29,no.3,2015.238Parrack,Dave,"Researchersaimtobuildeco-friendlyrobotsfrombiodegradablematerials",NewAtlas,2May2012.239EuropeanParliament,EthicalAspectsofCyber-PhysicalSystem,June2016.240Ibid.241Ibid.242Jibo,"RobotsforaSustainableFuture",14July2016.243Joppa,Lucas,"Howartificialintelligencecouldsavetheplanet",TheGuardian,30Jan2017.244PopulationMatters,"Theimpactofroboticsonfuturesocieties",briefingpaper,1May2016.


65

Limitations and challenges in impact identification AIandroboticsareprominenttopics,as(informally)evidencedbypolicy,academicandpopularmediaaccounts. However, the SEIA’s complex topic underscores a challenge our team faced: finding,pertinentliteraturerelatingspecificallytoimpacts.Asdescribedintheprecedingmethodologysection,wesupplementedpeer-reviewedacademicwithawider-rangeof sources, includingworkingdraftsandconferencepapersthatdirectlyaddressedtheSEIA’stopic.Tocounteractpotentialqualityissues,wescrutinisednon-peer-reviewedpublications,optingtoselectthosefocusedonimpactsandfromknownexperts.Giventhediversityofsourcesfromwhichweselectedliterature,wefound(sometimeswide)variancesbetweenthedepthofanalyses.Somepiecesofliteratureengageextensivelywithagiventopicandtechnology,othersapproachthetechnologyand/orimpactcasually.In addition to variations in depth of analysis of the different pieces of literature, the authors andorganisationsproducingsuchliteraturehadvariousinstitutionalmotivationsthatmayhavepromptedtheircreation.TheGreenObservatoryreportalsonotedthispointwhenitdescribedthe“downplayingoftheimpactsdependingonwhoauthoredthereports”.245The Green Observatory report also highlighted various authors’ “infatuation with all thingsrobotic/evangelism”.246 A 2016 report sponsored by Stanford University described “fantasticpredictions”247aboutthetechnologiesandtheirfutureimpactsthatsomeAIandroboticsliteraturemake.Thesefeatures–authors’infatuationwiththetechnologyandthepropensitytomakefantasticpredictions–representachallengetothedescriptionandevaluationof impactssincetheyrequiredemarcationoflegitimate,expectedimpactsandwriterhyperbole.Dignum identifiesanotherchallenge;namely, it isdifficult to findconsensuson the impactsof thistechnological field.248Weseethischallengeaffectingourwork ina limitedrespect.Consensuswasneverabarbywhichwejudgedtheidentificationorassessmentofimpacts.MorevitaltothisSEIAisanoverview, supportedby scholarly andpolicy researchand stakeholder-informedopinion, of themany,variousimpactsandtheirassessments.Dignum’spointdoes,however,indicateapossibletopicfor futureresearch.ResearcherscanexploreDignum’sperspectiveasaresearchquestion,suchas:Does expert consensus about social, economic and environmental impacts lead tomore accuratepredictionsaboutthem?As highlighted in this section’s introduction and consultation approach, the interviews provideopinions–learnedthoughtheymaybe–ofasmallnumberofindividuals.Moreinterviewswithotherexpertscouldaid indevelopingabroader,empirics-basedunderstandingofthe impacts.Again,weprovidetheseinterviewsassupplementstotheliterature.

245Sterniczky,etal.,opcit.,2017.246Ibid.247TheStanford2016reportstatesthat“ContrarytothemorefantasticpredictionsforAIinthepopularpress,theStudyPanelfoundnocauseforconcernthatAIisanimminentthreattohumankind.Nomachineswithself-sustaininglong-termgoalsandintenthavebeendeveloped,noraretheylikelytobedevelopedinthenearfuture.Instead, increasinglyuseful applicationsofAI,withpotentially profoundpositive impactsonour society andeconomyarelikelytoemergebetweennowand2030,theperiodthisreportconsiders.”Stone,etal.,opcit.,2016.248Dignum,opcit.,2017.


66

4.3 Assessment of impacts Inthissection,wesummariseandsynthesisetherespectiveviewsabouttheabove-describedimpactsofAIandrobotics.Wealsoincludeourinterpretations.Category of impacts – AI, robotics or AI and robotics Thefirstmatterconsideredwasthatofthetechnologicalfield–whatwecall“Category”–fromwhichthe identified impactsemanate.This is, inpart,amatterofterminology.Asnotedabove,SIENNA’sDescriptionofAction(DoA)delimitsthisSEIA’stopicandcoverage–impactsofAIandrobotics.Although this SEIA abides by agreed upon terminology and definitions of AI and robotics, not allauthorscombinedthetechnologicalfieldsinthesamewayastheSIENNAproject.Consequently,wenoteddescriptionsofimpacts(intheliterature)accordingtowhethertheycomefromAI,roboticsorAIandrobotics.Thegeneralconclusionsthatwederivefromtheliteratureisthat,byandlarge,authorsviewmostoftheimpactsascomingfromAIandrobotics,followedbythosecomingfromAIaloneandthenthosefromrobotics.Weaddheresomeofourinterpretationinordertohelpcontextualisethematterofcategorisationasit arose in the literature. There are clear distinctions between AI and robotics, aswell as overlapbetweenthefields:artificiallyintelligentrobots.However,wefoundthatauthorsdidnotalwaysdrawbrightlinedistinctionsormapoutcommonalitiesbetweenthetechnologicalfields.Forexample,RaoandVerweijconsidertheeconomicimpactsofAIandroboticscombined(i.e.,notAIandroboticsasartificiallyintelligentrobots).249FreyandOsborne,bycontrast,focusonAIandroboticsconceivedofasartificiallyintelligentrobots.250Thatis,theystayfaithfultoAIandroboticsasonlytheareaofoverlapbetween fields–artificially intelligent robots.Wedonotpresent thisasan indictmentagainstanyauthors.Rather,wepresentthistoacknowledgethecomplexityofthefieldsandhowsuchcomplexitycanaffectthescopeoftheirimpacts.AsRaoandVerweij’sworkhighlights,itispossible,topresenttheimpactsofAIandroboticscombinedwithoutconflatingthefields.Classifying the responses from interviewees into the same categories (i.e., AI and robotics, AI orrobotics),wefoundthattheyweregenerallyoftheviewthatAIandroboticswasthemainsourcefromwhichimpactsemanate.However,expertsworkinginAIorroboticsspecifically,asopposedtoexpertsinotherfieldswhoalsoexamineAIandrobotics,oftendrewsharpboundariesbetweenthefields.Forexample,intervieweeswithexpertiseinroboticsidentifiedareasofoverlapbetweenAIandroboticsbutclearlydelimitedimpactsissuingfromAIorrobotics.ThesameheldtruewithintervieweesworkinginAIspecifically.Recognisingthatinterviewees(asfoundintheliterature,too)sometimescollapseddistinctions between technological fields presents a question for future research or, at least,consideration: is itcommonopinionthatAIandroboticsareequivalenttechnologieswithoutmanydistinctionsbetweenthem?Wedonotendeavourtoanswertheabovequestionnow.Nevertheless,weviewthechallengesofcategorisationasapotentiallimitation–divergentterminologyorconceptionsoftechnologicalfieldscan stymie efforts to represent and classify their social, economic and environmental impacts inaccordancewithdistinctionsbetweentechnologies.

249RaoandVerweij,opcit.,2017.250FreyandOsborne,opcit.,2016.


67

Nature of the impacts The nature of an impact refers to whether an impact is or will be positive (beneficial), negative(adverse)orpositiveandnegativetoparties(e.g.,stakeholdersand/orcommunities).Theassessmentofthenatureofanimpactrepresentsan“allthingsconsidered”judgement,derived(respectively)frominterviewsand literature.Rather than lumpall impacts together in thetablesbelow,weseparatedimpacts by domain (i.e., social, economic and environmental) to facilitate easier analysis andcomparisonoftheimpacts.Again,weofferseparatetablesdependingontheirsource–theliteraturerevieworinterviews.Theinterviewtableincludesresponsestothefollowingquestion:“DoyouthinkthatAIandroboticswillhavebeneficialornegativeimpacts?”Apartfromthepresentationofthetables,wehighlightthegeneralvalueofassessingthenatureofimpacts.Tounderstandandprepareforimpacts,individualsandgroupsneedsomesenseofwhetherimpactswill deliver benefits, burdens or both benefits and burdens. The assessments in this sub-section offer a broad view of the nature of the impacts. Additional refinements appear below insubsequentsections.The nature of social impacts

Thefollowingtablesclassifybynaturethevariousidentifiedsocialimpacts.Fromtheliteraturereview

Natureofimpact SocialimpactsNegative Diminishindividuals’controlovertheirdata

DiminishprivacyIncreasediscriminationIncreaseharmandthreatofharmfromautonomousweaponsIncreaserulingclassdominationandwealth

Positive ImproveeldercareImprovehealthcareImprovesafetyofpersonnelininhazardousenvironmentsImprovetransportationsafety

Positiveandnegative AlterhumaninteractionsAlterlegalandregulatoryframeworksAltermoralconceptionsAlterunderstandingsofthescopeandlimitsofanaloguepersonhoodPromptunintendedconsequences

Table10.Thenatureofsocialimpactsasdescribedintheliterature(sourcescitedinSocialImpactstablefromsub-section4.2)Fromtheinterviews

Natureofimpact SocialimpactsNegative Diminishpluralisminmedia

DiminishprivacyIncreasecyberwarfareIncreasediscriminationIncreaseharmandthreatofharmfromautonomousweaponsIncreaserulingclassdominationandwealth


68

Natureofimpact SocialimpactsPositive Improvecybersecurity

ImprovecommunitiesthroughbettermatchingofpeoplewithsimilarinterestsImprovedecision-makingthroughdataanalyticsImproveeldercareImprovehealthcareImprovelanguagetranslationImprovevotingsecurityPromptnewformsofco-operationandinclusionReducerepetitivetasks

Positiveandnegative IncreasebigdataminingandanalysisIncreaseavailabletimeIncreasepoliceprofilingIncreasesurveillanceIncreaseavailabletime

Table11.ThenatureofsocialimpactsasdescribedintheinterviewsThe nature of economic impacts

Thefollowingtablesclassifybynaturethevariousidentifiedeconomicimpacts.Fromtheliteraturereview

Natureofimpact EconomicimpactsNegative Accelerateeconomicinequalityaroundtheworld

Increasecareerdisruptions(i.e.,multiplejobchanges)IncreasediscriminationIncreasejoblossesRevolutionisedigitalmarketing

Positive ImproveworkplacerelationshipsIncreasecompetitivenessIncreaseconsumerdemandIncreaseefficiencyIncreaseprofitability

Positiveandnegative Alterenterprises’organisationalstructuresAlterproductdesignprocesses

Table12.Thenatureofeconomicimpactsasdescribedintheliterature(sourcescitedinEnvironmentalImpactstablefromsub-section4.2)Fromtheinterviews

Natureofimpact EconomicimpactsNegative Accelerateeconomicinequalityaroundtheworld

ConcentrateeconomicpowerIncreasediscriminationIncreasejoblossesPrompttaxlosses


69

Natureofimpact EconomicimpactsReducecompetitivenessandmarketsharethroughover-relianceandinvestmentinsingletechnologylineStunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Positive EnhanceprivacyforconsumerservicesImproveaccesstoretailservicesImprovesupplychainsImproveworkplacerelationshipsIncreasecompetitivenessIncreaseconsumerdemandIncreasefarmingproductivity(yields)

Positiveandnegative Alterenterprises’organisationalstructuresAlterproductdesignprocesses

Table13.ThenatureofeconomicimpactsasdescribedintheinterviewsThe nature of environmental impacts Thefollowingtablesclassifybynaturethevariousidentifiedenvironmentalimpacts.Fromtheliteraturereview

Natureofimpact

Environmentalimpacts

Negative

Increaseharmfromincreasedrobotuse(theircomponentmaterialsgenerallytoxicandnon-biodegradable)

Positive

Reducepesticideuse

ReducewasteduringproductionReducewaterandenergyfootprintsReducenegativeeffectsofclimatechangeReduceresource-informationgaps(esp.helpingtounderstandland-usepatternsanddecisionsupport)ReducehumanpopulationReducepesticideuse

Table 14. The nature of environmental impacts as described in the literature (sources cited inEnvironmentalImpactstablefromsub-section4.2)Fromtheinterviews

Natureofimpact EnvironmentalimpactsNegative Increaseenergyconsumption

Table15.ThenatureofenvironmentalimpactsasdescribedintheinterviewsBased on the categorisation of impacts in each domain and separated by source – literature orinterviews–therearegroundsforguardedpessimismaboutAIandrobotics:amajorityoftheimpactsofAIandroboticsarenegative.Amoreoptimisticapproachtotheassessmentsinthissub-sectionmaybetoanswerthefollowing:Which,where,whenandbywhatmeanscanincreasesinpositiveimpactsand reductions of negative ones occur? Althoughwe do not settle thismatter, the following sub-sectionsoffersomeinsightsintootherimpactsthatmayassistindevelopingasolution.Applications and sectors of AI and robotics with the most positive (beneficial) or negative (adverse) impact (described by interviewees)


