JReliableIntellEnviron(2015)1:33–46 DOI10.1007/s40860-015-0002-6 ORIGINAL ARTICLE On resilient behaviors in computational systems and environments VincenzoDeFlorio1 Received:17December2014/Accepted:31March2015/Publishedonline:1May2015 ©SpringerInternationalPublishingSwitzerland2015 Abstract Thepresentarticleintroducesareferenceframe- peculiar domain-specific meanings [17]. To exacerbate the work for discussing resilience of computational systems. problem,eveninthecontextofthesamedisciplineoftenno Ratherthanapropertythatmayormaynotbeexhibitedbya consensushasbeensofarreachedastowhatarethepeculiar system,resilienceisinterpretedhereastheemergingresultof aspectsofresilienceandwhatmakesitdifferentfrom,e.g., adynamicprocess.Saidprocessrepresentsthedynamicinter- elasticity,dependability,orantifragility. playbetweenthebehaviorsexercisedbyasystemandthoseof Thepresentarticlecontributestowardsasolutiontothis theenvironmentitissettooperatein.Asaresultofthisinter- problem in several ways. First, in Sect. 2, we introduce pretation,coherentdefinitionsofseveralaspectsofresilience resilience, compare various of its domain-specific defini- can be derived and proposed, including elasticity, change tions,andderiveanumberofancillaryconceptsandworking tolerance,andantifragility.Definitionsarealsoprovidedfor assumptions. This allows resilience to be interpreted in measuresoftheriskofunresilienceaswellasfortheoptimal Sect.3asthepropertyemergingfromtheinteractionofthe matchofagivenresilientdesignwithrespecttothecurrent behaviors exercised by a system and the environment it is environmentalconditions.Finally,aresiliencestrategybased settooperatein.Theoutcomeofsaidinteractiondependson onourmodelisexemplifiedthroughasimplescenario. bothintrinsicandextrinsicfactors:the“traits”ofthesystem together with its endowment—the system’s peculiar char- Keywords Resilience · Elasticity · Computational acteristics as well as its current state and requirements. At antifragility·Behavior·Autonomiccomputing stake is the identity of the system, which we identify here with compliance to system specifications, including func- tionalandnon-functionalsystemrequirements. 1 Introduction Theresilienceinteractionismodeled byconsidering the behaviorsproducedbythesystemanditsenvironment.This Resilience is one of those “general systems attributes” that provides us with a unifying framework within which it is appeartoplayacentralroleinseveraldisciplines.Examples possibletoexpresscoherentdefinitionsofconceptssuchas includeecology,business,psychology,industrialsafety[32], elasticity, entelechism (change tolerance), and antifragility. microeconomics,computernetworks,security,management Bothsystemandenvironmentarefurthermodeledinterms science,cybernetics,controltheory,aswellascrisisanddis- ofthe“resilienceorgans”managingthefivemajorservices astermanagementandrecovery[17,22,35,42,50].Although ancillary to resilience [12,13]: perception, apperception, commontraitsareretained,ineachdisciplineresiliencetakes planning, executive, and knowledge management organs— correspondingtothefivemodulesofautonomicsystems[30]. B Afterthis,inSect.4,weintroducemeasuresoftheoptimality VincenzoDeFlorio vincenzo.defl[email protected]; ofagivenresilientdesignwithrespecttothecurrentenviron- vincenzo.defl[email protected] mental conditions: system supply and system–environment fit.OnesuchstrategyisdetailedandexemplifiedinSect.5 1 MOSAICResearchGroup,iMindsResearchInstitute, through an ambient intelligence scenario. Finally, major UniversityofAntwerp,Middelheimlaan1,2020Antwerp, Belgium resultsarerecalledandfinalconclusionsaredrawninSect.6. 123 34 JReliableIntellEnviron(2015)1:33–46 2 Basicconcepts its original shape upon unloading” [38]. Beyond the mentioned threshold, deformation is irreversible:“some The term “resilience” comes from Latin resil¯ıre, “to spring residual strain will persist even after unloading” (ibid.) back, start back, rebound, recoil, retreat”, and is often Here resilience is a property rather than a process, the intended and defined as the ability to cope with or recover key difference being the type of behavior exercised by fromchange.AsmentionedinSect.1,thisgeneraldefinition the“systems”.Systemidentityisinthiscaserepresented hasbeenspecializedindifferentdomains,ineachofwhich bytheshapeorthecharacteristicsofthematerial. ithastakendomain-specifictraits: – Incivilengineering,resilienceisaconstruction’sability to absorb or avoid damage in the face of a natural or – Inecology,resilienceoftenreferstoanecosystem’sability man-induced abnormal condition [29] such as flooding, torespondtoandrecoverfromanecologicalthreat[27]. hurricanes,orfirepower.Theconsiderationsdoneinthe “Recovering”meansheretheabilitytoreturntoasteady caseofmaterialscienceapplyalsotothiscase. statecharacterizingtheecosystembeforethemanifesta- – Finally,incomputerscience,resiliencehasbeendefined, tionofthethreat.Thementioned“steadystate”represents e.g., as the ability to sustain dependability when facing thepeculiarcharacteristicsoftheresilientecosystem—its changes[32].Thistranslatesintotheabilitytoavoidfail- identity. ureandatthesametimetheabilitytosustainthedelivery – Incomplexsocioecologicalsystems,resilienceistheabil- oftrustworthyservices.Systemidentityisinthiscaseas ity to absorb stress and maintain function in the face inthefollowingdefinition. of climate change [23]. “Absorbing stress” clearly cor- responds to the “recovering” ability found in ecology, Definition1 (Systemidentity)Intheframeworkofartificial, thoughinthiscontexttheidentityofthesystemliesinits computer-basedsystems,systemidentityisdefinedasasys- functionratherthaninitsstate.Anadditionalandpeculiar