Page 1
Good eveningandthanksforhavingmehere.TodayIwanttolookathowourrelationshiptotheworldchangeswhenwe’resurroundedbydevicesthatanticipateourneedsandactonthem.Thatmeansitsitsattheintersectionoftheinternetofthings,userexperiencedesignandmachinelearning,andalthoughpeoplehavedealtwithoneofthosedisciplinesbefore,Idon’tthinkthey’veeverbeencombinedinquitethewaystheyarenow,orwiththecurrententhusiasm.And,tobeclear:Iamneitherafanof,noracriticof,thesetechnologies.Ithinkthey’retoocomplextobereducedthatwayandtomaximizetheirpositiveimpactwehavetoactivelyengagewiththem,andthat’swhatthistalkistryingtodo.
Thetalkisdividedintoseveralparts:itstartswithanoverviewofhowIthinkInternetofThingsdevicesareprimarilycomponentsofservices,ratherthanbeingself-containedexperiences,howpredictivebehaviorenableskeycomponentsofthoseservices,andthenIfinishbyexploringsomespeculativeideasofwhatkindofimpactthey’regoingtohaveonus,asindividualsandasasociety.AtitscoreisanargumentthateverythingisgoingtobeconnectedtotheInternet,thatthosethingswilleachtrytopredictourimmediatefuture,andthatthisisgoingtofundamentallychangeourrelationshiptotheworld.
Acoupleofcaveats:- Mycurrentworkinthisfieldfocusesalmostexclusivelyontheconsumerinternetofthings,soIseemostthingsthroughthatlens.- IwanttopointoutthatfewifanyoftheissuesIraisearenew.Thoughtheterms“internetofthings”and“machinelearning”arehotrightnow,theideashavebeendiscussedinresearchcirclesfordecades.Searchfor“ubiquitouscomputing,”“ambientintelligence,”and“pervasivecomputing”andyou’llseealotofgreatthoughtinthespace.Ifyou’rereallyambitious,youcanreadtheArtificialIntelligenceandCyberneticsworksofthe50sand60sandyou’llbesurprisedbytheprescienceofthepeopleworkinginthisspacewhentheentireworld’scomputepowerwasaboutasmuchasmykeyfob.- Therearealotofideashere,andIwillalmostcertainlyunder-explainsomething.ForthatIapologizeinadvance.Mygoalhereistogiveyouageneralsenseofhowthesethepiecesconnect,ratherthananin-depthexplanationofanyoneofthepieces.- Finally,mostofmyslidesdon’thavewordsonthem,soI’llmakethecompletedeckwithatranscriptavailableassoonI’mdone.
0
Page 2
Let me begin by telling you a bit about my background. I�m a user experience designer. I was one of the first professional Web designers. This is the navigation for a hot sauce shopping site I designed in the spring of 1994.
1
Page 3
I’vealsoworkedontheuserexperiencedesignofalotofconsumerelectronicsproductsfromcompaniesyou’veprobablyheardof.
2
Page 4
Now, aquickaside.WhatisUserExperienceDesign?UXdesignisnotgraphicdesign,interfacedesign,ergonomics,industrialdesign,orproductdesign,butitincludesaspectsofallofthosethings.
UXdesignisahumanisticproblemsolvingapproachthatbringstogethertheneedsofpeopleandbusinessestocreatetechnologicalproductsthatarevaluableforbothgroups.It’smuchmoreaboutprocessthanmakingthingslookgood.
Thefieldisabout20yearsold.Thisishowitlookedabout15yearsago.
DiagrambyJessMcMullin.
3
Page 5
It’salittlemorecomplextoday,but’sroughlythesamething.
DiagrambyCoreyStern.
4
Page 6
Iwroteacoupleofbooksbasedonmyexperienceasadesigner.Oneisacookbookofuserresearchmethods,andtheseconddescribeswhatIthinkaresomeofthecoreconcernswhendesigningnetworkedcomputationaldevices.I’malsomarriedtooneoftheauthorsofthisbook,sothinkingabouttheimpactofthedesignofconnecteddevicesonpeopleiskindofafamilybusiness.
5
Page 7
Ialsostartedacoupleofcompanies.Thefirst,AdaptivePath,wasprimarilyfocusedontheweb, andwiththesecondone,ThingM,Igotdeepintodevelopinghardware.