70

Wenoticedthatourliteraturereviewdidnotclarifyasignificantissue;so,weusedtheinterviewsasanopportunitytohelpgetasenseofopinionsaboutit.Theissue:AIandroboticsconsistofavarietyofdistinctapplications,yetwelackedaclearsenseaboutwhichofthesemightyieldthemostpositive(i.e.,beneficial)ornegative(i.e.,adverse)impacts.Althoughthisissueissignificantbecauseaddressingithelpstocompletethepictureofthetechnologicalfield,itcanalsoaidstakeholders,suchaspolicy-makers and regulators,when crafting policy and regulations. For example, if a specific applicationprovessoodiousthatitconsistentlyproducesnegativeimpacts,policy-makerswillneedinformationaboutthatapplicationanditsimpactsinordertomakeinformedchoices.To address the above issue,wedecided to gain some insightby asking interviewees the followingquestion: “Which applications of AI and robotics do you thinkmight have themost beneficial oradverseimpactsonindividualsorsociety?”Accordingtointerviewees,theapplicationswiththemostadverse impact on individuals and society were autonomous weapon systems. Conversely,intervieweesnotedmedicaldiagnosticapplicationsasthemostbeneficialforindividualsandsociety.According to interviewees,other applicationswithpositive impacts includedautonomous cars andassistivemedicalapplications.We soughtadditional clarification from intervieweesaboutotheraspectsof themostpositiveandnegativeimpacts.Tothisend,weaskedthefollowingquestion:“AretherecertainsectorsorfieldsinwhichAIandroboticsmighthavethemostbeneficialoradverseimpacts?”Bysector,wemeanapartof an economy (e.g., service sector), and by field (as consistent throughout this report) wemeansubjectorareaofactivity(e.g.,scientificfield).Basedontheresponsestotheprecedingquestionaboutapplications,whichhighlightedthenegativeimpact of autonomous weapons’ applications and the positive impact of medical diagnosticapplications,ourexpectationsweretoreceivesimilaropinionsaboutthesectorsandfieldswiththemost positive and negative impacts. By and large, the interviewees’ responses seemed consistentacross the two questions (i.e., themost negative impact from autonomousweapons’ applicationsappears to coincide with the most negative impact from the military/defence sector). However,intervieweesdidnotalwaysexplainthereasons for theiropinions.Consequently,weavoidmakinginferencesabouttherelationshipbetweeninterviewees’responsestotherespectivequestionsAccordingtotheinterviewees,themilitary/defencesectoristheoneinwhichAIandroboticswillhavethemostnegativeimpact.Conversely,thehealthcaresectoristheoneinwhichAIandroboticswillhavethemostpositiveimpact.IntervieweesalsoexpressedtheviewthatAIandroboticswillgreatlybenefittheautomotiveandeducationalsectors.Timing of impacts and starting points (described by interviewees) Preparingforandpotentiallymitigatingnegativeimpactsseemstodependonbeingabletoanticipatethetimingoftheiroccurrence.So,too,doesrecognisingandpreparingforthepositiveimpactsofAIandrobotics.Asourcentralresearchquestionunderscores,thisSEIAalsoendeavourstoaccountforthetimingofsocial,economicandenvironmentalimpacts.Wefoundthatnotallliteraturediscussedthetimingofimpacts.Forexample,Pagallodescribedthe(social)impactthatroboticswouldhaveonprivacybutdidnotspecifywhentheimpactwouldoccur.251Bycontrast,SmithindicatedthatAI’sdiscriminatory(social)impactforcertainrecipientsofinsurance251Pagallo,opcit.,2013.


71

and social benefitswould happen in the future252. Still, Smith’swork framed the future impact inconditionalterms;namely,ifnoonetakesactiontostymiesuchdiscrimination,thenitwilloccur.OtherauthorsstatedthattheimpactsofAIandroboticsarecurrentlyoccurringandwillcontinuetodosointo the future – Makridakis places job losses along this current—future continuum.253 Somepublicationsmarkedimpactsascurrentlyhappening,asPwCdidwhenidentifyingthesocialimpactofAIandroboticstoincreaseproactivemanagementofhealthylifestyles254.WesummarisetheviewsfromtheliteratureasgenerallyleaningtowardstheimpactsofAIandroboticsoccurringnowandintothefuture.Next,authorsgenerallyviewedthefutureasthestartingpointofAIandrobotics’impacts.Wewantedtogetopinionsaboutmoredefinitetimeframesandstartingpointsofimpacts(evenwiththegeneralprognoses fromthe literaturedescribedabove).Toobtain such information,weaskedintervieweesthefollowingquestions:(1)“Whichimpactscanyouforeseeinthenext5-10years?”and(2)“Whichimpactscanyouforeseeinthenext20years?”Thetablesbelowpresenttheinterviewees’responses.Wedividedtheiranswersaccordingtostartingpoint/timeframeofimpacts,aswellasthenatureoftheimpacts(negative,positiveandpositiveandnegative).

5to10years

Negativeimpacts• increasejoblosses• increasepoliceprofiling

Positiveimpacts• increased industrial use of AI and robotics (e.g., in semi-conductors and

aircraft)• substantialbattlefielduseofautonomousweaponssystems• increaseduseofautonomousdrones• limiteduseofAIinvoting• morewidespreaduseofAIretailservices• increaseduseofautonomousvehicles• widespreadinstantspokenlanguagetranslation• widespreaduseofAIandroboticpersonalassistants• increaseduseofAIandroboticsinindustrialdesignprocess

Positiveandnegativeimpacts• widespreaduseofdatamining/databases• widespreaduseofAIinfinance

Table16.Responsesto“Whichimpactscanyouforeseeinthenext5-10years?”

20years

Negativeimpacts• widespreadbattlefielduseofrobotsandautonomousweaponssystems• substantialjoblossesduetoautomation

Positiveimpacts• highlyoptimisedfarmingsolutionsusingAIandrobotics• widespreaduseofdomesticrobotsandroboticsystemsinhomes

Positiveandnegativeimpacts• Complementarity255

252Smith,opcit.,2017.253Makridakis,opcit.,2017.254PwCGlobal,opcit.,2017.255Complementarityisaneffectbywhichproductivity,earnings,anddemandforhumanlabourincreasesbecauseofrobotsintheworkplace.Autor,DavidH.,“WhyAreThereStillSoManyJobs?TheHistoryandFutureofWorkplaceAutomation”JournalofEconomicPerspectives,vol.29,no.3,2015,p.5.


72

• limitedautonomousindustrialmanufacturing• radicalchangestoallinteractions• widespreadadoptionofautonomousvehicles

Table17.Responsesto“Whichimpactscanyouforeseeinthenext20years?”Althoughthespecificpredictions(especially)fromtheabovetablesofferuncertaininsightaboutthefuture, the interviewees’opinionsat leastunderscoreconfidence inthecontinuedexistence, ifnotprogressofAIandrobotics(andtheimpactstheywillhaveonsociety).Alltold,theinformationinthissub-sectionhighlightsandaffirmsanintuitionandclaimaboutAIandrobotics: the development, implementation and consequent impacts are happening now and willcontinuetodosointothefuture256.Asthemorecompletepictureofimpactsunfoldsinthesubsequentassessments,andespeciallywithrespecttothesuggestedmitigationmeasures,thecontemporaneoustemporaldimensionofimpactsshouldberemembered.Iftheopinionsfromintervieweesaboutthetiming of impacts is accurate, thenwork nowmust continue in order to ensure that the positiveimpactsincreaseandthedeleteriousonesdiminish.Geographical reach of the impacts Togainaperspectiveofthegeographicalreachofimpacts–i.e.,wheretheimpactsoccurorwilloccur–weroughlygroupedsemanticallyequivalent locationtermsused inthe literatureundercommonheadings.Forexample,wefoundmultipletermsintheliteraturedescribingtheoverarchingandwide-reachingspreadofimpacts(e.g.,international,global,worldwideetc.).Wegroupedtheseundertheheading‘global’,whichwedefineas“distributedamongstallcountries”.Wechoseagainstkeepingallthe termswe found in order to limit the proliferation of descriptors that did not addmeaningfulrefinementornuance.Basedontheimpactsidentifiedintheliteraturereview,themajorityofimpactsoccur(orwilloccur)globally.Thenextmostcommonheading,whichadmittedlyisnotaplace,was“Notdescribed”.Thisheadingreferstothoseauthors(andtheirpublications),forexampleDirican257,thatdidnotindicatewhereanimpactoccursorwilloccur.Onoccasion,publicationsmadefine-graineddistinctionsabouttheimpacts’geographicalreach.TheUnitedStatesandChinareceivedspecialattention.AuthorssuchasKania258andEzrachietal.259viewthesetwocountriesasmajorforcesbehindAIandroboticsandthus the sites of many of their impacts. It should be noted, however, that some fine-graineddistinctionsaboutthegeographyofimpactsreflectthesubjectofauthors’investigations,ratherthanexhaustiveanalysisaboutthefullgeographicalextentofimpacts.Forexample,Melkasetal.examinetheimplementationandimpactofroboticsoneldercareinFinland260.Other literature-baseddescriptionsofthegeographicalreachof impacts inthe literature,however,indicatelessprecisioninidentifyingexactlywhereorwhichcountrieswillfeelimpacts.Forexample,someauthorsacknowledge theglobal reachof impactsbutnevertheless seeunequalgeographicaldistributionsanddonotspecifypreciselocations(e.g.,“Global–variouscountriesdifferentially”—meaningtheeffectswilldistributeunevenlythroughtheworld).

256Brynjolfsson,Erik,AndrewMcAfee,JeffCummings,TheSecondMachineAge:Work,Progress,andProsperityinaTimeofBrilliantTechnologies,WWNorton,NewYork,2014,chs.2-4.257Dirican,opcit.,2015.258Kania,opcit.,2017.259EzrachiandStucke,opcit.,2015.260Melkas,etal.,opcit.,2016.


73

Weaskedintervieweesthefollowingquestionaboutthe“geographyofimpacts”:“Howfarandwide,geographically,couldindividualsandcommunitiesfeeltheimpacts?”Interviewees’responsesshowedthattheybelieve,byandlarge,thattheimpactsofAIandroboticswillbeglobalinreach.However,intervieweesalsoidentifiedspecificcountriesthatwillbegreatlyimpacted.IntervieweesnotedthattheUnitedStatesandChinastandtogainfromAIandrobotics,asdosmallercountriessuchasEstonia,whoseeconomycannimblyadaptandintegratetheAIandroboticsandtheirimpacts.Intervieweesfocussed attention on the negative impacts that AI and robotics will bring to poorer regions andcountriesoftheworld.Inthisregard,theconcentrationofbeneficialimpactswasseenasstayingwithwealthynations,whileadverseimpactsreachthepoorer,less-developedregionsandcountries.Althoughthemajorityviewabouttheglobalreachofimpactsroughlyfitswithourexpectations,the“Global–variouscountriesdifferentially”headingseemsthemostintuitivelyplausible.Inthissense,evenifthemajorityofimpactshaveaworldwidespread,theirrespectiveextentswould,ingeneral,distributedifferentiallyacrosscountries.Our intuition,however, seemsmisplacedaccording to thenumbersfromouranalysis.Effects of impacts on values ThissectiondetailswhetherandwhichimpactsofAIandroboticssupport(orenhance)ordiminish(orundermine)communityand/orsocietyvalues.Furthermore,weindicatethevaluesimpactedbyAIandrobotics thatauthorsor intervieweessawasbeingenhancedordiminishedbyspecific impacts.Toascertaininterviewees’viewsabouttheimpacts’effectsonvalues,weaskedthefollowingquestion:“Willtheimpactssupportorunderminetheaffectedcommunities’orsocietalvalues?”UnderstandingtheeffectsonvaluesthatAIandroboticswillhaveisimportantbothtohelpevaluatetheuseofAIandroboticsapplicationsandtoinformtheirresponsibledesignanddevelopment.Thisalsocouldhelpprovidesomeinsightintothetrade-offsthatmightemergewhenvaluescompetewitheachother(e.g.,autonomyandsafety)261orwhendecisionsneedtobemadeaboutprioritisingfundingorimplementingregulationtoprotectthevaluesthatneedtobesafeguarded.Social impacts – effects on values ThetablesbelowshowthekeyvaluesthatwillbeaffectedbythesocialimpactsofAIandrobotics.Fromtheliteraturereview

Affectedvalues ExamplesofrelatedidentifiedimpactsAccountability AlterlegalandregulatoryframeworksAutonomy Diminishindividuals’controlovertheirdata

IncreasejoblossesDemocracy Increasebias

IncreasediscriminationIncreaserulingclassdominationandwealth

Equality

IncreasebiasIncreasediscriminationIncreaserulingclassdominationandwealth

Fairness

IncreasebiasIncreasediscriminationIncreaserulingclassdominationandwealth

261AfulldiscussionisnotwithinthescopeofthislimitedexercisebutmightfallwithintheethicalimpactanalysisduringthelaterstagesoftheSIENNAproject.