tem’scompliancetoitssystemspecificationsandinparticular aspectofresilientsystemsinthisdomainisgivenbytheir to its functional and non-functional quality of service and abilitytoimprovesystematicallytheirsustainability. qualityofexperiencerequirements. – Inorganizationscience,(organizational)resilienceis“the capacity to anticipate disruptions, adapt to events, and As already remarked, with the change of the reference createlastingvalue”[3].Here,theaccentsareonproac- domaintheabovenotionsofresilienceareappliedtoaspec- tiveness and adaptation rather than on “springing back” trum of entities ranging from simple, passive-behaviored, toapaststateorfunction.Intuitively,onemaydeductthat individualobjectstocomplex,teleological,collectiveadap- theformerclassofbehaviorsismoreadvancedthanthe tive systems. By making use of the behavioral approach latterone.Onemayalsoobservehowinthiscasethedefi- introduced by Wiener et al. in [37], and briefly recalled in nitionbringstotheforegroundafundamentalcomponent Sect. 3.1, in what follows three major classes in this spec- ofsystemidentity,namelytheabilitytocreatevalue. trumareidentified. – In social science and human ecology, resilience is “the ability of groups or communities to cope with external 2.1 Elasticobjectsandsystems stresses and disturbances as a result of social, political andenvironmentalchange”[1].Here,therecoverystrat- Resilience shall be referred to as elasticity when the sys- egy of resilience is not made explicit. System identity tem under consideration is only capable of simple types implicitlyreferstotheidentityofgroupsandcommuni- of behaviors: passive behavior and active, purposeful, non- tiesandrangesfromsocioculturalaspectsuptotheability teleological behaviors. In the former case, the system shall tosurvive. bereferredtoasanobject. – Inpsychology,“resilienceistheprocessofadaptingwell The considerations made above with reference to, e.g., in the face of adversity, trauma, tragedy, threats or sig- material science, apply also in this case. In particular, for nificant sources of stress [...] It means ‘bouncing back’ bothobjectsandservo-mechanismsresilience(elasticity)is fromdifficultexperiences”[4].Animportantaspecthere representedasanintrinsicproperty:atrait. istheidentificationofresilienceasaprocess:“Resilience Elasticsystemsabletoexerciseactivebehaviorsarewhat should be considered a process, rather than a trait to be Boulding refers to in [7] as “servo-mechanisms”—systems had” [39]; see also [24,34]. “System” identity is in the whoseactionispredefinedandisnotmodifiedbytheinter- case of psychology the collection of beliefs about one- actionwithothersystems.Infact,servo-mechanismsdonot self. “interact with other systems outside of themselves” [26]— – In material science, resilience is a material’s property namely,theyarenotopensystems. to “stay unaffected by an applied stress” up to some Elastic systems and objects operate under the assump- threshold, called “point of yielding”, and to “return to tionthattheirenvironmentsarenotgoingtoexercisestress 123 JReliableIntellEnviron(2015)1:33–46 35 beyondtheirpointofyielding.QuotingN.N.Taleb,theyare Overshooting reminds of the condition of the shell-snail systemthat“donotcaretoomuch”abouttheirenvironments. that,“feelingalwaysindangerofbirds,livesconstantlyunder Another way to characterize elastic objects and systems itsshield”[6,p.147]—thuscarryingitsweightatalltimes isbyobservingthattheyhaveapredefinedandstaticpoint regardlessoftheactualpresenceorabsenceofbirds. of yielding. This introduces two syndromes, which we call “elasticundershooting”and“elasticovershooting.” 2.1.3 Observations 2.1.1 Elasticundershooting Elasticundershootingandovershootingmaybebetterunder- stood when considering a well-known result by Shan- Elastic objects or systems are characterized at design time non [45]: given an unreliable communication channel, it is by a static yielding point beyond which they permanently always possible to transfer information reliably through it lose their identity—for instance, they deformate; or break providedthatasufficientamountofinformationredundancy down;orfail;orbecomeuntrustworthy.Theyieldingpointis is foreseen. By means of the above-introduced terminol- thereforearesiliencethreshold.Whateverthecharacteristics ogy,Shannon’sresultmaybeformulatedasfollows:forany of an elastic object or system, there is always a non-zero communicationchannelwhoseobservedunreliabilityisy(t) probabilitythattheyieldingpointwillbeovercome. throughoutagiventimeintervalT,itispossibletodefinean elasticcommunicationprotocolwithayieldingpoint Definition2 (Elasticundershooting)Whenattimetanelas- tic object or system with yielding point Y is insufficiently Y > y(t) ∀t ∈T. resilientwithrespecttoanexperiencedconditionforwhich ayieldingpoint y(t)wouldberequired,weshallcallelastic Ideally, the choice of Y should be such that Y represents undershooting(orsimplyundershooting,whenthismaybe thesupremumofalltheunreliabilitysamples y(t)observed done without introducing ambiguity) the dynamic quantity y(t)−Y. duringT.Inthisidealcase,noundershootingisexperienced, althoughthesystemexhibitsacumulativeovershootingequal Undershootingisinfactasinawell-knownfairytale[28]: to onemaymaketheirhouseofstraw,ofwood,orevenofbricks; (cid:2) althoughmoreandmorerobust,eachhousewill“just”shift Y −y(t)dt. the yielding point farther away; but that is all: there is no T guaranteethat,soonerorlater,somethingstrongerwillshow upandblowthathousedown,whateverthematerialitismade In more concrete situations in which unreliability drifting of.Developmentandoperatingcosts,ontheotherhand,will isunbound,undershootingwouldoccureachtimethevalue growupproportionallytothechosenyieldingpoint—which chosenforY wouldbelessthantheobservedunreliabilityof bringsustothesecondsyndrome. thechannel. 