6
Page 8
TodayIworkforPARC,thefamousresearchlabthatinventedthepersonalcomputer,objectorientedsoftware,thetabletcomputer,andlaserprinter,asaprincipalinitsInnovationServicesgroup.Wehelpcompaniesreducetheriskofadoptingnoveltechnologiesusingamixofsocialresearch,designandbusinessstrategy.
7
Page 9
PARCalsostartedthinkingaboutwhatwecalltheIoT longbeforemost othercompanies.
ItwasatPARCin1971thatDickShoup,andearlyPARCresearcher,wrotethateventuallyprocessorswouldbeascommon,andasinvisibleaselectricmotors.Thisclearlyoutlinesthedestinyofconnectedcomputer:thateventuallyitwillbecomeasboringandascommonaselectricmotorsaretoday.
8
Page 10
Inthelate80s,alsoatPARC,MarkWeisercoinedthetermubiquitouscomputingtodescribeafuturewhenthenumberofcomputerssurpassedthenumberofpeopleusingthem.Inthischartfrom20yearsago,hepredictedthatwouldhappenaround2005.Hedidn’tlivetoseethatcrossover,buthewasbasicallyright—theiPhonelaunchedin2007—andwenowliveintheworldheenvisioned.
Essentially,whatwenowseeasanovelphenomenonhasbeenforseen bypeopleintheindustryformanydecades.Thequestionshavealwaysbeennotaboutwherewe’regoing,butwhenwe’llgetthere,andhow.
9
Page 11
Buttheendvisiondoesn’tappearallatonce.We’veonlystartedthetransitiontotheubiquitouscomputingworld,andassuch,we’reseeingalotofbadideasaboutwhattheInternetofThingsisanditisn’t.Essentially,everythingthatcanbeconnectedtotheInternetwillbe,whichincludesalotofthingsthatshouldn’tbe.TherearesomanybadideasnowthatthereareentireTumblrs dedicatedtomockingstupidIoTideas.Oneisaboutdumbsmartthingsandtheotherisjustaboutdumbsmartrefrigerators.
10
Page 12
Mostofthesethingsarebadideasbecausesimplyconnectingexistingstufftotheinternetdoesnotproducecustomervalue…
11
Page 13
Simpleconnectivityhelpswhenyou’retryingtomaximizetheefficiencyofafixedprocess,butthat’snotaproblemthatmostpeoplehave.We’vebeenabletosimplyconnectvariousdevicestoacomputersinceaTandyColorComputerscouldlightsoffandonoverX10in1983.Theproblemisthatthatwasn’tveryusefulthen,andit’snotveryusefulnow.IfyoureplacetheTandywithaniPhoneandthelampwithawashingmachine…
12
Page 14
…oraneggcarton,youstillhavethesameproblem,andit’sauserexperienceproblem.
TheUXproblemisthatendusershavetoconnectallthedotstocoordinatebetweenawidevarietyofdevices,andtointerpretthemeaningofallofthesesensorstocreatepersonalvalue.Formanysimplyconnectedproductsthereissolittleefficiencytobehadrelativetothecognitiveloadthatit’s justnotworthit.What’sworse,theextracognitiveloadisexactlyoppositetowhattheproductpromises,andcustomersfeelintenselydisappointed,perhapsevenbetrayed,whentheyrealizehowlittletheygetoutofsuchaproductThatmakesmostsuchproductseffectivelyWORSEthanuseless.
Thatpromisegapiswhatdistinguishesagadgetfromatool,whythiseggcartonisfunny,andwhyQuirkywhomadeit,filedforbankruptcyafterburningthroughhundredsofmillionsofdollars.
13
Page 15
How doyoucreateatoolthatreducescognitiveloadinsteadofcreatingit,thatexchangespeople’sprecioustimeforsignificantvalue?Oneapproachistocouplecloud-basedserviceswithpredictivemachinelearningmodelstoanticipatewhatbehaviorswillmaximizethechancesofadesirableoutcomeinagivensituation.