74

Affectedvalues ExamplesofrelatedidentifiedimpactsHumandignity Increasebias

IncreasediscriminationIncreaseharmandthreatofharmfromautonomousweaponsImproveeldercareImprovehealthcare

Justice

AlterlegalandregulatoryframeworksIncreasebiasIncreasediscriminationIncreaserulingclassdominationandwealth

Physicalintegrity(safety)

ImproveeldercareImprovehealthcareImprovesafetyofpersonnelininhazardousenvironmentsImprovetransportationsafetyIncreaseharmandthreatofharmfromautonomousweaponsIncreasejoblosses

Integrityofproperty(security)

IncreaseharmandthreatofharmfromautonomousweaponsIncreasejoblosses

Privacy Diminishindividuals’controlovertheirdataDiminishprivacy

Socialrelations,solidarityandcohesion

AlterhumaninteractionsAltermoralconceptionsAlterunderstandingsofthescopeandlimitsofanaloguepersonhoodIncreasejoblossesIncreaserulingclassdominationandwealth

Table18.Social impacts–effectsonvalues–asappearing inthe literature(sourcescited inSocialImpactstablefromsub-section4.2)Fromtheinterviews

Affectedvalues ExamplesofrelatedidentifiedimpactsAutonomy IncreasejoblossesDemocracy Improvevotingsecurity

IncreasebiasIncreasediscriminationIncreaserulingclassdominationandwealthPromptnewformsofco-operationandinclusion

Equality

ImprovevotingsecurityIncreasebiasIncreasediscriminationIncreaserulingclassdominationandwealthPromptnewformsofco-operationandinclusion

Fairness

ImprovevotingsecurityImproveprotectionsforworkersincountrieswithweakersafeguardsIncreasebiasIncreasediscriminationIncreasepoliceprofilingIncreaserulingclassdominationandwealthPromptnewformsofco-operationandinclusion

Justice

ImprovevotingsecurityImproveprotectionsforworkersincountrieswithweakersafeguardsIncreasebiasIncreasediscriminationIncreasepoliceprofilingIncreaserulingclassdominationandwealthPromptnewformsofco-operationandinclusion

Physicalintegrity(safety) Improveeldercare


75

Affectedvalues Examplesofrelatedidentifiedimpacts Improvehealthcare

IncreasecyberwarfareIncreaseharmandthreatofharmfromautonomousweaponsIncreasejoblossesIncreasepoliceprofiling

Integrityofproperty(security)

ImprovecybersecurityIncreasecyberwarfareIncreaseharmandthreatofharmfromautonomousweaponsIncreasejoblossesIncreasepoliceprofiling

Pluralism DiminishpluralisminmediaPrivacy Increasebigdataminingandanalysis

IncreasesurveillanceDiminishprivacy

Socialrelations,solidarityandcohesion

ImprovecommunitiesthroughbettermatchingofpeoplewithsimilarinterestsIncreaseavailabletimeImprovelanguagetranslationIncreasejoblosses

Well-being

Improvedecision-makingthroughdataanalyticsReducerepetitivetasks

Table19.Socialimpacts–effectsonvalues–asappearingintheinterviewsEconomic impacts – effects on values Thetablesbelowshowthekeyvaluesthatwillbeaffectedbytheeconomicimpacts.Fromtheliteraturereview

Affectedvalues ExamplesofrelatedidentifiedimpactsCompetitiveness Accelerateeconomicinequalityaroundtheworld

Alterenterprises’organisationalstructuresAlterindustrialdesignprocessesImproveworkplacerelationshipsIncreasecompetitivenessIncreaseefficiencyIncreaseprofitabilityReducecostsReducerisk

Efficiency AccelerateeconomicinequalityaroundtheworldAlterenterprises’organisationalstructuresAlterindustrialdesignprocessesImproveworkplacerelationshipsIncreasecompetitivenessIncreaseefficiencyIncreaseprofitabilityReducecostsReducerisk

Equality AccelerateeconomicinequalityaroundtheworldIncreasejoblosses

Democracy AccelerateeconomicinequalityaroundtheworldJustice Accelerateeconomicinequalityaroundtheworld

IncreasejoblossesFairness Accelerateeconomicinequalityaroundtheworld

IncreasejoblossesProductivity Accelerateeconomicinequalityaroundtheworld


76

Affectedvalues ExamplesofrelatedidentifiedimpactsAlterenterprises’organisationalstructuresAlterindustrialdesignprocessesImproveworkplacerelationshipIncreasecompetitivenessIncreaseconsumerdemandIncreaseefficiencyIncreaseprofitabilityRevolutionisedigitalmarketing

Profitability AccelerateeconomicinequalityaroundtheworldAlterenterprises’organisationalstructuresAlterindustrialdesignprocessesImproveworkplacerelationshipsIncreasecompetitivenessIncreaseconsumerdemandIncreaseprofitabilityReducecostsRevolutionisedigitalmarketing

Increasecareerdisruptions(i.e.,multiplejobchanges)IncreasediscriminationIncreasejoblosses

Table 20. Economic impacts – effects on values – as appearing in the literature (sources cited inEconomicImpactstablefromsub-section4.2)Fromtheinterviews

Affectedvalues ExamplesofrelatedidentifiedimpactsAccess ImproveaccesstoretailservicesCompetitiveness Alterindustrialdesignprocesses

ImproveworkplacerelationshipIncreasecompetitivenessIncreasefarmingproductivity(yields)IncreaseprofitabilityReduce competitiveness and market share through over-reliance andinvestmentinsingletechnologylineStunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Efficiency AlterindustrialdesignprocessesImprovesupplychainsImproveworkplacerelationshipsIncreasecompetitivenessIncreasefarmingproductivity(yields)Reduce competitiveness and market share through over-reliance andinvestmentinsingletechnologylineStunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Equality ConcentrateeconomicpowerIncreasejoblossesPrompttaxlossesStunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Justice ConcentrateeconomicpowerIncreasejoblosses


77

Affectedvalues ExamplesofrelatedidentifiedimpactsStunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Fairness ConcentrateeconomicpowerIncreasejoblossesPrompttaxlossesStunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Productivity AlterindustrialdesignprocessesImprovesupplychainsImproveworkplacerelationshipIncreasecompetitivenessIncreaseconsumerdemandIncreasefarmingproductivity(yields)StunttechnologicaldevelopmentinpoorerregionsduetohighcapitalcostsofAIandroboticsWeakentradeunions

Profitability AlterindustrialdesignprocessesImprovesupplychainImproveworkplacerelationshipsIncreasecompetitivenessIncreaseconsumerdemandPrompttaxlossesReduce competitiveness and market share through over-reliance andinvestmentinsingletechnologylineWeakentradeunions

Privacy EnhanceprivacyforconsumerservicesWell-being Concentrateeconomicpower

EnhanceprivacyforconsumerservicesIncreasediscriminationIncreasejoblossesPrompttaxlosses

Table21.Economicimpacts–effectsonvalues–asappearingintheinterviewsEnvironmental impact – effects on values Thetablesbelowshowthekeyvaluesthatwillbeaffectedbytheenvironmentalimpacts.Fromtheliterature

Affectedvalues Examplesofrelatedidentifiedimpacts

Ecologicalsustainability

Reduceresource-informationgaps(esp.helpingtounderstandland-usepatternsanddecisionsupport)

Environmentalintegrity

Increaseharmfromincreasedrobotuse(theircomponentmaterialsgenerallytoxicandnon-biodegradable)ReducewaterandenergyfootprintsReducefoodwasteReducepesticideuseReducewasteduringproductionReducenegativeeffectsofclimatechange

Regeneration ReducehumanpopulationResponsibleresourceuse Reduceresource-informationgaps(esp.helpingtounderstandland-use

patternsanddecisionsupport)


78

Table22.Environmentalimpacts–effectsonvalues–asappearingintheliterature(sourcescitedinEnvironmentalImpactstablefromsub-section4.2)FromtheInterviews

Affectedvalues ExamplesofrelatedidentifiedimpactsEcologicalsustainability IncreaseenergyusePollutioncontrol Increasepollutionthroughincreaseuseofcars

Table23.Environmentalimpacts–effectsonvalues–asappearingintheinterviewsThe above tables highlight a variety of values affectedby the social, economic and environmentalimpacts.Accordingtoourfindingsfromtheliteratureandinterviews,humanwell-being262(inmorewaysthanone)is(orwillbe)themostcommonlyaffectedsocialvalue.Themostcommonlyaffectedeconomicvaluesare(orwillbe)competitionandcompetitiveness,efficiencyproductivity,equalityandequityandprofitability.Environmentalintegritystandsoutasthemostcommonlyaffectedvalueoftheimpacts.Parties affected by the social, economic and environmental impacts OurliteraturereviewfocussedonanotheraspectoftheimpactsofAIandrobotics:affectedparties.Byaffectedparties(orparty),wemeanindividual(s),group(s)(e.g.,stakeholders,communities,societyetc.), entities (e.g., enterprises, government etc.) and/or some aspect of these (e.g., supply chain)changed or otherwise influenced by social, economic and/or environmental impacts. Below, weseparatelycategorisetheaffectedpartiesthatauthorsandintervieweesidentified.Togathertheinterviewees’viewsonthepartiesaffectedbytheimpactsofAIandrobotics,weaskedthisquestion:“Whodoyouthinkwillbemostaffectedbytheseimpacts?Ashighlightedinprecedingsub-sections,manyoftheidentifiedimpactsofAIandroboticsareofsuchnatureandgeographicalreachthattheyhavethepropensitytoaffect ‘all’parties– i.e., individualsand/orgroups throughout theworld (dependingon the implementation,uptakeanduseofAIandrobotics as well as effects of policies that support or restrict them). Nevertheless, authors andintervieweesalsopickedoutspecificaffectedparties;wepresentthembelowaccordingtoimpact(i.e.,social,economicorenvironmental).

Specific parties affected by social impacts Fromtheliteraturereview

• Communities and groups in society: might experience emergence of new forms of co-operationandinclusion

• Consumers:willseeeffectsofincreasedmass-customisationofproducts• Disfavoured stakeholders: will be affected by, e.g., bias of machines against specific

individualsand/orgroups

262 The importance of humanwell-being in relation to AI and robotics iswell borne out by The IEEEGlobalInitiative for Ethical Considerations in Artificial Intelligence and Autonomous Systems, op cit., 2018. Thedocumentaimsto“advanceapublicdiscussionabouthowwecanestablishethicalandsocialimplementationsforintelligentandautonomoussystemsandtechnologies,aligningthemtodefinedvaluesandethicalprinciplesthatprioritizehumanwell-beinginagivenculturalcontext”.


79

• Elderly: might be impacted positively (assistive AI and robotics for the elderly)263 andnegatively via invasions of privacy (e.g., remote electronic surveillance of elderly peoplebathingorchanging)264

• First-responders and other workers in hazardous environments:will reap benefits fromincreased safety of personnel in hazardous environments (e.g., robot replacement duringsearchandrescuemissions)265.

• Government (regulators, policy-makers): are affected by inadequate legal or regulatoryframework for handling potential enterprise malfeasance and antitrust practices andinadequatelegal/regulatoryframeworkfordeterminingliableagents266.

• Patientsandhealthcareproviders:willbepositivelyaffectedbyincreasingdemocratisationof access for healthcare, faster medical analysis and diagnosis and vast reduction – evenelimination–ofmedicalmisdiagnosis,benefitstoeldercareandtheinstitutionsthatprovideit,promotionofhealthierbehaviourinindividuals,andincreaseinhealthcaredecisionsbasedonevidenceandfreeofcognitivebiasesoroverconfidence267.

• Recipientsof insuranceand socialbenefits:mightexperience insuranceand socialbenefitdiscrimination (e.g., higher termination rate for benefit eligibility by religious group andincreasingautoinsurancepricesfornight-shiftworkers)268.

• Tenants(especiallyfromminorityethnicgroups):mightfacehousingdiscrimination(e.g.,alandlord relies on search results suggesting criminal history by race and matchingalgorithms)269.

• Workers:willbeaffectedbyjoblosses270.Fromtheinterviews• Communities and groups in society: might experience emergence of new forms of co-

operationandinclusion• Consumers:willseeeffectsofincreasedmass-customisationofproducts• Elderly:mightbeimpactedpositively(assistiveAIandroboticsfortheelderly)• Government(regulators,policy-makers):affectedbylossoftaxrevenue• Patients and healthcare providers:will be positively affected by increasing use of AI and

robotics• The inhabitants of poorer regions: many negative impacts of AI and robotics (and fewer

positiveones)willbe feltby the inhabitantsofpoorer regions,wheretherewillbe far lessdevelopmentofandwealthgeneratedbyAIandrobotics.

• Workers: will be affected by job losses, including through the movement of jobs viaoutsourcing.

Asidentifiedbothintheliteratureandinterviews,andofspecialnote,aretheimpactsonvulnerablepopulations:theelderly,(medical)patientsandlowerincomecountriesandpersons,especially.Theelderly and patients appear to benefit from improvements to healthcare and assistive medicalapplicationsofAIandrobotics.Thistracksthemostpositiveapplicationsandsectorsorfieldsidentifiedabove.Ifaccurate,thenanimportant,positivecontributionofAIandroboticsistheimprovementto

263PWCGlobal,opcit.,2017.264FreyandOsborne,opcit.,2016.265EuropeanParliament,opcit.,2016.266EzrachiandStucke,opcit.,2015.267PWCGlobal,opcit.,2017.268Smith,opcit.,2017.269Smith,opcit.,2017.270Stone,etal.,opcit.,2016.


80

thesevulnerablepopulations’conditionsandgeneralwell-being.Evenso, theelderly, inparticular,stand tohave their autonomyandprivacydiminishedor invadedby certain applicationsofAI androbotics(e.g.,remoteelectronicsurveillanceofelderlypeoplebathingorchanging)271.Suchgeneralimprovementsfortheelderlyandpatientsdonotseemtoholdforlowerincomepersonsandcountries,anotheraffectedpartywithinthevulnerablepopulationcategory.Here,considertheimpacts and affected parties of increased concentration of wealth in already wealthy andtechnologicallyadvancedcountries,diminishedcompetitivenessoflowerincomecountriesaswellasjoblosses.With respect to the economic impacts too, some impacts are of a global nature, i.e., have thepropensitytobefeltby‘all’.Theseinclude,e.g.,accelerationofeconomicinequalityaroundtheworld,improvedaccesstoretailservices,andimpactonlabourandwealthdistribution(thoughthismightalsospecificallyaffectworkersandcompanies).Specific parties affected by economic impacts Fromtheliteraturereview

• Consumers:might negatively experience a narrowing of choice (e.g., presented ads basedsolelyonpast“clicks”)272.

• Enterprises(all):Enterprises(largeandsmall)willfeeltheimpactsofAIandroboticsatvariouslevels. Examples of such impacts include: reduced risk; increased productivity throughaugmentation of existing labour force with AI and robotics; increased consumer demand(resulting from the availability of personalised and/or higher-quality AI-enhanced productsandservices)273;greaterprofitability;changesinorganisationalstructures;increasedefficiencyinfinancialtransactions,increasedworkflow274.

• Marketingcommunity:mightseearevolutionindigitalmarketingthroughchangestocriticalmarketingprocessessuchasleadnurturing,leadgenerationandsocialmedialistening(someofwhichisalreadyhappening)275.

• Medical/healthcareenterprisesandpatients:Costreductionsthroughrobot-assistedsurgicalproceduresandotherhealthcareimprovementsthroughAIandrobotics276.

• Theeconomy:mightseeimpactssuchassignificantchangesintheunemploymentrate,thePhilipsCurve,PurchasingPowerParity,GDP,inflation,money,managementandaccounting,loss of competitiveness and market share through over-reliance and investment in singletechnologyline277.