2.1.2 Elasticovershooting 2.2 Systementelechies Thechoiceoftheyieldingpointrepresentstheabilitytocope In Sect. 2, we have concisely reviewed a number of defi- withaworst-casescenarioregardlessofhowfrequentlysaid nitions of resilience emerged in the framework of diverse condition will actually manifest itself. Unless the environ- disciplinesanddomains.Severalofsuchdefinitionsexplic- mentalconditionsaredeterministicandimmutable,therewill itly require a resilient system to enact complex forms of always be a non-zero probability that the yielding point is behaviors: adapt reactively (see, e.g., in ecology and psy- morepessimisticthanwhattheexperiencedconditionwould chology) and adapt proactively (see, e.g., in organizational require.Inotherwords,anelasticsystemispreparedforthe science). Such behaviors correspond respectively to simple worst;butalsoitcostsandexpendsresourcesasiftheworst teleologicalbehaviorsandextrapolativeteleologicalbehav- wasactuallythereallthetime. iors(asdefinedin[37]andrecalledinSect.3.1):behaviors that are driven respectively by the current state and by the Definition3 (Elasticovershooting)Whenattimetanelastic hypothesized future state of an intended objective. Obvi- objectorsystemwithyieldingpointYisresilientwithrespect ously, in this case, resilience cannot be regarded as a trait toanexperiencedconditionforwhichayieldingpoint y(t) or attribute; rather, it is the emerging result of a process. wouldsuffice,weshallcallelasticovershooting(orsimply Resilient systems are in motion to actively pursue the per- overshooting, when this may be done without introducing sistence of their system identity. The two just mentioned ambiguity)thequantityY −y(t). aspectscorrespondtothetranslationthatJoeSachsprovides 123 36 JReliableIntellEnviron(2015)1:33–46 of the Aristotelian concept of entelechy1: “being-at-work- 2.3 Antifragilesystems staying-itself”[40,41].Anentelechyisasystemthatisable to persist and sustain its completeness through a resilient InthelightofthediscussioninSects.2.1and2.2,onemay process.“Completeness”hereistobeintendedasthecharac- observethatmostofthereporteddefinitionsofresiliencecor- teristicsthatmakeofasystemwhatitis:its“definition”—or, respondtoeitherelasticobjects/systemsortoentelechies. inotherwords,itssystemidentity.Becauseofthis,weshall Anexceptionmaybefoundintheclassofcomplexsocioe- refertothesystemsinthisresilienceclassastoentelechies. cologicalsystems.Therewehavesystemsthat“areatwork Theverynatureofentelechiesrequiresthemtobeableto to stay themselves” (thus, they are entelechies), though are “interactwithothersystemsoutsideofthemselves”,namely endowedwithanadditionalfeature:theability“toimprove to be open systems [26]. Such systems do not “want tran- systematically their sustainability”. Wiener et al. did not quility” nor expect their environments to be stable or stay explicitlyconsiderbehaviorsincludingthatability[37].The thesame.Onthecontrary,theyassumeconditionswillvary, mostcloselyrelatedoftheirbehavioralclasses—onecould and adjust their function to the observed conditions or to sayitsgenusproximum[9]—isgivenbyteleologicalbehav- speculatedfutureconditionsoftheirenvironments. iors. Another way to distinguish entelechies from elastic As discussed in more detail in Sect. 3.1, teleological objects and systems is by observing that entelechies are systems are those characterized by a feedback loop: their characterized by dynamic and adaptive points of yielding. behavior Undershootingsandovershootingsarestillpossible,though “iscontrolledbythemarginoferroratwhichthe[sys- withaslightlydifferentformulation: tem]standsatagiventimewithreferencetoarelatively Definition4 (Entelechial undershooting) When, at time specificgoal”[37]. t, an entelechy with yielding point Y(t) is insufficiently Duetoitspurelybehavioralnature,theapproachfollowedin resilientwithrespecttoanexperiencedconditionforwhich the cited work does not cover organizational, architectural, a yielding point y(t) would be required, we shall call ent- and structural aspects. Because of this, no account is given elechial undershooting (or simply overshooting, when this onthemodificationsthatateleologicallybehavioredsystem may be done without introducing ambiguity) the dynamic wouldapplytoitselftoachieveitsgoal. quantity y(t)−Y(t). Atleastthefollowingfourcasesmayoccur: Definition5 (Entelechialovershooting)When,attimet,an entelechy with yielding point Y(t) is resilient with respect 1. Thefeedbackloopispurelyexogenous:thesystemaction toanexperiencedconditionforwhichayieldingpoint y(t) is simply steered towards the goal (in its current or wouldsuffice,weshallcallentelechialovershooting(orsim- hypothesizedfutureposition.) plyovershooting,whenthismaybedonewithoutintroducing 2. The feedback loop is both exogenous and endogenous. ambiguity)thequantityY(t)−y(t). Internalchangesonlyconcernsthe“knobs”,namelythe parametersofthesystem.Thiscasecorrespondstopara- An exemplary entelechy is given by an adaptive com- metricadaptation. municationprotocolforthereliablecommunicationoveran 3. The feedback loop is both exogenous and endogenous; unreliablechannelcharacterizedby y(t)unreliability.Such the internal changes adapt the structure of the system. protocolwouldcontinuously“beatwork”soastoestimate This corresponds to system reconfigurations (namely pastandcurrentvaluesof y(t)andextrapolatewiththema structuraladaptation).Adaptationsarephenotypicaland futurestatey(t(cid:4)).Oncethisspeculatedfuturevalueisknown, donotaffecttheidentityofthesystem.Furthermore,the theprotocolwould“stayitself”bychoosingayieldingpoint experienceleavesnotraceontheidentityofthesystem. Y(t(cid:4))ascloseaspossiblebutstillgreaterthan y(t(cid:4)). 