14
Page 16
WhenI talkaboutservices,I’mtalkingaboutthinkingofhardwaredevicesasphysicalrepresentativesofcloudservices,whichmakesthemverydifferentthantraditionalconsumerelectronics.Historically,acompanymadeanelectronicproduct,sayaturntable,theyfoundpeopletosell itforthem,theyadvertiseditandpeopleboughtit.Thatwastraditionallytheendofthecompany’srelationshipwiththecustomeruntilthatpersonboughtanotherthing,andallofthevalueoftherelationshipwasinthedevice.WiththeIoT,thesaleofthedeviceisjustthebeginningoftherelationshipandphysicalthingholdsalmostnovalueforeitherthecustomerorthemanufacturer.
15
Page 17
Value now shifts to services and the devices, software applications and websites used to access it—its avatars—become secondary. A camera becomes a really good appliance for taking photos for Instagram, while a TV becomes a nice Instagram display that you don’t have to log into every time, and a phone becomes a convenient way to check your friends’ pictures on the road.
Hardware, physical things, become simultaneously more specialized and devalued as users see “through” each device to the service it represents. The avatars exist to get better value out of the service.
16
Page 18
Amazon reallygetsthis.Here�satellingolderadfromAmazonfortheKindle. It’ssaying�Look,usewhateverdevice youwant.Wedon�tcare,aslongyoustayloyaltoourservice.Youcanbuyourspecializeddevices,butyoudon�thaveto.�
17
Page 19
WhenFirewasreleased5yearsago,JeffBezosevencalled itaservice.
18
Page 20
AmazonDashisaservicethat’senabledbydedicateddevices.ADashbuttonisanetworkedcomputerwhoseonlypurposeistobeanavatarforamacaroniandcheeseservice.
19
Page 21
Mostlarge-scaleIoT productsareserviceavatars.Theyusespecialized sensorsandactuatorstosupportaservice,buthavelittlevalue—ordon’tworkatall—withoutthesupportingservice.SmartThings,whichwasacquiredbySamsung, clearlystatesitsserviceofferingrightupfrontontheirsite.Thefirstthingtheysayabouttheirproductlineisnotwhatthefunctionalityis,butwhateffecttheirservicewillachievefortheircustomers.Theirhardwareproducts’functionality,howtheywilltechnicallysatisfytheservicepromise,isalmostanafterthought.
20
Page 22
ComparethattoX10,theirspiritualpredecessorthat’sbeeninthebusinessfor30years.AllthatX10tellsisyouiswhatthedevicesare,notwhattheservicewillaccomplishforyou.Idon’tevenknowifthereISaservice.WhyshouldIcarethattheyhave“modules”?Ishouldn’t,andIdon’t.
21
Page 23
Ithink therealvalueconnectedservicesofferistheirabilitytomakesenseoftheworldonourbehalf,toreducecognitiveloadbyenablingpeopletointeractwithdevicesatahigherlevelthansimpletelemetry,atthelevelofintentionsandgoals,ratherthandataandcontrol.Humansarenotbuilttocollectandmakesenseofhugeamountsofdataacrossmanydevices,ortoarticulateourneedsassystemsofmutuallyinterdependentcomponents.Computersaregreatatit.
22
Page 24
Theydothisthroughprocessesthathavemanynames,butI’lllumpthemallunderMachineLearning,whichisabigpartofwhatusedtobecalledArtificialIntelligence.Manyofthecoreideasheregobacktothe1950sandit’sthebasisofeveryemailspamfilter,soifyou’vehadyourspamautomaticallyfiltered,you’veexperiencedthevalueofmachinelearning.
AbigpartofMachineLearningispatternrecognition.Wehumansevolvedverysophisticatedfacultiestorapidlyidentifyvisualimagesinallkindsofdifficultconditions.Youlookatapictureofanorangeonaredplateandyoucantellinstantlythatit’snotasunset,butuntilrecentlythatwasreally,reallyhardforacomputer.BecauseofacombinationofMoore’sLawandsomebreakthroughs,computershavegottenmuchbetteratpatternrecognitioninthelastcoupleofyears.
Foracomputer,recognizingsomethingstartswithaprocesswheresomebasicattributesofanimageareextracted,suchastheshapeofboundariesbetweenclustersofpixels,orthedominantcolorofapatchofanimage.Thesearecalledfeaturesinmachinelearning.Byexamininglotsandlotsofexamplesoffeaturesinanimage,amachinelearningsystembuildsastatisticalmodelofwhatthatclusterrepresents.