• Workers: may see changed and/or improved workplace relationships (e.g., emotionalintelligence and conventional intelligence will merge to facilitate improved workplace

271FreyandOsborne,opcit.,2016.272Smith,opcit.,2017.273RaoandVerweij,opcit.,2017.274Hislop,etal.,opcit.,2017.275Dirican,opcit.,2015.276FreyandOsborne,opcit.,2016;PwCGlobal,opcit.,2017.277Preimesberger,opcit.,2017;IEEE,opcit.,2018;Dirican,opcit.,2015;BerrimanandHawksworth,opcit.,2017;Arntzetal.,opcit.,2016;UNCTAD,opcit.,2017;Orchestrate,opcit.,2017;Hawking,opcit.,2016;Wisskirchen,etal.,opcit.,2017.


81

relationships)278, job losses (workers being replaced byAI and robots)279; employment andincome distribution changes through various channels, increased economic inequality,disruptionsinhowhumanlabourisaugmented,employmentdiscrimination(e.g.,filteringjobcandidatesbyraceorgenetic/healthinformationandfilteringcandidatesbyworkproximityleadstoexcludingminorities)280.

Fromtheinterviews

• Consumers:positively,consumerswillseeprivacyenhancementsforconsumerservices.• Enterprises(all):Enterprises(largeandsmall)willfeeltheimpactsofAIandroboticsatvarious

levels.• Farmers and agriculturists: might see increased farming productivity (yields) through

adaptive,smarteroptimisation.• Government(regulators,policy-makers):mightbeaffectedviataxlosses.• Medical/healthcareenterprisesandpatients:Generalimprovementstohealthcarethrough

AIandrobotics• Poorerand/ortechnologically lessadvancedcountriesandregions:ThecapitalcostsofAI

androboticsareveryhigh,makingpoorerregionslesscapableoftechnologicaldevelopment.Countrieswithlowerlabourcostsmayloseouttoautomation.

• Pre-existing conglomerates: There might be a further concentration of wealth to a fewcompanies.

• Supplychain:mightseeimprovementsinoperationsandmanagement.• Technologyownersandinvestors:willexperiencefurtherconcentrationofwealthfromtheir

assetsandinvestments.• Theeconomy:lossofcompetitivenessandmarketsharethroughover-relianceandinvestment

inasingletechnologyline.• Workers: may see changed and/or improved workplace relationships (e.g., emotional

intelligence and conventional intelligence will merge to facilitate improved workplacerelationships), job losses (workers being replaced by AI and robots); weakening of tradeunions;employmentandincomedistributionchangesthroughvariouschannels.

Specific parties affected by environmental impacts Fromtheliteraturereview

• Scientistsandresearchers:willseebenefitsfromthereductionofresource-informationgaps(especiallyhelpingtounderstandland-usepatternsanddecisionsupport)281.

• Societyandtheplanet:willbenefitfromreducedwaterandenergyfootprints282,decreasedhumanpopulation283andthereductionofnegativeeffectsofclimatechange284.Consequently,societyandtheplanetwillbeadverselyaffectedbyanincreaseinharmfromincreasedrobotuse(theircomponentmaterialsaregenerallytoxicandnon-biodegradablethougheffortsarebeingmadeandmightinthefuturebemadetochangethis)285.

278Preimesberger,opcit.,2017;279BerrimanandHawksworth,opcit.,2017;Arntzetal.,opcit.,2016;Dirican,opcit.,2015;UNCTAD,opcit.,2017;Hawking,opcit.,2016;Wisskirchen,etal.,opcit.,2017.280Smith,opcit.,2017.281Joppa,opcit.,2017.282EuropeanParliament,opcit.,2016.283PopulationMatters,opcit.,2016.284Jibo,opcit.,2016.285Parrack,opcit.,2012.


82

Fromtheinterviews• Energy providers: an increase in energy consumption (from the increased use of AI and

robotics).• Societyandtheplanet:possibledeleteriouseffectsofincreasedenergyconsumption

Inadditiontothespecificaffectedpartiesthemselves,wethinktheabovebringsoutapointthatbearsrepeating:impactsdonotaffectallpartiesequallyevenifimpactsingeneralaffectallparties.Notably,somemembersofvulnerablepopulationsbenefitfromimpactswhileothersufferfromthem.Thenextsub-sectionalsopresentsassessmentsthatmayhelptoclarifyandbetterdelimitthepartiesaffectedbysocial,economicandenvironmentalimpactsofAIandrobotics.Resilience and vulnerability of the affected parties to the impacts (described by interviewees) Togetasenseofthedegreetowhichaffectedpartiesmightflourishordeclineinlightoftheimpacts,weaskedintervieweesaboutaffectedparties’resilienceandvulnerabilitytoimpacts.Inaccordancewith the definition from the European Commission’s Communication on the EU Approach toResilience, resiliencemeans“theabilityofan individual,ahousehold,acommunity,acountryoraregiontowithstand,toadapt,andtoquicklyrecoverfromstressesandshocks”.286Resiliencehastwocrucialaspects:theinherentstrengthofapartyanditscapacitytobounceback.Byvulnerability,wemeantheinverseofresilience:theweaknessandsusceptibilityofapartytonegativeimpactsanditsability(ifany)torecoverfromnegativeimpacts.Consideringsuchmattersofresilienceandvulnerabilitycanpotentiallyhelptoframefutureresearchandpolicyinitiatives.Inthisrespect,theviewsofintervieweesareastepinbetterunderstandingnotonlythatcertainpartieswillbeaffected,butalsothatcertainaffectedpartiesmaybetteradaptandadjusttotheimpactsofAIandrobotics.Here,wesuggestthatresilientpartiesmayhavecertainvaluesorresourcesthat,ifdistributedtovulnerableones,couldincreasewell-beingandminimisepotentialharmsfromimpacts.Togaintheviewsofintervieweesonthesubjectsofresilienceandvulnerability,weposedthesetwoquestions:(1)“Howresilientarethepotentiallyaffectedcommunities?”(2)“Howvulnerablearetheyto the adverse impacts?” The table below summarises the interviewees’ responses to these twoquestions.Greaterresilience

• Westernsocietiesaremoreresilient• Communitiesbasedonstrongunifyingvaluesareresilient• Communitiesofinterestwillberesilient• Isolatedcommunitiesaremoreresilient• Peoplewithopenattitudesandtheabilitytorapidlyadapttochangearemostresilient

Conditionalresilience(i.e.,dependentonotherconditionsbeingmet)• Resilienceonlyifpublicsectorplaysalargeroleinaidingparties• Resiliencedependsonwhetherwe can introduce themeansand channel the technologies for

publicgood• Globalresilienceifcompetitiveblocsagreetoregulation

286EuropeanCommission,CommunicationfromtheCommissiontotheEuropeanParliamentandtheCouncil–TheEUApproachtoResilience:LearningfromFoodSecurityCrises,COM(2012)586Final,Brussels,3.10.2012.http://ec.europa.eu/echo/files/policies/resilience/com_2012_586_resilience_en.pdf


83

Greatervulnerability• Theglobalsouthismorevulnerable• Poorcountriesaremorevulnerable• Workerswholosejobsareveryvulnerable• Societiesinwhichmoreindividualslackeducationaremorevulnerable• Communitiesworldwidewillseeclashofvalues(thusmorevulnerable)

Table24.ResilienceandvulnerabilityofaffectedpartiesThequalitiesoftheresilientandvulnerablepartiesdescribedintheabovetablearelargelyinlinewithourabovefindings.Thatis,thebeneficiariesofpositiveimpacts–predominantlywealthy,educatedand technologically advanced populations – are also (or will also be) more resilient. Conversely,according to interviewees’ responses, the unemployed, lower income, less educated and lessdevelopedpopulationsaremorevulnerable.Weseenoveltyinthetablewithrespectto(a)thespecificvaluesofcertainresilientpartiesand(b)the“conditional resilience” category. Regarding the former, some interviewees (n=3) were keen todistinguishthevalues(andthecommunitiesassociatedwiththem)thatproduceordeepenresilience.Such values include religion, bondsdeveloped through long-termassociationand co-location (e.g.,neighboursandcompatriots).Isolatedcommunities(whetherbyvirtueofchosenseparation,suchascertainreligioussects,orluck-basedgeographicalisolation)alsofitwithinthisframeworkofresilience.Whethersuchinterest-,value-,orgeography-basedqualitiesequatetoorincreaseresilienceagainstthenegativeimpactsofAIandroboticsisapossibleareaoffutureresearch.The“conditionalresponses”categoryalsoprovesavaluablegroundforfutureresearch.Theresponsesroughly echo the larger pool ofmitigationmeasures proposed below by experts; namely,withoutregulationguidingthedevelopmentandimplementationofAIandrobotics,vulnerabilitywillincreaseand resilience decrease (or remain the same). This general conclusion needs evaluation in futureresearch,asdoesevaluationofwhichregulationsmayleadtothemostequitabledistributionsofAIandrobotics’benefits.Costs of the impacts Inthissub-section,wedeterminewhobearsthecosts–whetherprivate,opportunityorsocial–ofthenegativeimpactsofAIandrobotics.Althoughtherearecost-bearersassociatedwithpositiveimpacts,expertsprovidedscantdescriptionofthem.Consequently, thefollowingassessmentonlydescribescost-bearersofnegativeimpacts.Inordertogatheropinionsfromintervieweesaboutcosts,weposedthefollowingquestion:“Whatmightbethecostsoftheimpactsandwhomightbearthosecosts?”Negative social impacts – cost-bearers Thetablesbelowdepictsthecosts-bearersofnegativesocialimpactsstartingfromthegenerallevelandmovingontothemorespecific.Fromtheliteraturereview


84

Costbearer(s) SocialimpactsIndividualsandsociety Alterhumaninteractions

AlterlegalandregulatoryframeworksAltermoralconceptionsAlterunderstandingsofthescopeandlimitsofanaloguepersonhoodDiminishprivacyImprovehealthcareImprovetransportationsafetyIncreasebiasIncreasebigdataminingandanalysisIncreasediscriminationIncreaseharmandthreatofharmfromautonomousweaponsIncreaserulingclassdominationandwealthPromptunintendedconsequences

Individuals Diminishindividuals’controlovertheirdata

Table25.Negativesocialimpacts–cost-bearers–asdescribedintheliterature(sourcescitedinSocialImpactstableinsub-section4.2)Fromtheinterviews

Costbearer(s) SocialimpactsIndividualsandsociety Diminishindividuals’controlovertheirdata

DiminishpluralisminmediaDiminishprivacyImprove communities through better matching of people with similarinterestsImprovecybersecurityImprovedecision-makingthroughdataanalyticsImproveeldercareImprovehealthcareImprovelanguagetranslationIncreaseavailabletimeIncreasebiasIncreasebigdataminingandanalysisIncreasecyberwarfareIncreasediscriminationIncreaseharmandthreatofharmfromautonomousweaponsIncreasejoblossesIncreasepoliceprofilingIncreasesurveillancePromptnewformsofco-operationandinclusionReducerepetitivetasks

Individuals Diminishindividuals’controlovertheirdata

Minoritycommunities Increasepoliceprofiling

Patients Improvehealthcare

Workersandsociety ImproveprotectionsforworkersincountrieswithweakersafeguardsImprovesafetyofpersonnelininhazardousenvironments

Elderly ImproveeldercareMedia Reducepluralisminmedia

Table26.Negativesocialimpacts–cost-bearers–asdescribedintheinterviewsNegative economic impacts – cost-bearers Fromtheliteraturereview


85

Cost-bearers EconomicimpactsIndividualsandsociety

IncreaseconsumerdemandIncreasediscriminationIncreaseeconomicinequality

Business IncreasecompetitivenessIncreaseprofitabilityReduceriskReducecosts

Workersandsociety Increasecareerdisruption(i.e.,multiplejobchanges)Increasejoblosses

Businessesandworkers AlterindustrialdesignprocessesImproveworkplacerelationships

Businessmanagement ChangesinorganisationalstructuresTable27.Negativeeconomic impacts – cost-bearers –described in the literature (sources cited inEconomicImpactstableinsub-section4.2)Fromtheinterviews

Cost-bearers EconomicimpactsIndividualsandsociety

ConcentrateeconomicpowerEnhanceprivacyforconsumerservicesImproveaccesstoretailservicesIncreaseconsumerdemandIncreasediscrimination

Governmentsandsociety PrompttaxlossesBusiness Improvesupplychains

IncreasecompetitivenessReducecosts

Workersandsociety IncreasejoblossesTradeunionsandworkers WeakentradeunionsBusinessesandworkers Alterindustrialdesignprocesses

ImproveworkplacerelationshipsBusinessmanagement ChangesinorganisationalstructuresBusinesses,nationaleconomies,economicblocs,workers

Reduce competitiveness and market share through over-reliance andinvestmentinsingletechnologyline

Poorercountriesandregions

Stunt technological development in poorer regions due to high capitalcostsofAIandrobotics

Table28.Negativeeconomicimpacts–cost-bearers–describedintheinterviewsNegative environmental impacts – cost-bearers Thetablesbelowpresentthecost-bearersofthenegativeenvironmentalimpacts.Fromtheliteraturereview

Cost-bearer(s) EnvironmentalimpactsIndividuals,society

Increase harm from increased robot use (their component materialsgenerallytoxicandnon-biodegradable)Reducehumanpopulation

Table29.Negativeenvironmentalimpacts–cost-bearers–fromtheliteraturereview(sourcescitedinEconomicImpactstableinsub-section4.2)Fromtheinterviews