4. The feedback loop is both exogenous and endogenous; Moreformally,thechoiceforY(t(cid:4))wouldbesuchthat theinternalchangesadaptanyofthefollowingaspects: thefunction;thestructure;thearchitecture;andtheorga- 0<Y(t(cid:4))−(cid:2)(y(t(cid:4)))<ε, (1) nization of the system. Changes are genotypical: they are persisted and modify permanently the nature of the where (cid:2)(y(t(cid:4))) represents a prediction of y(t(cid:4)) and ε > 0 system. expresses a safety margin to cover for inaccuracies in the prediction. Itisimportanttoremarkthat,whileincases1–3thesystem is“atworktostayitself”[40,41],incase4thesystemis“at 1 QuotingSachs,“Entelecheiameanscontinuinginastateofcomplete- worktogetbetter”.Atleastinthecaseofcomplexsocioeco- ness,orbeingatanendwhichisofsuchanaturethatitisonlypossible logicalsystems,teleologicalbehaviorsbelongtothisfourth tobetherebymeansofthecontinualexpenditureoftheeffortrequired tostaythere.”[40]. category:throughtheirexperience,thosesystemselaboratea 123 JReliableIntellEnviron(2015)1:33–46 37 feedbackthatisalsoendogenousandaffectsthegenotypical reached with a consumption of resources lower than in the ability“toimprovesystematicallytheirsustainability”.The originalalgorithm. feedbackthusaffectstheidentityofthesystem.Ratherthan At the same time, it is important to observe how the adapting,thesystemevolves. introduced interleaving would affect several peculiar char- In the case of complex socioecological systems, said acteristicsoftheprotocol—forinstance,itwouldintroduce evolution leads to an improvement in sustainability: those jitter (viz., a drifting in the periodicity of the messages). systems “enhance the level of congruence or fit between Repercussions on the validity of the specifications become themselves and their surroundings” [47]. This matches the thenpossible.Forinstance,iftheprotocolwereintendedfor concept introduced by Taleb in [48]: antifragile systems. ateleconferencingservice,theintroducedjitterwouldaffect Quoting from the cited reference, “Antifragility is beyond thequalityofexperienceoftheusersofthatservice.Embed- resilienceorrobustness.Theresilientresistsshocksandstays dingthesameprotocolinaserviceinsensibletoperiodicity thesame;theantifragilegetsbetter.” drifting(suchasafiletransferringservice)wouldnottrans- In what follows, we distinguish explicitly this class of lateinalossofsystemidentity. teleologicalbehaviorsandsystemsbyreferringtothemasto antifragilesystems,whichwedefineasfollows: 2.4 Afewobservations Definition6 (Antifragile system) We shall call a system “antifragile” if it is able to exercise teleological behaviors Asasummaryofthediscussioninthissection,wecanderive that evolve the system and its identity in such a way as to hereanumberofobservations: systematicallyimprovethefitwiththeirenvironment. Byconsideringthejustenunciateddefinition,wecanobserve 2.4.1 Resilienceisarelativefigure that Asobservedin[16],resilienceisadynamicpropertywhose – Antifragilesystemsarenotnecessarilymoreresilientthan emergenceisinfluencedbyatleastthefollowingtwofactors: entelechiesorelasticobjectsandsystems.Asitisthecase forthoseentities,alsoantifragilesystemsarecharacter- ized by a yielding point—a resilience threshold beyond 1. Theintrinsiccharacteristicsofthesystem:inparticular, which they would fail; break down; or become untrust- whetherthesystemiselastic,entelechial,orantifragile. worthy. 2. The extrinsic “level of congruence or fit between [the – Antifragilesystemsmutatetheirsystemidentity.Byrefer- system]and[its]surroundings”[47]. ring to Definition 1, this means that the behaviors of antifragilecomputer-based systemsmaydriftoutsideof Thefirstfactorisabsoluteinnatureandtellshow“evolved” whatprescribedintheirspecifications.Scenariossuchas the class of the system is. The second factor is a relative those that Stephen Hawking [25] and many others [33] one, and tells how the system’s behavior is able to match arewarningofbecomemoreconcretewhenconsidering theconditionscurrentlyexpressedbytheenvironment.This thisparticularcharacteristicofantifragilesystems. secondfactormakesofresiliencearelativefigure.Whatever – Antifragilesystemsmustpossesssomeformofawareness asystem’sstructure,organization,architecture,capabilities, oftheircurrentandpastsystem–environmentfit;inpar- and resources, that system is resilient only so long as its ticular,theymustbeabletocreateandmaintainamodel implementation matches the conditions currently exercised oftheriskoflosingoneormoreaspectsoftheirsystem bytheenvironment2. identity. Going back to the communication protocol presented in 2.4.2 Resilienceistheproductofaninterplaybetween Sect.2.2,anexemplaryantifragilesystemwouldbeaproto- asystemanditsenvironment colthat,inadditiontobeingabletoestimatequantity y(t(cid:4)), also learns how to mutate its own algorithm so as to profit AsacorollarytowhatmentionedinSect.2.4.1,weobserve from the characteristics of the environment. As an exam- thatresilienceisnotaproperty,butrathertheproductofaa ple,insteadofsending,say,Y redundantcopiesforeachof process.Suchprocesscorrespondstothedynamicinterplay the packets of its messages, the protocol could realize that betweentwoentities:asystemanditsenvironment. abetterstrategy(withrespecttothecurrentbehaviorofthe channel) would be that of interleaving the transmission of packets of different messages. This would result in a more 2 Possibly,thefirstScholartohavedistinguishedintrinsicandextrinsic efficientstrategysuchthatahigheryieldingpointwouldbe factorstowardstheemergenceofresiliencewasvonLeibniz[15]. 