Basicformsofthiskindofimagerecognitionhasbeenusedindustriallyfordecades.Mostoftheorangesthatcomefromthecentralvalleyarescanned360timestoseparateoneswithblemishesfromoneswithout.LegohasacompletelyautomatedfactorythatinjectionmoldsamillionLegobricksanhour,examineseverysinglepiece,automaticallysorts,bagsandboxesthem,allusingcomputervision.That’srelativelyold.
Imagesfrom:Region-basedConvolutionalNetworksforAccurateObjectDetectionandSemanticSegmentation,R.Girshick,J.Donahue,T.Darrell,J.Malik,IEEETransactionsonPatternAnalysisandMachineIntelligence
Real-TimeImageandVideoProcessing:FromResearchtoRealitybyKehtarnavaz andGemadia
23
Page 25
What’snewisaclassofsystemsthatunderstandthecontentofimages.Theydon’tjustlookatfeatures,butclustersoffeatures,andclustersofclustersoffeatures,andtheycannowidentifyanorangefromthesettingsun,orapersonfromanairplane,orapolarbearfromadalmatian.
ThisiswhyFacebookasksyoutosaywhoisinanimage.It’snotjustforyou,it’sfortheirfacerecognizer.
Nowhere’stheinterestingpart:we’rebuilttoidentifypatternsinvisualphenomena,butwe’reprettybadatidentifyingtheminotherkindsofsituations.Forexample,ifyou’veevertriedtounderstandsomeone’sfoodsensitivities,it’sreallyhardtoextractwhatthatpersonisreactingto,evenifyoukeepverycarefultrackofwhatthey’veeaten.We’rejustnotbuiltforit.Itwasneverevolutionarilysufficientlyimportant,sowedidn’tevolveanorganforit.
Computers,ontheotherhand,don’tcare,andnowthatwe’vefoundreallygoodwaystofindpatternsinvisualimages,thesesametechniquescanfindpatternsinanything.
Insteadofamatrixofpixels,whatifyouhadamatrixofmedicalprescriptions,witheachrowasthehistoryofoneperson’sprescriptionsfromthefirsttimethatpersonwenttothedoctorforaproblem,throughwhentheywereprescribedcertainthings,towhentheygotbetter,ortheydidn’t.Thesamekindofsystemcouldlearnthetypicalpatternforprescribing,say,awheelchair.Itwouldessentiallyseethegeneralshapeofthesequencefortheprescriptionofachairovertimeandacrossmanypeople.
Thenifyousawawheelchairbeingprescribedthatwasoutsideofthetypicalpattern,youcouldidentifyit.That’scalledanomalydetection.That’sinfactexactlyhowwebuiltasystemtoidentifyMedicarefraudforthestateofCalifornia.Peopleareterribleatthatstuff,butcomputersaregreat.
24
Page 26
Whenoneofthedimensionsistimeandanotheristheoutcomeofaseriesofactionsyoucanmakeapatternrecognizerthatassociatesasequenceofactionswithasetofstatisticalprobabilitiesforpossibleoutcomesbasedondatacollectedacrossawidevarietyofsimilarsituations.Inotherwords,becausepeopleandmachinesbehaveinfairlyconsistentways,thesemachinelearningsystemscanincreasinglypredictthefutureandattempttoadaptthecurrentsituationtocreateamoredesirableoutcome.
25
Page 27
Theinterestingthingisthatthisnotjusttheory.
PredictionandresponseisattheheartofthevaluepropositionmanyofthemostcompellingIoT services,startingwiththeNest.TheNestsaysthatitknowsyou.Howdoesitknowyou?Itpredictswhatyou’regoingtowantbasedonyourpastbehavior.
26
Page 28
Amazon’sEchospeaker saysit’scontinuallylearning.Howisthat?Predictivemachinelearningbasedonyouractionsandyourwords.
27
Page 29
The Birdi smartsmokealarmsaysitwilllearnovertime,whichisagainthesamething.
28
Page 30
Jaguar, learning…ANDintelligent.
29
Page 31
TheEdyn plantwateringsystemadapts toeverychange.Whatisthatadaptation?Predictivemachinelearning.
30
Page 32
Canary,ahomesecurity service.
31
Page 33
Cocoon,anotherhomesecuritysystem knows.Howdoesitknow?Machinelearning.