Cost-bearer(s) EnvironmentalimpactsIndividuals,society Increaseenergyconsumption


86

Cost-bearer(s) Environmentalimpacts

Table30.Negativeenvironmentalimpacts–cost-bearers–fromtheinterviewsInexaminingthematterofcosts,againrepresentedascost-bearers,wenotetheextenttowhichsomecosts(suchasthoseresultingfromimpactssuchasjoblosses)tosocietyappearandarestressedinexperts’analysisonthetopics.Althoughprivatecoststoindividualsalsofactorintothis,manyexperts’analysis of impacts focus on the broader implications of such widely adopted and implementedtechnology.Inthisrespect,individualsmayincursomeprivatecostsofjoblossesortheharmfuleffectsofpollution,butthewidespreadandestimablygreatestcostswillmostlybethesocial,infrastructuralandeconomiconesbornebysociety.Mitigation measures for negative impacts Weencounteredavarietyofspecificproposals forhowtomitigate thenegative impactsofAIandrobotics.However,policyandregulatorymeasureswerethemostoft-madesuggestionformitigationofadverseimpactsofAIandrobotics.Inordertofaithfullyrepresentassessmentswithoutneedlessproliferationofterms,wegroupedresponsesintothreecategories.Theyare:(1)policyandregulatorymitigation measures, (2) technological and industry-level measures, (3) society-level mitigationmeasures(non-policyorregulatory).Policy and regulatory measures include international, supranational, national, regional and localpoliciesandregulationsthataimtolessennegativeimpacts.Mainexamplesofthesemeasuresinclude:Fromtheliteraturereview

• dataprotectionregulationsthattakes intoaccountthreatstoprivacyfromAIandrobotics’applications287

• policyandregulationthatpreservethereasonableexpectationofprivacy288Fromtheinterviews

• international agreements regarding the implementation and use of particular applications(e.g.,autonomousweapons)

• taxationofrobot-workersforpurposesofmaintainingpublicrevenuesfordisplacedhumanworkers

• introductionofuniversalbasicincomeschemestosupportdisplacedhumanworkersTechnological and industry-level mitigation measures are those that, through the creation andimplementationof technologyby industryaimto lessennegative impacts.Mainexamplesof thesemeasuresinclude:Fromtheliterature

• thedesignofethicalalgorithms289• implementationofqualitycontrol290• oversightsystemsofAIandroboticsapplications291

287EuropeanDataProtectionSupervisor,opcit.,2016.288Pagallo,opcit.,2013289Dignum,opcit.,2017;Smith,opcit.,2017;Bossman,opcit.,2016;Burgess,etal.,opcit.,2018.290PwCGlobal,opcit.,2017.291EuropeanParliament,opcit.,2016;PwCGlobal,opcit.,2017.


87

Fromtheinterviews• diversificationofinvestmentsintechnologiestopreventlossofmarketshareorcompetitive

advantageSociety-level mitigation measures (non-policy or regulatory) represents a more diffuse set ofmeasuresthataimtoreducenegativeimpactsthroughtheeffortsofdiversemembersandgroupsofsociety.Mainexamplesofthesemeasuresinclude:Fromtheliterature:

• the co-ordination of non-governmental global elite to diminish the impact of autonomousweapons292

• multiplenon-governmentalorganisations’effortstoreviseandupdatetheoriesabouthumaninteractions293

Fromtheinterviews

• increasethepossibilitiesforcommunitiesofinterestthriveIn linewith these three categories, themajority of proposedmeasures (both in the literature andinterviews)fitwithinthe‘Policyandregulatory’grouping.Thiswasfollowedbythe'Technologicalandindustry-levelmitigationmeasures’,withonlyafewsociety-levelmitigationmeasuresproposed.It should be noted that only sometimes did interviewees provide highly detailed description ofmitigation measures, preferring broad descriptions such as regulation (without explaining orsuggestingwhichregulationsinparticularmightmitigateAIandrobotics’impacts).Regardlessoftherelativedetailorbreadthofproposedmitigationmeasures(fromeithertheliteratureorinterviews),theyallhelptoframesolutionsthatmayaddressall-pervasivetechnologiessuchasAIandroboticsandtheircurrentandexpectednegativeimpacts.Inthisrespect,themessagewedistilledisthis:collectiveactionbythemajorsocialandeconomicinstitutionsofsocietycanmitigatenegativeimpacts.Here,itisusefultorecalltheimpacts’cost-bearers:SocietyasawholewillbearmostoforthegreatestcostsofAIandrobotics’negativeimpacts.Theroleofpolicyandregulatorymeasuresis(atleastinpart)tolimitsuchcoststosocietyatlarge–doingsoostensiblyrequirestheco-operationand co-ordinating power of major institutions, such as national governments and internationalorganisations.4.4 Conclusions As described throughout this section, there is a wide range social, economic and environmentalimpactsstemmingfromAIandrobotics.Here,wepresentsomeconclusionsforamplifyingpositiveimpacts and minimising negative ones. To effectively prioritise the action needed to address theimpacts,weofferourconclusionsinaccordancewithwhentheyshouldbeadoptedorapplied:short-medium-andlong-term.Conclusionsfortheshort-termtomedium-term(fiveto10years)

• The effects of negative impacts without mitigation, especially for extremely destructiveapplicationssuchasautonomousweapons,willpotentiallyeffacethepositiveimpactsfromAIandrobotics.Ifthetimingofimpactspresentedaboveisaccurate,thenregulationandother

292Gubrud,opcit.,2016.293Burgess,etal.,opcit.,2018.


88

mitigation measures need implementation now to mitigate negative impacts currentlyoccurringandprojectedtodosointhefuture.

• Socialcostsofnegative impactsmeritspecialattention,andalleffortsto limitthemshouldbeginassoonaspossible.

• Job lossesandthethreattosocialandeconomicstability forworkersneedprioritisation inregulatoryandpolicydiscussions.

• EducatingindividualsaboutthepositiveimpactsofAIandroboticscanaidintheiracceptanceand reduce impacts associated with social strife, isolation and disaffection. However,presenting only positive attributes and impacts of AI and robotics may be perceived asdisingenuousordishonest.Consequently, increasing societal acceptanceofAI and roboticsshouldalsoincludediscussionsoftheirnegativeimpacts(andmitigationmeasurestakentoreducethem).

Conclusionsforthelong-term(20yearsandafter)

• Recognisethatdecisionsoccurringintheshort-andmedium-termswilllargelyconstructthestateoftheaffairsinthelong-term.Asaruleofthumb,avoidanyapplicationofAIandroboticswhoseexpectedlossesoutweightheprobablebenefitsinthelongrun.

InconcludingthisSEIA,werepeatthatAIandroboticsholdpromiseforaidinghumanbeingsinawiderangeofendeavours.BasedontheanalysisinthisSEIA,webelievethatthemorecompletepictureofAIandroboticsandtheirimpactsincludesthepossibilityofgreatburdens,too.Thisisnottostrikeachordofdoom,butrathertosaythereisaneedforconscientiousactiontoreduceAIandrobotics’negativeimpactsandmaximisethepositiveones.


89

5. Conclusion Thisdeliverablehaspresentedastate-of-the-artreviewofthefieldsofAIandrobotics. Itofferedathoroughanalysis of both fields in termsof their central concepts, their history, their present andanticipatedtechnologiesandapplications,aswellasasocio-economicimpactassessment(ofpresentandexpectedimpacts)oftheirtechnologies.

Wedeterminedthatbothfieldsareverybroadanddifficulttodefine.Inthem,wefoundthatthereexist a plethora of different approaches and areas of research,many ofwhich promising to bringsignificantchangeintermsofintroducingnoveltechnologiesandapplications.Forexample,weexpecttoseefurtheradvancesinAIareassuchasmachinelearning,neuralnetworks,computervisionandnatural languageprocessing,aswellas inrobotminiaturization,robotcontrol,robotsensing,cloudconnectivityandbiomimicry.AllofthiswillpavethewaytowardsAIsystemsandrobotsthataremoreintelligent,ubiquitous,autonomous,mobile,flexibleandsociable.

OuranalysishascoveredabroadarrayofpresentandfutureapplicationsforAIandrobotics.Itwasfound that applications are in such varied domains as transportation, infrastructure, healthcare,finance and insurance, security (military and law enforcement), retail and marketing, media andentertainment,companionship,science,education,manufacturingandagriculture.Therehavealreadybeenmajordevelopmentsinthesedomains,andouranalysisindicatesweareatthecuspofseeingmore significant development still, including fully autonomous cars, parcel delivery drones andconversationalAIsystemsincustomerservice,justtonameafew.

Oursocio-economicimpactassessmentidentifiedawiderangeofsocial,economicandenvironmentalimpactsstemmingfromAIandrobotics.Theserangefromnegativeimpacts,suchasthoseexpectedforautonomousweaponssystems,topositiveimpacts,suchasthoseexpectedformedicaldiagnosticapplications of AI and robotics.Weproposed that regulation andothermitigationmeasures needimplementationnowtomitigatethenegativeimpactsthatarecurrentlyoccurringandprojectedtodosointhefuture.Wefurthersuggestedthatthesocialcostsofnegativeimpactsmeritspecialattention,andalleffortstolimitthemshouldbeginassoonaspossible.Inaddition,wefoundthatjoblossesandthe threat to social andeconomic stability forworkers needprioritisation in regulatory andpolicydiscussions.Finally,wedeterminedthateducatingindividualsaboutthepositiveandnegativeimpactsofAIandroboticsmayaidintheiracceptanceandmayreduceimpactsassociatedwithsocialstrife,isolationanddisaffection.

Weendbynotingthatconsideringandassessingthefutureapplicationsandimpactsofcomplexandpervasive technologies such as AI and robotics is a speculative endeavour. Some of the expectedapplicationsandimpactswedescribedmaynotmaterialise,andwemayhaveomittedsomeimportantfutureapplicationsandimpactsthatarepresentlyhardtoforesee.


90

References • AccessScienceEditors,“ActiveTrafficManagement:AdaptiveTrafficSignalControl,”

AccessScience,2014.http://accessscience.com/content/active-traffic-managementadaptive-traffic-signal-control/BR0106141

• Ackerman,Evan,“RoboticConstructionPlatformCreatesLargeBuildingsonDemand,”IEEESpectrumRobotics,April26,2017.http://spectrum.ieee.org/automaton/robotics/industrial-robots/robotic-construction-platform-creates-large-buildings-on-demand

• Allison,PeterRay,“WhatDoesaBombDisposalRobotActuallyDo?”BBC,July15,2016.http://bbc.com/future/story/20160714-what-does-a-bomb-disposal-robot-actually-do

• Altman,Micah,AlexandraWood,DavidR.O’Brien&UrsGasser,“PracticalApproachestoBigDataPrivacyOverTime”,BrusselsPrivacySymposium,draftpaper,6November2016.

• AmericanGrippersInc.,“RoboticsandAutomatedAssemblyRelieveRepetitiveStrainInjuryinWorkers,”RoboticIndustriesAssociation:RoboticsOnline,February12,2016.http://agi-automation.com/2016/02/robotics-and-automated-assembly-relieve-repetitive-strain-injury-in-workers/

• Anandan,TanyaM.,“RobotsandHealthcareSavingLivesTogether,”RoboticIndustriesAssociation,November23,2015.http://robotics.org/content-detail.cfm/Industrial-Robotics-Industry-Insights/Robots-and-Healthcare-Saving-Lives-Together/content_id/5819

• Appenzeller,Tim,“TheAIRevolutioninScience,”ScienceMagazine,July7,2017.http://sciencemag.org/news/2017/07/ai-revolution-science

• Arntz,M.,T.GregoryandU.Zierahn,“TheRiskofAutomationforJobsinOECDCountries:AComparativeAnalysis”,OECDSocial,EmploymentandMigrationWorkingPapers,No.189,OECDPublishing,Paris,2016.http://dx.doi.org/10.1787/5jlz9h56dvq7-en

• Arvizo,Sarah,“ZestFinanceIntroducesMachineLearningPlatformtoUnderwriteMillennialsandOtherConsumerswithLimitedCreditHistory,”BusinessWire,February14,2017.http://businesswire.com/news/home/20170214005357/en/ZestFinance-Introduces-Machine-Learning-Platform-Underwrite-Millennials

• Asaro,Peter.“HowJustCouldaRobotWarBe?,”inAdamBriggle,KatinkaWaelbersandPhilipBrey(eds.),CurrentIssuesinComputingAndPhilosophy,IOSPress,Amsterdam,2008,pp.50-64.

• Ashok,Asokan,“TheImpactofArtificialIntelligenceinHealthcare,”Medium,August24,2017.http://medium.com/@Unfoldlabs/the-impact-of-artificial-intelligence-in-healthcare-4bc657f129f5

• Atherton,KelseyD.,“Meet‘WasteShark’,theDroneThat’sPickingupGarbagefromtheOceans,”PopularScience,September12,2016.http://popsci.com/waste-shark-is-garbage-collecting-sea-drone

• Autor,DavidH.,“WhyAreThereStillSoManyJobs?TheHistoryandFutureofWorkplaceAutomation”JournalofEconomicPerspectives,vol.29,no.3,2015.

• Barmpounakis,EmmanouilN.,EleniI.Vlahogianni,andJohnC.Golias,“UnmannedAerialAircraftSystemsforTransportationEngineering:CurrentPracticeandFutureChallenges,”InternationalJournalofTransportationScienceandTechnology,Vol.5,No.3,October2016,pp.111-122.PDF

• BBCTechnology,“PepperRobottoWorkinBelgianHospitals,”BBCNews:Technology,June14,2016.http://bbc.com/news/technology-36528253

• Bernier,Catherine,“WhatIsaCollaborativeRobot?,”RobotIQ,September20,2013.https://blog.robotiq.com/bid/66463/What-is-a-Collaborative-Robot

• Berriman,RichardandJohnHawksworth,Willrobotsstealourjobs?ThepotentialimpactofautomationontheUKandothermajoreconomies,PwCreport,2017.https://www.pwc.co.uk/economic-services/ukeo/pwcukeo-section-4-automation-march-2017-v2.pdf


91

• Bock,T.,T.Linner,andW.Ikeda,“ExoskeletonandHumanoidRoboticTechnologyinConstructionandBuiltEnvironment,”Intech,2012.http://cdn.intechopen.com/pdfs-wm/25773.pdf

• Boden,Margaret,JoannaBryson,DarwinCaldwell,KerstinDautenhahn,LilianEdwards,SarahKember,PaulNewman,VivienneParry,GeoffPegman,TomRodden,TomSorrell,MickWallis,BlayWhitby,AlanWinfield,“Principlesofrobotics:regulatingrobotsintherealworld”,ConnectionScience,vol.29,no.2,2017.https://doi.org/10.1080/09540091.2016.1271400

• Bogue,Robert,“RobotsinHealthcare,”IndustrialRobot:AnInternationalJournal,Vol.38,No.3.2011,pp.218-223.