123 38 JReliableIntellEnviron(2015)1:33–46 2.4.3 Theenvironmentisasystem naturalevents”,namely“theexaminationoftheoutputofthe objectandoftherelationsofthisoutputtotheinput”3.The “Environment”isinterpretedhereandinwhatfollowssim- term“object”inthecitedpapercorrespondstothatof“sys- ply as another system; in particular, as a collective system tem”.Inthatrenownedtext,theauthorspurposely“omitthe (in other words, a “system-of-systems”) taking different specificstructureandtheintrinsicorganization”ofthesys- shapes, including for instance any combination of cyber- temsunderscrutinyandclassifythemexclusivelyonthebasis physical things; biological entities such as human beings; of the quality of the “change produced in the surroundings servo-mechanisms;andintelligentambientsabletoexercise by the object”, namely the system’s behavior. The authors complexteleologicalbehaviors. identifyinparticularfourmajorclassesofbehaviors4: 2.4.4 Resilienceisaninterplayofbehaviors βran:Randombehavior.Thisisanactiveformofbehav- ior that does not appear to serve a specific purpose or Resilience is one of the possible outcomes of an interplay reachaspecificstate.Asourceofelectromagneticinter- of behaviors. If and only if the interplay between the sys- ferenceexercisesrandombehavior. tem behaviors and the environmental behaviors is one that βpur: Purposeful, non-teleological behavior. This is preservesthesystemidentitythenthesystemwillbecalled behavior that serves a purpose and is directed towards resilient.AsdiscussedinSect.3.1,behaviorsmayrangefrom aspecificgoal.Inpurposefulbehavior,a“finalcondition therandombehaviorstypicalofelectromagneticsourcesup towardwhichthemovement[oftheobject]strives”can tothe“intelligent”,cyberneticbehaviorscharacterizing,e.g., beidentified.Servo-mechanismsprovideanexampleof humanbeingsandcomplexambientenvironments. purposefulbehavior. βrea:Reactive,teleological,non-evolutivebehavior.This 2.5 Preliminaryconclusions isbehaviorthat“involve[s]acontinuousfeedbackfrom the goal that modifies and guides the behaving object”. Amajorconclusionhere—andastartingpointforthetreatise Examplesofthisbehaviorincludephototropism,namely innextsection—isgivenbytheintuitivenotionthatevaluat- thetendencythatcanbeobserved,e.g.,incertainplants, ingresiliencemustbedonenotmerelyconsideringasystem’s to grow towards the light, and gravitropism, viz., the intrinsiccharacteristics;rather,itshouldbedonebyexpress- tendencyofplantrootstogrowdownwards.Asalready ing in some convenient form the dynamic fit between the mentioned(seeSect.2.2),reactivebehaviorsrequirethe systemanditsenvironment.Thismaybeobtained,e.g.,by system to be open [26] (i.e., able to continuously per- comparingtheresilienceclassofthesystemwiththedynam- ceive, communicate, and interact with external systems icallymutatingresilienceclassoftheenvironment.Another, andtheenvironment)andtoembodysomeformoffeed- moredetailedmethodcouldbebycomparingthebehaviors backloop.Classβreaisnon-evolutive,meaningthatthe ofsystemandenvironment.Yetanotherapproachcouldbeto experiencedchangedoesnotinfluencetheidentityofthe applythebehavioralcomparisonmethodtospecificorgans system(cf.Sect.2.3). ofthesystemanditsenvironment—forinstancethoseorgans βpro: Proactive, teleological, non-evolutive behavior. thatarelikelytoplayasignificantroleintheemergenceof Thisisbehaviordirectedtowardstheextrapolatedfuture resilienceoritsopposite. state of the goal. The authors in [37] classify proactive Inwhatfollows,wefocusonthelastmentionedapproach. behavior according to its “order”, namely the amount of context variables taken into account in the course of 3 Resilientbehaviors,organs,andmethods the extrapolation. As class βrea, so class βpro is non- evolutive. Toproceedwiththepresenttreatise,wefirstrecallinSect.3.1 themajorclassesofbehaviorsaccordingtotheclassicdiscus- ByconsideringtheargumentsinSect.2.3,afifthclasscan sionbyRosenblueth,Wiener,andBigelow[37].Afterthis,in beadded: Sect.3.2,fivemajorservicesthatplayakeyroletowardsthe emergence of resilience areidentified. Finally, inSect. 3.3, weusetheconceptsintroducedsofartoreformulatedefini- βant:teleologicalevolutivebehaviors.Thisisthebehav- ioremergingfromantifragilesystems(seeSect.2.3for tionsforelasticity,entelechism,andantifragility. moredetailedonthisclassofsystems). 3.1 Behavioralclassification 3 If not otherwise specified the quotes in the present section are As already mentioned, Rosenblueth, Wiener, and Bigelow from[37]. introducedin[37]theconceptofthe“behavioristicstudyof 4 Forthesakeofbrevitypassivebehaviorshallnotbediscussedhere. 123 JReliableIntellEnviron(2015)1:33–46 39 Eachoftheabovefiveclassesmayseetheirsystemsoper- π) and (2) β is based on a set of context figures that 2 ateinisolationorthroughsomeformofsocialinteraction.To extendsβ ’s. 1 differentiatethesetwocases,weaddthefollowingattribute: 2. (π(β )≤π(β ))∧(∃(F,G):β =β|F|∧β =β|G|∧ 1 2 1 1 2 2 F (cid:2)G). σ(b): True when b is a social behaviors. This attribute Thiscaseisequivalenttocase1,theonlydifferencebeing identifies behaviors based on the social interaction inthenotationothebehavior. with other systems deployed in the same environment. 3. (π(β1)=π(β2))∧(σ(β1)=false)∧(σ(β2)=true).In Examples of such behaviors include, among others, otherwords,β1 ≺β2alsowhenbothβ1andβ2belongto mutualistic,Commensalistic,parasitic,co-evolutive,and thesamebehavioralclass,thoughβ2isasocialbehavior co-opetitive behaviors [5,8]. For more information, the whileβ1isnot. readerisreferredtoBoulding’sdiscussioninhisclassic paper [7] and, for a concise survey of social behaviors, Foranytworesilientsystemsp andp ,respectively,char- 1 2 to[17]. acterized by behavioral classes β and β , if β ≺ β then 1 2 1 2 p is said to exhibit “systemically inferior” resilience with 1 The resilience classes introduced in Sect. 2 can now be respectto p . 