32
Page 34
Here’sfoobot,anairqualityservice.
[Ialsolikehowoneof itsimplicitservicepromisesistoidentifywhenyourkidsaresmokingpot.]
33
Page 35
Silk’sSenseadapts
34
Page 36
Mistbox sprayswaterintoyourairconditionertoreduceyourenergybill.You’dthinkthat’saprettysimpleprocess,butno,it’salwayslearning.
35
Page 37
Anumberofcompaniesaremakingchipsthatmakemachinelearningmuchcheaperandmorepower-efficient,whichmeansthatit’sgoingtobeveryeasytoinstallitineverydevice,fromstreetlightstomedicalequipmenttotoys.It’snotjustlikely,it’sinevitable.Here’sonethatwasannouncedacoupleofweeksago.
36
Page 38
Here’saKickstarter for an“AIButler”thatpostedearlierthismonth.Whatdoesitdo?Idon’tknow,butitlearns.
37
Page 39
Theidealscenariothesethingspaintisprettyseductive.Imagineaworldofespressomachinesthatstartbrewingasyou’rethinkingit’sagoodtimeforcoffee;officelightsthatdimwhenit’ssunnytosaveenergy,andmacandcheesethatneverrunsout.Theproblemisthatalthoughthevaluepropositionisofabetteruserexperience,it’sunspecificinthedetails.Previousmachinelearningsystemswereusedinareassuchaspredictivemaintenance andfinance.Theyweremadebyandforspecialists.Nowthatthesesystemsareforgeneralconsumers,wehavesomesignificantquestions.Howexactlyhowwillourexperienceoftheworld,ourabilitytouseallthecollecteddata,becomemoreefficientandmorepleasurable?
We’restillearlyinourunderstandingofpredictivedevices,sorightnowtheproblemsareworsethansolutions.IwanttostartbyarticulatingtheissuesI’veobservedinourwork.
38
Page 40
We’veneverhadmechanicalthingsthatmakesignificantdecisionsontheirown.Asdevicesadapttheirbehavior,howwilltheycommunicatethatthey’redoingso?Dowestickasignonthemthatsays“adapting”,likethelightonavideocamerasays“recording”?Shouldmychairvibratewhenadjustingtomyposture?Howwillusers,orjustpassers-by,knowwhichthingsadapt?Icouldendupsittinguncomfortableforalongtime,waitingformychairtochange,beforerealizingitdoesn’tadaptonitsown.Howshouldsmartdevicessettheexpectationthattheymaybehavedifferentlyinwhatappearstobeidenticalcircumstances?
How doweknowHOWintelligentthesedevicesare?Peoplealreadyoftenprojectmoresmartsondevicesthanthosedevicesactuallyhave,soacoupleofaccuratepredictionsmayimplyamuchbettermodelthanactuallyexists.Howdoweknowwe’renotjust homesteadingtheuncannyvalleyhere?
ChairbyRaffaello D'Andrea,MattDonovanandMaxDean.
39
Page 41
Theironyinpredictivesystemsisthat they’reprettyunpredictable,atleastatfirst.Whenmachinelearningsystemsarenew,they’reofteninaccurate,whichisnotwhatweexpectfromourdigitaldevices.60%-70%accuracyistypicalforafirstpass,buteven90%accuracyisn’tenoughforapredictivesystemtofeelright,sinceifit’smakingdecisionsallthetime,it’sgoingtobemakingmistakesallthetime,too.It’sfineifyourhouseisacoupleofdegreescoolerthanyou’dlike,butwhatifyourwheelchairrefusestogotoadrinkingfountainnexttoadoorbecauseit’sbeentrainedondoorsanditcan’ttellthat’snotwhatyoumeaninthisoneinstance?Forallthetimesasystemgetsitright,it’sonthemistakesthatwejudgeitandacouplesuchinstancescanshatterpeople’sconfidence.Anxietyisakindofcognitiveload,andalittledoubtaboutwhetherasupposedlysmartsystemisgoingtodotherightthingisenoughtoturnaUXthat’srightmostofthetimeintoonethat’smoretroublethanit’sworth.Whenthathappens,it’s lostyou.