• Bossmann,Julia,“Top9ethicalissuesinartificialintelligence”,WorldEconomicForum,21October2016.https://www.weforum.org/agenda/2016/10/top-10-ethical-issues-in-artificial-intelligence/

• Bourzac,Katherine,“AIinSpace,”IEEERobotics,August21,2017.http://spectrum.ieee.org/tech-talk/robotics/artificial-intelligence/ai-in-space

• Bowen,AndrewD.,DanaR.Yoerger,andChristTaylor,etal.,“TheNereusHybridUnderwaterRoboticVehicleforGlobalOceanScienceOperationsto11,000mDepth,”IEEEXplore,June30,2009.

• Brandes,DennisandLucZobrist,“ManandMachine:RobotsontheRise?TheImpactofAutomationontheSwissJobMarket”,Deloittereport,2015.https://www2.deloitte.com/content/dam/Deloitte/ch/Documents/innovation/ch-en-innovation-automation-report.pdf

• Brey,Philip,andJohnnyHartzSøraker.“PhilosophyofComputingandInformationTechnology,”inA.Meijers(Ed.),PhilosophyofTechnologyandEngineeringSciences(pp.1341-1408),VolIX,inD.Gabbay,P.ThagardandJ.Woods(Eds.),HandbookofthePhilosophyofScience,Elsevier,Amsterdam,2009.

• Brynjolfsson,Erik,AndrewMcAfee,andJeffCummings,TheSecondMachineAge:Work,Progress,andProsperityinaTimeofBrilliantTechnologies,W.W.Norton,NewYork,2014.

• Burgess,J.Peter,LucianoFloridi,AuréliePolsandJeroenvandenHoven,TowardsaDigitalFuture,EuropeanDataProtectionSupervisor,EthicsAdvisoryGroupreport,25January2018.https://edps.europa.eu/sites/edp/files/publication/18-01-25_eag_report_en.pdf

• Carey,Bjorn,“MaidenVoyageofStanford’sHumanoidRoboticDiverRecoversTreasuresfromKingLouisXIV’sWreckedFlagship,”StanfordNews,April27,2016.http://news.stanford.edu/2016/04/27/robotic-diver-recovers-treasures/

• Chowdhry,Amit,“HowArtificialIntelligenceisGoingtoAffecttheFinancialIndustryin2018,Forbes,February26,2018.http://forbes.com/sites/amitchowdhry/2018/02/26/clinc-artificial-intelligence/#80f299f481f9

• Chu,Eric,JonathanDunn,andDebRoy,etal.,“AIinStorytelling:MachinesasCocreators,”McKinsey&CompanyMedia&Entertainment,December2017.http://mckinsey.com/industries/media-and-entertainment/our-insights/ai-in-storytelling

• Cohn,Pamela,AlastairGreen,MeredithLangstaff,andMelanieRoller,“CommercialDronesareHere:TheFutureofUnmannedAerialSystems,”McKinsey&Co.,December2017.http://mckinsey.com/industries/capital-projects-and-infrastructure/our-insights/commercial-drones-are-here-the-future-of-unmanned-aerial-systems

• CommunicationsoftheACM,Vol.19,No.3,1976,pp.113–126.• Creswell,JohnW.,ResearchDesign:Qualitative,Quantitative,andMixedMethod

Approaches,4thEd.,Sage,LosAngeles,2014.• Davis,Alex,“MercedesPromisesSelf-DrivingTaxisinJustThreeYears,”WIRED,April2,2017.

http://wired.com/2017/04/mercedes-promises-self-driving-taxis-just-three-years/• Deahl,Dani,“SuspectCaughtinChinaatMusicConcertAfterBeingDetectedbyFacial

RecognitionTechnology,”TheVerge,April12,2018.


92

http://theverge.com/2018/4/12/17229530/china-facial-recognition-technology-suspect-music-festival

• DeloitteDevelopmentLLC,“FutureofRoboticsTechnology,”Government2020,n.d.http://government-2020.dupress.com/driver/robotics-technology/

• Dent,Steve,“IntelUnveilsanAIChipThatMimicstheHumanBrain,”Engadget,September26,2017.https://www.engadget.com/2017/09/26/intel-loihi-neuromorphic-chip-human-brain/

• Dickson,Ben,“HowArtificialIntelligenceisShapingtheFutureofEducation,”PCMag,November20,2017.http://pcmag.com/article/357483/how-artificial-intelligence-is-shaping-the-future-of-educati

• Dignum,Virginia,“DesigningAIforhumanvalues”,ICTdiscoveries,Specialissueno.1,25September2017.https://www.itu.int/en/journal/001/Documents/itu2017-1.pdf

• Dijkshoorn,Nick,“SimultaneousLocalizationandMappingwiththeAR.Drone,”UniversiteitvanAmsterdam,July14,2012.PDF.

• Dirican,Cüneyt,“TheImpactsofRobotics,ArtificialIntelligenceonBusinessandEconomics”,Procedia-SocialandBehavioralSciences,vol.195,3July2015.

• Edwards,LilianandMichaelVeale,“Enslavingthealgorithm:froma‘righttoanexplanation’toa‘righttobetterdecisions’?”draftpaper,October13,2017.https://ssrn.com/abstract=3052831

• ElectroTrainingandServicingInstitute,“RoboticsTechnology,”ETASI,n.d.https://etasieducation.com/course/robotics-technology/

• Espinoza,Javier,“DeliveryRobotsHittheStreets,butSomeCitiesOptOut,”FinancialTimes:ArtificialIntelligenceandRobotics,January31,2018.http://ft.com/content/0a2a5a76-e0ea-11e7-a0d4-0944c5f49e46

• EuropeanCommission,CommunicationfromtheCommissiontotheEuropeanParliamentandtheCouncil–TheEUApproachtoResilience:LearningfromFoodSecurityCrises,COM(2012)586Final,Brussels,3.10.2012.http://ec.europa.eu/echo/files/policies/resilience/com_2012_586_resilience_en.pdf

• EuropeanCommission,GuidanceforassessingsocialimpactswithintheCommissionimpactassessmentsystem,2009.http://ec.europa.eu/smart-regulation/impact/key_docs/docs/guidance_for_assessing_social_impacts.pdf

• EuropeanCommission,ImpactAssessmentGuidelines,15January2009.http://ec.europa.eu/smart-regulation/impact/commission_guidelines/docs/iag_2009_en.pdf

• EuropeanCommission,ImpactAssessmentTools.http://iatools.jrc.ec.europa.eu/bin/view/IQTool/WebHome.html

• EuropeanDataProtectionSupervisor,"ArtificialIntelligence,Robotics,PrivacyandDataProtection",Backgrounddocumentforthe38thInternationalConferenceofDataProtectionandPrivacyCommissioners,October2016.https://edps.europa.eu/data-protection/our-work/publications/other-documents/artificial-intelligence-robotics-privacy-and_en

• EuropeanParliament,EthicalAspectsofCyber-PhysicalSystem,June2016.http://www.europarl.europa.eu/thinktank/en/document.html?reference=EPRS_STU(2016)563501

• Ezrachi,Ariel,andMauriceE.Stucke,“ArtificialIntelligence&Collusion:WhenComputersInhibitCompetition”,OxfordLegalStudiesResearchPaperNo.18/2015;UniversityofTennesseeLegalStudiesResearchPaperNo.267,2015.http://dx.doi.org/10.2139/ssrn.2591874

• Faggella,Daniel,“AirbnbMachineLearning—HowDataandSocialScienceMakeitallWork,”TechEmergence,2017.http://techemergence.com/airbnb-machine-learning-how-data-and-social-science-make-it-all-work/


93

• Faggella,Daniel,“ArtificialIntelligenceinMarketingandAdvertising—5ExamplesofRealTraction,”TechEmergence,2018.http://techemergence.com/artificial-intelligence-in-marketing-and-advertising-5-examples-of-real-traction/

• Faggella,Daniel,“ArtificialIntelligenceinRetail—10PresentandFutureUseCases,”TechEmergence,2018.http://techemergence.com/artificial-intelligence-retail/

• Faggella,Daniel,“ExamplesofArtificialIntelligenceinEducation,”TechEmergence,2017.http://techemergence.com/examples-of-artificial-intelligence-in-education/

• Faggella,Daniel,“FromPasttoFuture,TracingtheEvolutionaryPathjofFinTech—AConversationwithBradBailey,TechEmergence,December2,2017.http://techemergence.com/from-past-to-future-tracing-the-evolutionary-path-of-fintech/

• Faggella,Daniel,“MachineLearninginFinance—PresentandFutureApplications,”TechEmergence,March27,2018.http://techemergence.com/machine-learning-in-finance-applications/

• FinancialStabilityBoard,ArtificialIntelligenceandMachineLearninginFinancialServices,November1,2017.http://fsb.org/wp-content/uploads/P011117.pdf

• ForexamplesseeSMPRobotics:http://smprobotics.com/application_autonomus_mobile_robots/transport-robots/

• Forsyth,DavidA.,andJeanPonce,ComputerVision:AModernApproach,PrenticeHall,UpperSaddleRiver(NewJersey),2003.

• Foster,Alana,“ApplyingArtificialIntelligencefromProductiontoDelivery,”IBCTechAdvances,January23,2018.http://ibc.org/tech-advances/ai-transforming-media-production-/2638.article

• Frey,CarlB.,andMichaelA.Osborne,“TechnologyatWorkV2.0:TheFutureIsNotWhatItUsedtoBe”,CitiGPSReport,2016.http://www.oxfordmartin.ox.ac.uk/publications/view/2092

• Galeon,DomandKristinHouser,“IBM’sWatsonAIRecommendsSameTreatmentasDoctorsin99%ofCancerCases,”Futurism,October28,2016.http://futurism.com/ibms-watson-ai-recommends-same-treatment-as-doctors-in-99-of-cancer-cases/

• Garfield,Leanna,“TheWorld’sBiggestMeatProducerisPlanningtoTestoutRobotButchers,”BusinessInsider,October26,2016.http://businessinsider.com/jbs-meatpacking-testing-robot-butchers-2016-10?international=true&r=US&IR=T

• Geraci,RobertM.,ApocalypticAI:VisionsofHeaveninRobotics,ArtificialIntelligence,andVirtualReality,Oxford,OxfordUP,2010.

• Girimonte,Daniela,andDarioIzzo,ArtificialIntelligenceforSpaceApplications,EuropeanSpaceAgency,2007.PDF.https://www.esa.int/gsp/ACT/doc/AI/pub/ACT-RPR-AI-2007-ArtificialIntelligenceForSpaceApplications.pdf

• Glant,Morton,andRobertKissell,Multi-AssetRiskModeling:TechniquesforaGlobalEconomyinanElectronicandAlgorithmicTrading,AcademicPress,2014,pp.258.

• Glaser,April,“TheIndustrialRoboticsMarketWillNearlyTripleinLessthan10Years,”Recode,June22,2017.http://recode.net/2017/6/22/15763106/industrial-robotics-market-triple-ten-years-collaborative-robots

• Goepfert,JessicaandMichaelShirer,“IDCForecastsWorldwideSpendingonRoboticstoReach$135Billionin2019DrivenbyStrongSpendingGrowthinManufacturingandHealthcare,”BusinessWire,February24,2016.http://businesswire.com/news/home/20160224005169/en/IDC-Forecasts-Worldwide-Spending-Robotics-Reach-135

• Grosz,Barbara,AlanMackworth,RussAltman,TomMitchell,EricHorvitz,DeidreMulligan,andYoavShoham,“ArtificialIntelligenceandLifein2030,”September2016.https://ai100.stanford.edu/sites/default/files/ai_100_report_0831fnl.pdf


94

• GuardianCities,“TheAutomatedCity:DoweStillNeedHumanstoRunPublicServices?”TheGuardian,2016.http://theguardian.com/cities/2016/sep/20/automated-city-robots-run-public-services-councils

• Gubrud,Mark,“WhyShouldWeBanAutonomousWeapons?”,IEEESpectrum,1June2016.• Guice,Jon,“DesigningtheFuture:TheCultureofNewTrendsinScienceandTechnology,”

ResearchPolicy,Vol.28,No.1,1999,pp.81–98.• Gupta,SatyandraK.SixRecentTrendsinRoboticsandTheirImplications,”IEEESpectrum,

September8,2015.https://spectrum.ieee.org/automaton/robotics/home-robots/six-recent-trends-in-robotics-and-their-implications

• Hardesty,Larry,“WhydoSomeNeighbourhoodsImprove?DensityofHighlyEducatedResidents,RatherthanIncomeorHousingCosts,PredictsRevitalization,”MITNews,July6,2017.http://news.mit.edu/2017/highly-educated-residents-neighborhoods-improve-0706

• Haugeland,John,ArtificialIntelligence,theVeryIdea,MITPress,Cambridge,1987.• Hawking,Stephen,"Thisisthemostdangeroustimeforourplanet",TheGuardian,1Dec

2016.https://www.theguardian.com/commentisfree/2016/dec/01/stephen-hawking-dangerous-time-planet-inequality

• Hempel,Jessi,“HowBaiduWillWinChina’sAIRace—and,Maybe,theWorld’s,”WIRED,August9,2017.http://wired.com/story/how-baidu-will-win-chinas-ai-raceand-maybe-the-worlds/

• HighwaysReporters,“RobotPotholeFillersBeingDeveloped,”TransportNetwork,February11,2015.http://highwaysmagazine.co.uk/robot-pothole-fillers-being-developed/

• Hislop,Donald,CrispinCoombs,StanimiraTanevaandSarahBarnard,“Impactofartificialintelligence,roboticsandautomationtechnologiesonwork”,CIPDRapidevidencereview,2017.