2 characterizedintermsoftheabovebehavioralclasses:elas- It is important to observe that ≺ is about the intrinsic tic systems correspond to βpur; non-evolving entelechies resiliencecharacteristicsofthesystem(seeSect.2.4.1).Par- exerciseeitherβreaorβprobehaviors;while,asalreadymen- tialorder≺doesnottellwhichsystemis“moreresilient”;it tioned,βantpertainstoantifragilesystems. highlightsthatforinstancesystem“dog”isabletoexercise We shall define π as a projection map returning, for behaviorsthatarelesscomplexthanthoseofsystem“man”. eachoftheabovebehaviorclasses,anintegerin{1,...,5} This tells nothing about the extrinsic “level of congruence (π(βran) = 1, …, π(βant) = 5). Aim of π is twofold: orfit”[47]thatforinstancea“man”ora“dog”mayexhibit itassociatesaninteger“identifier”toeachbehavioralclass inagivenenvironment.Asexemplified,e.g.,in[14],when and it introduces an “order” among classes. Intuitively, the a threat is announced by ultrasonic noise, a “dog” able to higheristheorderofclass,themorecomplexisthebehavior. perceive the threat and flee could result more resilient than In what follows, it is assumed that behaviors manifest a“man”.Theuseofsentinelspecies[44]isinfactasocial themselves by changing the state of measurable properties. behaviorbasedonthisfact.Anapplicationofthisprinciple Asanexample,behaviormaytranslateintoavariationinthe isgiveninSect.5. electromagneticspectrumperceivedasachangeinluminos- ityorcolor.Contextfiguresarethetermthatshallbeusedto 3.2 Resilienceorgans refertothosemeasurableproperties. Foranybehaviorβ dependentonasetofcontextfigures x Asdonein[16],itisconjecturedherethatreasoningabouta F,notationβF willbeusedtodenotethatβ isexercisedby x x system’sresilienceisfacilitatedbyconsideringthebehaviors considering the context figures in F. Thus if, for instance, of the system organs that are responsible for the following F = (speed,luminosity), then βrFea refers to a reactive abilities: behaviorthatrespondstochangesinspeedandlight. Foranybehaviorβ andanyintegern > 0,notationβn x x willbeusedtodenotethatβ isexercisedbyconsideringn M:theabilitytoperceivechange; x A:theabilitytoascertaintheconsequencesofchange; contextfigures,withoutspecifyingwhichones. Asanexample,behaviorβp|Fro|,with F definedasabove, P: the ability to plan a line of defense against threats identifiesanorder-2proactivebehavior,whileβpFro saysin derivingfromchange; E:theabilitytoenactthedefenseplanbeingconceived addition that that behavior considers both speed and lumi- instepP; nositytoextrapolatethefuturepositionofthegoal. K:and,finally,theabilitytotreasureuppastexperience Now,theconceptofpartialorderamongbehaviorsisintro- andsystematicallyimprove,tosomeextent,abilitiesM, duced. A,P,andE. Definition7 (Partial order of behaviors) Given any two behaviors β1 and β2, β1 ≺ β2 if and only if either of the Theseabilitiescorrespondtothecomponentsoftheso-called followingconditionsholds: MAPE-Kloopofautonomiccomputing[30].Inthecontext ofthepresentpaper,thesystemcomponentsresponsiblefor 1. (π(β )≤π(β ))∧(∃(F,G):β =βF∧β =βG∧F thoseabilitiesshallbereferredtoas“resilienceorgans”. 1 2 1 1 2 2 (cid:2) G). In other words, β ≺ β if (1) β belongs to a Thefollowingnotationshallbeusedtorefertoorgan O 1 2 1 behavioralclassthatisatmostequaltoβ ’s(viafunction ofsystems:s.O(forO∈{M,...,K}). 2 123 40 JReliableIntellEnviron(2015)1:33–46 Definition8 (Cyberneticclass)Foranysystems,the5-tuple cybernetic class of the system and the dynamically evolv- corresponding to the behaviors associated to its resilience ing cybernetic class of the environment (see Sect. 2.4.1). organs, Thiscomparisonrepresentsasystem–environmentfit,inturn indicatingthepropertyemergingfromtheinterplaybetween (s.M,s.A,s.P,s.E,s.K), the current state of the system and the current state of the environment—inotherwords,thesystemislikelytoeither shallbereferredtoasthecyberneticclassofs. preserveorloseitspeculiarfeaturesbecauseoftheinterac- tionwiththeenvironment.Intheformercase,thesystemshall Twoobservationsareimportantforthesakeofourdiscussion. becalledas“resilient;”inthesecondcase,“unresilient.” 3.3 Againonelasticity,entelechism,andantifragility Observation1 (Intrinsic resilience) A system’s cybernetic class puts to the foreground how intrinsically resilient that Themodelandapproachintroducedthusfarprovideuswitha systemis(seeagainSect.2.4.1)andmakesiteasiertocom- conceptualframeworkforalternativedefinitionsofelasticity, parewhethercertainresilienceorgans(orthewholesystem) entelechism,andantifragility—namelytheresilienceclasses are(resp.,is)systemicallyinferiorto(thoseof)anothersys- introducedinSect.2. tem. Definition9 (Elasticity) Given a system s and its cyber- As an example, the adaptively redundant data structures neticclassC ,withs deployedinenvironmente,s shallbe s describedin[21]havethefollowingcyberneticclass: called“elastic”withrespecttoeifsisresilient(inthesense expressed in Observation 2) and if the behaviors in C are s C =(β ,β1 ,β ,β ,∅), 1 pur pro pur pur purposeful(βpur)anddefined,onceandforall,atdesignor deploymenttime. while the adaptive N-version programming system intro- Elasticity corresponds to simple, static behaviors that ducedin[10,11]is make use of a system’s predefined internal characteristics C =(β ,β2 ,β ,β ,β ). and resources so as to mask the action of external forces. 