Photo CCBY2.0photo2011PopCultureGeektakenbyDougKline:https://www.flickr.com/photos/popculturegeek/6300931073/
40
Page 42
Thelastissuecomesasaresultoftheprevioustwo:control.Howcanwemaintainsomelevelofcontroloverthese devices,whentheirbehaviorisbydefinitionstatisticalandunpredictable?
Ontheonehandyoucanmangleyourdevice’spredictivebehaviorbygivingittoomuchdata.WhenIvisitedNestoncetheytoldmethatnoneoftheNestsintheirofficeworkedwellbecausethey’reconstantlyfiddlingwiththem.Inmachinelearningthisiscalledovertraining.Theotherhand,ifIhavenodirectwaytocontrolitotherthanthroughmyownbehavior,howdoIadjustit?AmazonandNetflix’srecommendationsystems,whicharemachinelearningsystemsforpredictingwhatyoumaylike,giveyousomecontextaboutwhytheyrecommendedsomething,butwhatdoIdowhenmyonlyinterfaceisagardenhose?
41
Page 43
Asinterestingastheseissuesare,Ithinkthat, moreimportantly,whattheyrepresentisthatwe’reentering intoanewrelationshipwithourdeviceecosystem,aseachangeinourrelationshiptothebuiltworld.
42
Page 44
Thinkofasewingmachine.It’sverycomplex,butitstillonlyactsinresponsetous.
43
Page 45
Computersacting autonomously erodethissimpletool/userrelationship.
Atthedawnofcomputinginthelate1940scyberneticistslikeNorbertWienerphilosophizedabouttheincreasinglycomplexrelationshipbetweenpeopleandcomputers,andhowitwasfundamentallydifferentthanthewayweinteractwithotherkindsofmachines.Developersworkinginsupervisorycontrolofmanufacturingmachinesandroboticshavehadtodealwiththesequestionspragmaticallyforabout30years,butthankstotheInternetofThings,thisisnowaproblemthateveryonewillhavetograpplewithgoingforward.
Here’sadiagrambythegreatsTomSheridanandBillVerplank from1978,inwhichtheyillustratefourwaysthatsemi-autonomouscomputersandhumanscanworktogethertosolveaproblem.
44
Page 46
By2000Sheridanexpandedtheseideastocreatethisframework,todefineaspectrumofresponsibilitybetweenpeopleandcomputers.Itrangesfromhumansdoingallthework(thisisyouwritinganessay)tocomputersdoingalltheworkcompletelyautonomously(thisisyourcar ’sfuelinjectioncontroller).Ofcoursethegoalistogetasystemtolevel9or10.That’sthemaximumreductionincognitiveload.However,forasystemtoqualifyforthat,ithastobeverystable,itseffectsneedtobehighlypredictableand,equallyimportantly,it’sroleneedstobeadequatelyembeddedinsociety.ItneedstobeOKforacomputertotakeonthatlevelofresponsibility.Attheairportwetrustthemonorailcomputerstoworkwithouthumanintervention,butwedon’ttrusttheplaneautopilottodothat,eventhough-–asIunderstandit—planescanbasicallyflythemselvesthesedays.
PredictiveIoT devicesgenerallyfallbetween5and7onthisscalerightnow.Theproblemisthatthisistheexactrangewhereyou’remaximizingsomeone’scognitiveload,butnotnecessarilydoingalltheworkforthem,sotheresultoftheautomationhadbetterbeworthit.Thisfundamentallyundermineswhatweexpectfromourtools,andwhenthattoolistryingtoanticipatewhatwe’retryingtodo,itfundamentallychangesourworkingrelationshipwithit.
45
Page 47
DannyHillis oftheLongNowtalksabouthowwehavegonepasttheEnlightenmentideawherewethoughtthatwecouldunderstandandcontroleverything,andbuilttoolsthatreflectedthatview.Inhisperspective,wearenolongerincontrolasmuchasweareentangledwiththem.
AnneGalloway,aNewZealandresearcherwholooksattheintersectionofanimalsanddigitaltechnology,callsittheendofhumanexceptionalism.Otherswouldsayit’sjustthePostmoderncondition,therecognitionthatthecomplexityoftheworldisbeyondourabilitytocontrol,andwehavetolearntocoaxandcoexist,ratherthancommandandcontrol.