• Hollick,Victoria,“LeadingtheWayinHorticulturalRobotics,”UniversityofSydney,October6,2016.http://sydney.edu.au/news-opinion/news/2016/10/06/horticultural-robotics-centre-will-revolutionise-aussie-farming.html

• Hornyak,Timothy,“HowRobotsareBecomingCriticalPlayersinNuclearDisasterCleanup,”Science,March3,2016.sciencemag.org/news/2016/03/how-robots-are-becoming-critical-players-nuclear-disaster-cleanup

• Houser,Kristin,“TheSolutiontoOurEducationCrisisMightbeAI,”Futurism,December11,2017.http://futurism.com/ai-teachers-education-crisis/

• IBM.“AAAI-17Workshop:Human-AwareArtificialIntelligence(HAAI-17),”IBM,July25,2016.https://researcher.watson.ibm.com/researcher/view_group.php?id=7446

• IEEE,GlobalInitiativeforEthicalConsiderationsinArtificialIntelligenceandAutonomousSystems,EthicallyAlignedDesign:AVisionforPrioritizingHumanWellbeingwithArtificialIntelligenceandAutonomousSystems(AI/AS),version2,2018.http://standards.ieee.org/develop/indconn/ec/ead_v2.pdf

• Intel.“Intel-poweredAIHelpsFindMissingChildren”,Intel,http://intel.com/content/www/us/en/analytics/artificial-intelligence/article/ai-helps-find-kids.html

• InternationalFederationofRobotics,TheImpactofRobotsonProductivity,EmploymentandJobs,April2017.http://ifr.org/img/office/IFR_The_Impact_of_Robots_on_Employment.pdf

• InternationalFederationofRobotics.“WorldRoboticsReport2016,”IFR,2016.https://ifr.org/news/world-robotics-report-2016

• InternationalOrganizationforStandardization,“ISO8373:2012(en):Robotsandroboticdevices—Vocabulary,”n.d.https://www.iso.org/obp/ui/#iso:std:iso:8373:ed-2:v1:en

• Jang,Kitae,KoohongChung,andHwasooYeo,“ADynamicPricingStrategyforHighOccupancyTollLanes,”TransportationResearchPartA:PolicyandPractice,No.67,2014,pp.69–80.


95

• Jibo,"RobotsforaSustainableFuture",14July2016.https://blog.jibo.com/2016/07/14/robots-for-a-sustainable-future

• Joh,ElizabethE.,“PolicingPoliceRobots,”UCLALawReview,2016.• Joppa,Lucas,"Howartificialintelligencecouldsavetheplanet",TheGuardian,30January

2017.• Jung,Scott,“RoboticDogsGuidetheBlind:TheFutureofFetching,”TechInnovationIntelIQ,

August5,2014.http://iq.intel.com/the-future-of-fetching-robotic-dogs-guide-the-blind/• Kania,ElsaB.,“BattlefieldSingularity:ArtificialIntelligence,MilitaryRevolution,andChina’s

FutureMilitaryPower”,CenterforNewAmericanSecurity,November2017.http://s3.amazonaws.com/files.cnas.org/documents/Battlefield-Singularity-November-2017.pdf?mtime=20171129235804

• Kania,Elsa,“ThePolicyDimensionofLeadinginAI”,Lawfare,19October2017.https://www.lawfareblog.com/policy-dimension-leading-ai

• Kaushik,Preetam,“IsArtificialIntelligencethewayForwardforPersonalFinance,”Wired.http://wired.com/insights/2014/02/artificial-intelligence-way-forward-personal-finance/

• Kazerooni,H.,“HumanAugmentationandExoskeletonSystemsinBerkeley,”InternationalJournalofHumanoidRobotics,Vol.04,No.03,2007,pp.575–605.

• Kim,Gene,“ThisRobotWillGrowalltheFoodYouNeedinyourBackyard,”BusinessInsider,July21,2016.http://businessinsider.com/farming-robot-farmbot-automatically-grow-vegetables-backyard-garden-2016-7?international=true&r=US&IR=T

• Knuth.Donald.“ComputerScienceandItsRelationtoMathematics,”AmericanMathematicalMonthly,Vol.81,No.4,1974,pp.323-343.

• KPMGAdvisory.“SocialRobots,”KPMG,2016.https://assets.kpmg.com/content/dam/kpmg/pdf/2016/06/social-robots.pdf

• Lattimer,BrianY.,“RoboticsinFirefighting,”SFPE:EmergingTrendsinEngineering,2015.http://sfpe.org/?page=fpe_et_issue_100

• Lin,Patrick,KeithAbneyandGeorgeA.Bekey,“CurrentTrendsinRobotics:TechnologyandEthics,”RobotEthics:TheEthicalandSocialImplicationsofRobotics,MITPress,2012.

• Lohr,Steve,“BringingEfficiencytotheInfrastructure,”TheNewYorkTimes,April29,2009.http://nytimes.com/2009/04/30/business/energyenvironment/30smart.html

• Lou,Harbing,“AIinVideoGames:TowardaMoreIntelligentGame,”HarvardUniversity,August28,2017.http://sitn.hms.harvard.edu/flash/2017/ai-video-games-toward-intelligent-game/

• Lou,Harbring,“AIinVideoGames:TowardaMoreIntelligentGame,”HarvardUniversity,August28,2017.http://sitn.hms.harvard.edu/flash/2017/ai-video-games-toward-intelligent-game/

• Louie,DennisR.andJaniceJ.Eng,“PoweredRoboticExoskeletonsinPost-StrokeRehabilitationofGait:AScopingReview,”JournalofNeuroEngineeringandRehabilitation,June8,2016.http://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-016-0162-5

• Lune,HowardandBruceL.Berg,QualitativeResearchMethodsfortheSocialSciences,9thEdition,Essex,Pearson,2017.

• Ma,Jie,“NissanPlanstoIntroduceFullyAutonomousDrivingCarsin2022,”BloombergTechnology,December6,2017.http://bloomberg.com/news/articles/2017-12-06/nissan-plans-to-introduce-fully-autonomous-driving-cars-in-2022

• Machi,Vivienne,“’IronMan’SuitonTrack,butHurdlesRemain,”NationalDefense,May22,2017.http://nationaldefensemagazine.org/articles/2017/5/22/iron-man-suit-on-track-but-hurdles-remain

• MackenzieValleyEnvironmentalImpactReviewBoard,IssuesandRecommendationsforSocialandEconomicImpactAssessmentintheMackenzieValley,Report,2007


96

• MackenzieValleyEnvironmentalImpactReviewBoard,Socio-EconomicImpactAssessmentGuidelines,2007.

• Makridakis,Spyros,“TheforthcomingArtificialIntelligence(AI)revolution:Itsimpactonsocietyandfirms”,Review,Futuresvol.90,2017.https://doi.org/10.1016/j.futures.2017.03.006

• Marinov,Bobby,“MilitaryWearableExoskeletonResearch,”ExoskeletonReport,October3,2016.http://exoskeletonreport.com/2016/10/military-wearable-exoskeleton-research/

• Matthews,Kayla,“ThreeWaysRobotsCouldImproveInfrastructureinDevelopingCountries,”EngineeringforChange,September10,2017.http://engineeringforchange.org/news/robots-could-improve-infrastructure-developing-countries/

• May2012.https://newatlas.com/research-biodegradeable-eco-robots/22382/• Medler,DavidA.“ABriefHistoryofConnectionism,”NeuralComputingSurveys.Vol.1,1998,

pp.61–101.http://www.blutner.de/NeuralNets/Texts/Medler.pdf• Melkas,Helinä,LeaHennala,SatuPekkarinenandVilleKyrki,"Humanimpactassessmentof

robotimplementationinFinnishelderlycare",InternationalConferenceonServiceology,2016.

• Meyer,Jared,“UberandLyftAreChangingtheWayAmericansMoveAboutTheirCountry,”NationalReview,June7,2016.http://nationalreview.com/article/436263/uber-lyft-ride-sharing-services-sharing-economy-are-future

• Moor,James,TheTuringTest:TheElusiveStandardofArtificialIntelligence,KluwerAcademicPublishers,Dordrecht,2003.

• Morgan,Blake,“HowArtificialIntelligenceWillImpacttheInsuranceIndustry,”Forbes,July25,2017.http://forbes.com/sites/blakemorgan/2017/07/25/how-artificial-intelligence-will-impact-the-insurance-industry/#5255ab2e6531

• Muncaster,Phil,“SayHello,or你好,toChina’sSiri”,MITTechnologyReview,November16,2012.technologyreview.com/s/507416/say-hello-or-to-chinas-siri/.FormoreinformationonSenseTime,seetheirwebsite:http://sensetime.com/

• Muoio,Danielle,“BMWPlanstoTakeonMercedesbyReleasingaFullyDriverlessCarby2021,”BusinessInsider,March17,2017.http://businessinsider.nl/bmw-to-rival-mercedes-with-level-5-driverless-car-in-2021-2017-3/?international=true&r=US

• Nanalyze.“8CompaniesUsingAIforLawEnforcement,”Nanalyze,November12,2017.http://nanalyze.com/2017/11/8-companies-ai-law-enforcement/

• NationalResearchCouncil,DescribingSocioeconomicFuturesforClimateChangeResearchandAssessment:ReportofaWorkshop:PanelonSocioeconomicScenariosforClimateChangeResearchandAssessment.CommitteeontheHumanDimensionsofGlobalChange,DivisionofBehavioralandSocialSciencesandEducation.Washington,DC,TheNationalAcademiesPress,2010.

• Newell,Allen,andHerbertA.Simon.“ComputerScienceasEmpiricalInquiry:SymbolsandSearch,”CommunicationsoftheACM,Vol.19,No.3,1976,pp.113–126.

• OpenAccess:Government,“RiseoftheRobotsinthePublicSector,”AdjacentDigitalPoliticsLTD,December15,2017.http://openaccessgovernment.org/rise-robots-public-sector/40598/

• Orchestrate,“TheimpactofArtificialIntelligence&MachineLearningonthefutureofMarketing”,1December2017.http://www.orchestrate.com/blog/impact-artificial-intelligence-machine-learning-future-marketing/

• Outing,Steve,“YourFutureCompanioninYourOldAgeCouldbeaRobot,”MarketWatch,October13,2017.http://marketwatch.com/story/your-future-companion-cool-socially-adeptand-a-robot-2017-10-13


97

• Owen-Hill,Alex,“Top10RoboticApplicationsintheAgriculturalIndustry,”RobotIQ,August1,2017.http://blog.robotiq.com/top-10-robotic-applications-in-the-agricultural-industry

• Pagallo,Ugo,“Robotsinthecloudwithprivacy:Anewthreattodataprotection?”ComputerLaw&SecurityReportvol.29no.5,2013.https://www.researchgate.net/publication/259123308_Robots_in_the_cloud_with_privacy_A_new_threat_to_data_protection

• Panzarino,Matthew,“DisneyhasBegunPopulationitsParkswithAutonomous,Personality-DrivenRobots,”TechCrunch,February8,2018.http://techcrunch.com/2018/02/08/disney-has-begun-populating-its-parks-with-autonomous-personality-driven-robots/

• Parrack,Dave,"Researchersaimtobuildeco-friendlyrobotsfrombiodegradablematerials",NewAtlas,2

• Pekkeriet,E.J.andE.J.vanHenten,E.J.,“CurrentDevelopmentsofHigh-TechRoboticandMechatronicSystemsinHorticultureandChallengesfortheFuture,”GreenhouseTechnologyGroup:WageningenUniversity,2011.http://edepot.wur.nl/169021

• Pengel,Bob,LudolphyWentholt,ValeriaKrzhizhanovskaya,etal.,“TheUrbanfloodEarlyWarningSystem:SensorsandCoastalFloodSafety,”2010.http://edepot.wur.nl/197116

• PetaPixel,“Google'sImageCaptioningAICanDescribePhotoswith94%Accuracy,”PetaPixel,September23,2016.https://petapixel.com/2016/09/23/googles-image-captioning-ai-can-describe-photos-94-accuracy/

• Petersen,Sandra,SusanHouston,andHuanyingQin,etal.,“TheUtilizationofRoboticPetsinDementiaCare,”JournalofAlzheimersDisease,Vol.55,No.2,November19,2017,pp.596-574.PDF.

• Petrov,Andrey,MatthewBerman,JessicaGraybill,PhilipCavin,MatthewCooney,VeraKuklina,RasmusOleRasmussen,“MeasuringImpacts:AReviewofFrameworks,MethodologiesandIndicatorsforAssessingSocioEconomicImpactsofResourceActivityintheArctic”,ReSDAGapAnalysisReport#3,2013.

• Pettit,Harry,“Drone-CatchingDronesArmedwithNetscouldbeUsedtoBolsterSecurityatthisWeek’sPyeongchangWinterOlympics,”Dailymail,February6,2018.http://dailymail.co.uk/sciencetech/article-5357575/Drone-catching-drones-armed-nets-hit-Winter-Olympics.html

• Piazza,Bailey,“MachineMilitaries:TheFutureofArtificialIntelligenceandNationalSecurity,”DiplomaticCourier,January23,2018.http://diplomaticourier.com/machine-militaries-future-artificial-intelligence-national-security/

• PopulationMatters,"Theimpactofroboticsonfuturesocieties",briefingpaper,1May2016.https://www.populationmatters.org/documents/impact_of_robotics.pdf

• Preimesberger,Chris,“ExpertsPredictHowAIWillImpactUsin2018,”eWeek,2017.4December2017.http://www.eweek.com/innovation/predictions-2018-how-ai-machine-learning-will-impact-us

• ProceedingsoftheLondonMathematicalSociety,Series2,42,1937,pp.230-265.• PSFKLabs,“5AITrendsRedefiningtheRetailExperience,”PSFK,2018.

http://psfk.com/2018/01/5-ai-retail-trends-redefining-experience.html• PwCGlobal,WhyAIandroboticswilldefineNewHealth,globalsurveyreport,June2017.

https://www.pwc.com/gx/en/industries/healthcare/publications/ai-robotics-new-health/ai-robotics-new-health.pdf

• Pylyshyn,ZenonW.,ComputationandCognition:TowardaFoundationforCognitiveScience,MITPress,Cambridge,MA,1989.

• Rao,AnandS.andGerardVerweij,SizingthePrize:What’stherealvalueofAIforyourbusinessandhowcanyoucapitalise?PwCAIanalysisreport,2017.https://www.pwc.com/gx/en/issues/analytics/assets/pwc-ai-analysis-sizing-the-prize-report.pdf


98

• Rizzotto,Lucas,“TheFutureofEducation:HowAIandImmersiveTechWillReshapeLearningForever,”MediumFuturePi,June23,2017.http://medium.com/futurepi/a-vision-for-education-and-its-immersive-a-i-driven-future-b5a9d34ce26d

• RobotInstituteofAmerica.NBS/RIAroboticsresearchworkshop:ProceedingsoftheNBS/RIAWorkshoponRoboticResearchheldatGaithersburg,MD.,1979.

• Russell,Stuart,andPeterNorvig.ArtificialIntelligence:AModernApproach,3rdEdition,PrenticeHall,2010.

• Rutz,DominikandRainerJanssen(eds),Socio-EconomicImpactsofBioenergyProduction,Cham,Springer,2014.

• Sample,Ian,"ArtificialintelligencerisksGM-stylepublicbacklash,expertswarn",TheGuardian,1November2017.https://www.theguardian.com/science/2017/nov/01/artificial-intelligence-risks-gm-style-public-backlash-experts-warn

• Sandberg,Anders,andNickBoström.“WholeBrainEmulation:ARoadmap,”TechnicalReport#2008-3,FutureofHumanityInstitute,OxfordUniversity,2008.http://www.fhi.ox.ac.uk/Reports/2008-3.pdf

• Schmidt,Adrien,“HowAIImpactsEducation,”Forbes,December27,2017.http://forbes.com/sites/theyec/2017/12/27/how-ai-impacts-education/#5e93879c792e

• ScienceNewsStaff,“AIisChangingHowWedoScience.GetaGlimpse,”ScienceMagazine,July5,2017.http://sciencemag.org/news/2017/07/ai-changing-how-we-do-science-get-glimpse

• Searle,JohnR.Minds,brains,andprograms,BehavioralandBrainSciences,Vol.3,No.3,1980,pp.417–457.

• Sennaar,Kumba,“AIinAgriculture—PresentApplicationsandImpact,”TechEmergence,November16,2017.http://techemergence.com/ai-agriculture-present-applications-impact/

• Sennaar,Kumba,“AIinAgriculture—PresentApplicationsandImpact,”TechEmergence,November16,2017.http://techemergence.com/ai-agriculture-present-applications-impact/

• Sennaar,Kumba,“AIinMovies,Entertainment,andVisualMedia—5CurrentUse-Case,”TechEmergence,November20,2017.http://techemergence.com/ai-in-movies-entertainment-visual-media/

• Sennaar,Kumba,“ExamplesofAIinRestaurantsandFoodServices,”TechEmergence,2018.http://techemergence.com/ai-in-restaurants-food-services/

• Sennaar,Kumba,“MachineLearninginGenomics—CurrentEffortsandFutureApplications,”TechEmergence,2018.http://techemergence.com/machine-learning-in-genomics-applications/

• Sennaar,Kumba,“MachineLearninginGenomics—CurrentEffortsandFutureApplications,”TechEmergence,2018.http://techemergence.com/machine-learning-in-genomics-applications/

• Shademan,Azad,RyanS.Decker,andJustinD.Opfermann,etal.,“SupervisedAutonomousRoboticSoftTissueSurgery,ScienceTranslationalMedicine,Vol.8,No.337,May4,2016,pp.337-364.

• Sharkey,Noel,AimeevanWynsberghe,ScottRobbins,andEleanorHancock,OurSexualFuturewithRobots:AFoundationforResponsibleRoboticsConsultationReport,2017.http://responsible-robotics-myxf6pn3xr.netdna-ssl.com/wp-content/uploads/2017/11/FRR-Consultation-Report-Our-Sexual-Future-with-robots-1-1.pdf

• Shin,Hoo-Chang,HolgerR.Roth,MingchenGao,etal.,“DeepConvolutionalNeuralNetworksforComputer-aidedDetection:CNNArchitectures,DatasetCharacteristicsandTransferLearning,”IEEETransactionsonMedicalImagingVol.35,No.5,2016,pp.1285–1298.

• Silver,David,andDemisHassabis.“AlphaGo:MasteringtheancientgameofGowithMachineLearning.”GoogleResearchBlog,January27,2016.https://research.googleblog.com/2016/01/alphago-mastering-ancient-game-of-go.html


99

• Simonite,Tom,“AICanHelpAppleWatchPredictHighBloodPressure,SleepApnea,”Wired,November13,2017.http://wired.com/story/ai-can-help-apple-watch-predict-high-blood-pressure-sleep-apnea/

• Sims,ArthurG,andKennethWDobinson,“TheSydneyCoordinatedAdaptiveTraffic(SCAT)SystemPhilosophyandBenefits,”IEEETransactionsonVehicularTechnology.Vol.29,No.2,1980,pp.130–137.

• Siwicki,Bill,“AIChatbotsMightbetheMoney-SaversHospitalsareLookingFor,”HealthcareITNews,September18,2017.http://healthcareitnews.com/news/ai-chatbots-might-be-money-savers-hospitals-are-looking

• Smith,Lauren,"UnfairnessbyAlgorithm:DistillingtheHarmsofAutomatedDecision-Making",FutureofPrivacyForum,2017.

• Sonsev,Veronika,“WillAIbetheFutureofRetail?”ForbesTech,December15,2017.forbes.com/sites/veronikasonsev/2017/12/15/predicting-the-future-of-retail-a-vc-perspective/#6c9941387ff1

• Stephen,Andrew,“AIisChangingMarketingasWeKnowIt,andThat’saGoodThing,”ForbesNetwork,October30,2017.http://forbes.com/sites/andrewstephen/2017/10/30/ai-is-changing-marketing-as-we-know-it-and-thats-a-good-thing/#2ea5197cdc40

• Sterling,Greg.“Googlesays20%ofmobilequeriesarevoicesearches,”SearchEngineLand,May18,2016.https://searchengineland.com/google-reveals-20-percent-queries-voice-queries-249917

• Sterniczky,Aaron,AdamOstolski,IoanaBanach-Sirbu,JaakkoStenhäll,JanPhilippAlbrecht,JaroslavFiala,LuisEstebanRubio,MohssinElGhabri,NóraDiószegi-Horváth,PieterPekelharing,TiffanyBlandin“GreenObservatory:RoboticsandArtificialIntelligence,”2017.https://www.greeneuropeanjournal.eu/green-observatory-robotics-and-artificial-intelligence/

• Strickland,Eliza,“FordAssemblyLineWorkersTryOutExoskeletonTechtoBoostPerformance,”IEEESpectrum,November10,2017.http://spectrum.ieee.org/the-human-os/biomedical/bionics/ford-assembly-line-workers-try-out-exoskeleton-tech-to-boost-performance

• Synchrono,By,“2017TopTenTrendsinModernDemand-DrivenManufacturing,”SupplyChain247,June14,2017.http://supplychain247.com/paper/2017_top_ten_trends_in_modern_demand_driven_manufacturing

• Szondy,David,“RobearRobotCareBearDesigntoServeJapan’sAgingPopulation,”NewAtlas:Robotics,February27,2015.http://newatlas.com/robear-riken/36219/

• Tarantola,Andrew,“RobotCaregiversareSavingtheElderlyfromLivesofLoneliness,”Engadget,August29,2017.http://engadget.com/2017/08/29/robot-caregivers-are-saving-the-elderly-from-lives-of-loneliness/

• Thankappan,Jeevan,“SaudiUsesAIforCrowdControlatGrandMosqueinMecca,”TahawulTech,December5,2017.http://tahawultech.com/news/saudi-uses-ai-crowd-control-grand-mosque-mecca

• TharsusGroup,“2018TrendsinRobotics,”Tharsus,n.d.https://www.tharsus.co.uk/2018-trends-in-robotics/

• Thomas,Jim,“HowCorporateGiantsareAutomatingtheFarm,”NewInternationalist,November1,2017.http://newint.org/features/2017/11/01/agriculture-robots

• Thomsen,Michael,“Microsoft'sDeepLearningProjectOutperformsHumansinImageRecognition,”Forbes,ForbesMagazine,February19,2015.https://www.forbes.com/sites/michaelthomsen/2015/02/19/microsofts-deep-learning-project-outperforms-humans-in-image-recognition/#4fb4d93d740b


100

• Tilley,Jonathan.“Automation,robotics,andthefactoryofthefuture,”McKinseyInsights,September2017.https://www.mckinsey.com/business-functions/operations/our-insights/automation-robotics-and-the-factory-of-the-future

• TowardsDataScience.“ApplyingAIandMachineLearningtoMediaandEntertainment,”TowardsDataScience,2017.http://towardsdatascience.com/applying-ai-and-machine-learning-to-media-and-entertainment-8d3ebc4b68f7

• Turing,Alan.“ComputingMachineryandIntelligence,”Mind,Vol.49,1950,pp.433-460.• UK-RAS,RoboticandAutonomousSystemsforResilientInfrastructure,UK-RASNetwork,

2017.PDF.• VanHooijdonk,Richard,“WillRobotsbeourKids’FutureToys,Teachers,andCompanions?”

August3,2016.http://richardvanhooijdonk.com/en/will-robots-kids-future-toys-teachers-companions/

• Vanclay,Frank,AnaMariaEsteves,IlseAucampandDanielFranks,“SocialImpactAssessment:Guidanceforassessingandmanagingthesocialimpactsofprojects”,InternationalAssociationforImpactAssessment,Fargo,ND,2015.https://www.iaia.org/uploads/pdf/SIA_Guidance_Document_IAIA.pdf

• Vaqar,Ali,“GardenSpaceisaNewPersonalGardeningRobotthatWatersyourPlantsandScaresAwayAnimals,”WonderfulEngineering,November,2017.http://wonderfulengineering.com/gardenspace-personal-gardening-robot-waters-plants-well-protects/

• Vis,Martijn,Anne-SophieDörnbrackandSébastienHaye,“Socio-EconomicImpactAssessmentTools”inDominikRutzandRainerJanssen(eds),Socio-EconomicImpactsofBioenergyProduction.Cham:Springer,2014.

• Wales,Jimmy,“ElonMuskWillMakeDriverlessCarsaRealitySoonerthanyouThink,”WIRED,January18,2018.http://wired.co.uk/article/fully-autonomous-cars-are-almost-here

• Walker,Jon,“FastFoodRobots,Kiosks,andAI:UseCasesfrom6RestaurantChainGiants,”TechEmergence,January27,2018.http://techemergence.com/fast-food-robots-kiosks-and-ai-use-cases/

• Walker,Jon,“IndustrialUsesofDrones—5CurrentBusinessApplications,”TechEmergence,July6,2017.http://techemergence.com/industrial-uses-of-drones-applications/

• Walker,Kris,“EnvironmentallyFriendlyRobots,”AZOCleantech,February15,2013.http://azocleantech.com/article.aspx?ArticleID=368

• Ward,Tom,“NASA:AIwillLeadtheFutureofSpaceExploration,”Futurism,June27,2017.http://futurism.com/nasa-ai-will-lead-the-future-of-space-exploration/

• West,DarrellM.,“Whathappensifrobotstakethejobs?Theimpactofemergingtechnologiesonemploymentandpublicpolicy”,BrookingsInstitution,2015.https://www.brookings.edu/wp-content/uploads/2016/06/robotwork.pdf

• Wisskirchen,Gerlind,BlandineThibaultBiacabe,UlrichBormann,AnnemarieMuntz,GundaNiehaus,GuillermoJiménezSoler,BeatricevonBrauchitsch,ArtificialIntelligenceandRoboticsandTheirImpactontheWorkplace.TheIBAGlobalEmploymentInstitute,2017.

• Wood,Laura,“ArtificialIntelligenceMarketinEnergy&Utilities-Forecast(2017-2022)-ResearchandMarkets,”BusinessWire,December29,2017.http://businesswire.com/news/home/20171229005229/en/Artificial-Intelligence-Market-Energy-Utilities---Forecast

• Yao,Mariya,“FutureFactories:HowAIEnablesSmartManufacturing,”TopBots,October27,2017.http://medium.com/topbots/future-factories-how-ai-enables-smart-manufacturing-c1405f4ec0e6

• Zingman,Alissa,ScottEarnest,BrianLowe,andChristineBranche,“ExoskeletonsinConstruction:WillTheyReduceorCreateHazards?”CentersforDiseaseControland


101

Prevention,June15,2017.http://blogs.cdc.gov/niosh-science-blog/2017/06/15/exoskeletons-in-construction/


102

Annexes

Annex 1: Interview questions

1. DoyouthinkthatAIandroboticswillhavebeneficialornegativeimpacts?

2. Which applications of AI and robotics do you think might have the most beneficial oradverseimpactsonindividualsorsociety?

3. AretherecertainsectorsorfieldsinwhichAIandroboticsmighthavethemostbeneficialoradverseimpacts?

4. Whichimpactscanyouforesee:

o Inthenext5-10years?

o Inthenext20years?

5. Whodoyouthinkwillbemostaffectedbytheseimpacts?

6. Howfarandwide,geographically,couldindividualsandcommunitiesfeeltheimpacts?

7. Willtheimpactssupportorunderminetheaffectedcommunities’orsocietalvalues?

8. Howresilientare thepotentiallyaffectedcommunities?Howvulnerableare they to theadverseimpacts?

9. Whatmightbethecostsoftheimpactsandwhomightbearthosecosts?

10. Doyouthinkitispossibletomitigatethenegativeimpacts?Ifso,how?

11. Arethereexistingmitigationmeasuresthathaveworkedforthesetypesofimpacts?Ifso,howcanweusethem?