2 pur pro pur pur pur Those characteristics and resources take the shape of vari- Bycomparingtheabove5-tuplesC andC onemayeasily ousformsofredundancywhichisusedtomask,ratherthan 1 2 tolerate,change. realizehowthemajorstrengthofthosetwosystemsliesin theiranalyticorgans,bothofwhicharecapableofproactive Definition10 (Entelechism)Givenasystemsanditscyber- behaviors(βpro)—thoughinasimplerfashioninC1.Another neticclassC ,withs deployedinenvironmente,s shallbe s noteworthydifferenceisthepresenceofaknowledgeorganin referredtoas“entelechy”withrespecttoeifsisresilient(in C ,whichindicatesthatthesecondsystemisabletoaccrue 2 the sense expressed in Observation 2) and if the behaviors and make use of the past experience to improve—to some correspondingtotheAandPorgansofC areeitheroftype s extent—itsaction5. βreaorβpro.Entelechism,orchangetolerance,isdefinedas Please note that not all the behaviors introduced in thepropertyexhibitedbyanentelechy. Sect. 3.1 may be applied to all of a system’s organs. For instance,itwouldmakelittlesensetohaveaperceptionorgan As evident from their definition, entelechies are open, behaverandomly(unlessonewantstomodel,e.g.,theeffect context-aware systems; able to autonomically repair their ofcertainhallucinogenicsubstancesinchemicalwarfare6.) state in the face of disrupting changes; and able to guar- anteetheirsystemidentity.Aknowledgeorganmayormay Observation2 (Extrinsic resilience) A system’s cybernetic notbe present and, depending on that,the feedback organs classputstotheforegroundalsohowextrinsicallyresilient mayormaynotbestateful—meaningthat“memory”ofthe thatsystemisiftheabovecomparisonisdonebetweenthe experiencedchangesisorisnotretained. Definition11 (Computationalantifragility)Givenasystem 5 Thepresenceofaknowledgeorgandoesnotmeanthatitssystemis s and its cybernetic class C , with s deployed in environ- antifragile.Inthecaseathand,forinstance,thesystemdoesnotmutate s itsystemidentity—itstaysanN-versionprogrammingsystem. ment e, we shall say that s is computationally antifragile 6 See for instance the interview with Dr. James S. Ketchum, in the withrespecttoewhenthefollowingconditionsareallmet: SixtiesaleadingpsychiatristattheArmyChemicalCenteratEdgewood Arsenal in Maryland, US: “With BZ [3-quinuclidinyl benzilate], the 1. The awareness organ of s, A, is open to the system– individualbecomesdelirious,andinthatstateisunabletodistinguish environment fit between s and e. In particular, as sug- fantasyfromreality,andmaysee,forinstance,stripsofbaconalong theedgeofthefloor.”[46]. gestedinSect.2.3,thismeansthatAimplementsamodel 123 JReliableIntellEnviron(2015)1:33–46 41 of the risk of losing one or more aspects of the system identityofs.Onesuchmodelisexemplified,e.g.,bythe distance-to-failure function introduced in [19] and dis- cussedin[12]. 2. Throughout time intervals in which the behavior of e isstable,theplanningorgan P isabletomonotonically improveextrinsicresilience(seeObservation2)andthus optimizerisk/performancetrade-offs. Fig. 1 Exemplificationofturbulence,namelythedynamicevolution 3. Organ P evolves through machine learning or other ofenvironmentalorsystemicbehavior.Abscissasrepresenttime,“now” machine-orientedexperientiallearning[31],leadingsto beingthecurrenttime.Ordinatesarethebehaviorclassesexercisedby evolvetowardsevergreaterintrinsic(systemic)resilience eithertheenvironmentorthesystem (seeObs.1). 4. OrganKisstatefulandpersistslessonslearnedfromthe function” would be by encoding the characteristics of each experienceandits“conceptualization.” behaviorontothebitsofabinaryword,withthethreemost significantbitsencodingtheprojectionmapofthebehavior Computationallyantifragilesystemsaresystem–environment andthebitsfromthefourthonwardencodingthecardinality fit-aware systems; able to embody and persist systemic ofthesetofcontextfiguresortheorderofthebehavior.When improvements suggested by the match of the current sys- thebehaviorsbelongtothesameclassalthoughconsidering temidentitywiththecurrentenvironmentalconditions.The twodifferentcontextsets,say F andG,thenasimplerfor- learningactivitypossiblyimpliesa4-stagecyclesimilarto mulationofadistancewouldbeabs(|F|−|G|).Letuscall the one suggested by Kolb in [31], executed concurrently distonesuchmetricfunction. withtheresiliencebehaviorsoftheM,A,P,andEorgans. It is now possible to propose a definition of two indica- torsfortheextrinsicqualityofresilience:Thesystemsupply relativetoanenvironmentandthesystem–environmentfit. 4 Approach Definition12 (System supply) Let us consider a system s As already mentioned, a methodological assumption in the deployedinanenvironmente,characterizedrespectivelyby present article is that the evolution of an environment may behaviorsβs(t)andβe(t).Letusassumethatβ andβ are 1 2 be expressed as a behavior. Said behavior may be of any such that either β ≺ β or β ≺ β . Given a behavioral 1 2 2 1 of the types listed in Sects. 