46
Page 48
Becausesoonerthanweexpect, we’llbelivingwithhundredsofdevicesandservicestryingtomodelusandpredictwhatwillbegoodforus,andmostofthemwillrequireourattention.Theywillwantustoverifythings,touploadthings,toconfirmthings.Theywillwantustovalidatetheirexistence.Andtheywillbewrongalot.Ifyouhave100devicesandeachdevice is99%accurate—andmostpredictivealgorithmsrarelyachievethatlevelofaccuracy,atleastnotatfirst—thenoneisalwayswrong.
So howdoweengagewiththisworld?Howdoweapproachwranglingallthesethinkingtools?
47
Page 49
Youcanthinkaboutworkingsurroundedbyabunchofapprenticeassistants,asinamiddleageguild.
48
Page 50
…oryoucantakeananimistviewofassumingeverythingintheworldhasaconsciousness.PhilVanAllenofArtCenterhasrecentlystartedadvocating anapproachlikethis.Well,maybenotlikeTHIS.
ImagefromMiyazaki’sPrincessMononoke.
PhilVanAllen:https://medium.com/@philvanallen/rethink-ixd-e489b843bfb6#.6jszlfw9p
49
Page 51
I’dliketoexplorefarmingasametaphor,andnotbecauseofthesuperficialironyofusingpre-Enlightenmenttechnologytotalkaboutapost-Enlightenmentproblem.
Ireallywanttocreateausefulwayofthinkingaboutthechallengeofsmarttoolssowecandesignabetterrelationshipwiththemfromthebeginning.
50
Page 52
Farmingis oneofouroldesttechnologies,oneofthemostadvanced,andoneofthemostbrutalontheland,peopleandanimalsinvolved.Butitgotushere.
Also,anadmission:I’macitykid,myfamilyhasbeenlivingin citiesgoingbackmanygenerations.Ihavenotraisedsomuchasasingleedibleplantorownedapet,thoughIdohavechildren,butIdon’tthinkit’sthesame.Butthelongnowaskedmetodosomethingbrandnewandforageneralaudience,andthisiswhereIendedup,soifthistalkhasn’tgoneofftherailsforyouyet,it’llprobablygoofftherailsnow.
51
Page 53
For mefarmingisausefulmetaphorabouthowtosimultaneouslymanipulate thestateofmanyautonomous,independent,similarthings,foryourgain.Afarmerdoesn’traiseanearofcorn,sheraisesafieldofcorn,andsheisnotincontroloftheircropsasmuchassheisinsymbiosiswiththem.
Shereducesthecomplexityoffarmingbyplantingmanycopiesofthesameplant,anddividingherlandintoregionsforeachkindofplant.Rightnowslikeeachplantistotallydifferentandrequiresatotallydifferenttechniquetoworkwithit.
Sheselectscropsthatthriveinaspecificsetofconditionsandwhichcansynergisticallyusethesamerawmaterialtomaximizethevalueofthatmaterial.Whatifhadmultiplealgorithmsusingtheinformationfromthesamesensors—sayallthecamerasandtemperaturesensorsinyourenvironment—thenfusingtheirresults?
52
Page 54
Afarmerusesspecializedtoolstoworkonmanyplantsatthesametime,whetherit’saplow,aharvesterorascarecrow.That’swhyshechoosesmanyofthesamething.Inthealgorithmanalogy,howcanwegrouplargenumbersofalgorithmsandworkonthemallatonce?
Sheexpectspests.Rightnoweveryoneisshockedwhentheirsmartfridgestartspostingspambecauseit’sbeenhacked.That’skindoflikeafungusinfection,andfarmershavetoolsforthatandtrytomaintaingoodpracticestominimizeit,butwhenithappens,nooneissurprised.
Shedoesn’texpecttoextractthevaluefromitimmediately—thatmaytakemonthsoryears—yetsheknowsshewillhavetomaintainitthatwholetimeregardless.Rightnowweexpectourdigitalproductstoworkimmediatelyorwethinkthey’renotworthwhileordefectiveiftheydon’t.Whatifwedesignedthingssothattheywouldonlybeusefulafterwehadlivedwiththemforalongtime,butthenthey’dbeREALLYuseful?
53
Page 55
Anotheraside:machinelearningalgorithmsarepatternrecognizers,sotheyneedtoknowwhichpatternsareimportant.Wheneveryoumarkemailasspamusingyouremailprogram,youaredoingwhat’scalledtrainingthealgorithmtounderstandwhatyouconsiderspam.
54
Page 56
Similarlywhenyoumakeachoiceusingvirtuallyanyadigitaldeviceorservice,you’retraininganalgorithm.Facebookasksyoutolabelpeopleinyourpicturestotrainitsalgorithmstoassociateasetoffacialfeatureswiththepersonyoulabeled.
Whathappenswhenyoutrainasingleanimal?Whatareyourmechanismsofcontrol?Whatareyourexpectations?
Well,youexpectthatitwillrequiretimeanditwillrequireacombinationofbothpositiveandnegativereinforcement.Then,youexpectthatitwillregularlymisbehaveandyouhavetoreinforcewhatyouteachit.Conversely,youcanexpectthatitwillprobablylearnabitfromotheranimalswithoutyouhavingtotelliteverythinganditsbehaviorwillsurpriseyouingoodwaysinadditiontobadways.
Imagesource:http://countingsheep.info/permalamb.html (AnneGalloway’sCountingSheepproject)
55
Page 57
Butwhathappenswhenafarmerhasalotofanimalstocontrol?Shecan’ttrainallofthemindividually,sooverthelast10000yearsshe’sdevelopedsometoolsformanagingthem.
First,sheselects animalsthatworkwellingroups.Ouralgorithmsarecurrentlybuiltoneatatimeandtheexpectationisthatourinteractionwiththemwillbeindividual.Thatdoesn’tscale.Weneedalgorithmsthatareexperiencedwelltogether,orelsewe’renotherdingsheep,we’reherdingcats.Next,shehasacrook.Whenyouneedtoassertcontrol,youneedaclearwaytodothatwhichworksonawidevarietyofanimalsandweneedconsistentwaystoassetimmediatecontroloverawidevarietyofsmartdevices.Shehas adog, whichisasmarterentitythatalso needstobetrained,butoncetrainedcanbeusedtoautonomouslycontrolmultipleotherindependententitiesitself.Shecanhandofftheworktoanassistant.Infarmingawholeclassofpeoplewhocantakeresponsibilityforallofthethingsandwhocanworktogether.Responsibilitycanbedelegated.AsTomCoatesofThington pointsout,mostIoT systemsarenotbuiltformanypeopletocontrolthemsimultaneously,eventhoughtheireffectsareoftenexperiencedinsharedenvironments.
56
Page 58
Todaywedon’thaveanInternetofThings,wehavemanyAOLsofthings.They’vebeenintentionallymademutuallyincompatibleandalthoughsomemaybecuteontheirown,whenyouhavealotofthem,andtheyhavetobedealtwithindividually,it’sabigproblem.
57
Page 59
Ithinkin 1000years,maybe100years fromnow,thisentirediscussionwillseemabsurd,likearguingaboutwhetherironisagoodthingorabadthing.We’llseeitasjustthewaytheworldis.Ourbodiesaregoingtobesemi-autonomouscomponentsthatwehavesomecontrolover,inanecosystemthatcombinesotherbiologicalanddigitalsemi-autonomouscomponents.Everythingisgoingtohavesomecontroloverandbecontrolledbyotherthings.
Someofthemaresmarterthanothers,somearemoreautonomousthanothers,someareevensmarterthanweareincertainways,somehavepositivesymbioticrelationships,someareparasites.Theboundariesbetweenmindsandbodies,betweennaturalandartificial,andbetweenhumanandnon-humanwillhavebeeneroded.Ourworldwillhavereconfigureditselfaroundassumptionsthateverythingismuchpermeableandmuchlessclearlydelineatedthanwehadfooledourselvesintobelieving.Wearenotasgods.Weare,andalwayshavebeen,animalsinanecosystem.
Anditwon’tallbegood.Therewillprobablybeterriblethingsthathappentopeople’sbodies,mindsandsocieties.Theremayalso,hopefully,begoodthings.
Image:CamillePissarro,“Shepherdesses,”1887
58
Page 60
Thisisthe lookingglassthatwe’vemade,andit’stimeforustostepthrough,andexplorethefield beyond,becausewehavenochoicebuttoengagewithit,tomakeitbewhatwewantittobe,whatweneedittobe,becauseitisnotandroidswhowilldreamingofelectricsheep,itwillbeus.
59