3.1 and 3.3 and it may result metric function dist defined as above, the following value inthedynamicvariationofanumberof“firingcontextfig- shallbecalledassupplyofs(t)withrespecttoβe(t): ures”.Infact,thosefigurescharacterizeand,inasense,set theboundariesofanecoregion,namely“anareadefinedby supply(s,e,t) ⎧ its environmental conditions” [2]. An environment may be the behavioral result of the action of, e.g., a human being ⎪⎪⎨dist(βs(t),βe(t)) if βe(t)≺βs(t) −dist(βs(t),βe(t)) if βs(t)≺βe(t) (a “user”); or the software agents managing an intelligent = (2) ⎪⎪⎩0 if βe(t)andβs(t) ambient; or for instance it may be the result of purpose- expressthesamebehaviors. less(random)behavior—suchasasourceofelectromagnetic interference. As a consequence, an environment may for Supplycanbepositive(referredtoas“oversupply”),neg- instance behave (or appear to behave) randomly, or it may ative(“undersupply”),orzero(“perfectsupply”). exhibit a recognizable trend; in the latter case the variation ofitscontextfiguresmaybesuchthatitallowsfortracking Observation3 Oversupplyandundersupplyprovideaquan- orspeculation(extrapolationoffuturestates).Moreover,an titative formulation of the notions of overshooting and environmentmayexhibitthesamebehaviorforarelatively undershootinggiveninSect.2. longperiodoftimeoritmayvaryfrequentlyitscharacter. Givenanenvironment(orasystem),thedynamicevolu- Definition13 (System–environmentfit)Giventhesamecon- tionoftheenvironmental(resp.systemic)behaviorshallbe ditionsasinDefinition12,thefollowingfunction: referredtoas“turbulence”.DiagramssuchastheoneinFig.1 (cid:7) maybeusedtorepresentthedynamicbehavioralevolution fit(s,e,t)= 1/(1+supply(s,e,t)) ifsupply(s,e,t)≥0 ofeitherenvironmentsorsystems. −∞ otherwise. Whenevertwobehaviorsβ andβ aresuchthatβ ≺β , 1 2 1 2 itispossibletodefinesomenotionofdistancebetweenthe shallbereferredtoas“system–environmentfitofs andeat two behaviors. One way to define such “behavioral metric timet.” 123 42 JReliableIntellEnviron(2015)1:33–46 ior exercised by the environment is evenly matched by the featuresofthesystem.Durings ands ,wehaveovershoot- 2 3 ings:thesystemicfeaturesaremorethanenoughtomatchthe currentenvironmentalconditions.Itisacaseofoversupply. Correspondingly,fitisratherlow.Ins ,theoppositesituation 5 takes place: the systemic features—for instance, pertaining to a perception organ—are insufficient to become aware of allthechangesproducedbytheenvironment.Inparticular, herechangesassociatedwithcontextfigure5goundetected. This is a case of undersupply (that is to say, undershoot- ing),correspondingtoalossofidentity:the“worstpossible” Fig. 2 Exemplificationofsupplyandsystem–environmentfit system–environmentfit. Theabovedefinitionexpressessystem–environmentfitas 4.1 Supply-andfit-awarebehaviors a function returning 1 in the case of best fit; slowly scal- ing down with oversupply; and returning −∞ in case of Wheneverapartialorder“≺”existsbetweenasystem’sand undersupply. The reason for the infinite penalty in case of itsenvironment’sbehaviors,itispossibletoconsidersystem undersupplyisduetothefactthatitsignifiesanundershoot- behaviorsofthefollowingforms: ingor,inotherwords,alossofsystemidentity. Thejustenunciatedformulationisofcoursenottheonly possible one: an alternative one could be, for instance, by 1. Either b = βpFro or b = βaFnt, with σ(b) = false and with F including figures that provide a measure of the havingsupply2 insteadofsupplyinthedenominatoroffit riskofunresilience,expressed,e.g.,throughsupplyand inDefinition13.Anotherformulationcouldextendoptimal fit. fit to a limited region of oversupply as a safety margin to Such behavior corresponds to condition 11 in Defini- coverforinaccuraciesinthepredictionofthebehaviorofthe tion11,namelyoneofthenecessaryconditionstocom- environment.Thisissimilartotheroleofεin(1). putationalantifragility.Whenexercisedbysystemorgans Figure2exemplifiesasystem–environmentfitinthecase foranalysis,planning,andknowledgemanagement,this of two behaviors βs and βe with s (cid:2) e. Environment e behaviortranslatesinthepossibilitytobecomeawareof affects five context figures identified by integers 1,...,5 andspeculateaboutthepossiblefutureresiliencerequire- whilesaffectscontextfigures1,...,4.Thesystembehavior ments. If this is coupled with the possibility to revise is assumed to be constant, thus for instance if s is a per- one’s system organs by enabling or disabling, e.g., the ception organ then it constantly monitors the four context ability to perceive certain context figures depending on figures 1,...,4.Onthecontrary,βe varies withtime.Five theextrapolatedfutureenvironmentalconditions,thena timesegmentsareexemplified(s ,...,s )duringwhichthe 1 5 system could use this behavior to improve proactively followingcontextfiguresareaffected: its own system–environment fit—possibly mutating its featuresandfunctions. s :Contextfigures1,...,4. 1 Anexemplarysystembasedonthisfeatureisgivenbythe s :Contextfigure1andcontextfigure4. 2 already mentioned adaptively redundant data structures s :Contextfigure4. 3 of[21]andtheadaptiveN-versionprogrammingsystem s :Contextfigures1,...,4. 4 introduced in [10,11]. Those systems make use of the s :Contextfigures1,...,5. 5 so-calledreflectivevariables[20]toperceivechangesin a“distance-to-failure”function.Suchfunctionbasically Context figures are represented as boxed integers, with an measurestheprobabilityoffailureofavotingschemeat empty box meaning that the figure is not affected by the thecoreofthereplicationstrategiesadoptedinthemen- behavior of the environment and a filled box meaning the tionedsystems.Inotherwords,suchfunctionestimates figure is affected. The behavior of the environment is con- theprobabilityofundersupplyforvoting-basedsoftware stantwithinatimesegmentandchangesatthenextone.This systems. is shown through the sets at the bottom of Fig. 2: for each 2. Behaviorbdefinedasincase1,butwithσ(b)=true. segmentt ∈{s ,...,s }thesupersetise(t )whilethesub- Inthiscase,theanalysis,planningandknowledgeorgans s 1 5 s set is s(s ), namely e(s )∩s. The relative supply and the areawareofothersystemsinphysicalorlogicalproxim- t t system–environmentfitalsochangewiththetimesegments. ityandmayusethisfacttoartificiallyaugmentorreduce Durings ands thereisperfectsupplyandbestfit:thebehav- their system features by establishing / disestablishing 1 4 123
Description: