Statistics of the lifespan of complex systems Or: is there senescence? Anders Sandberg Future of Humanity Institute Oxford Martin School University of Oxford
StatisticsofthelifespanofcomplexsystemsOr:istheresenescence?
AndersSandbergFutureofHumanityInstitute
OxfordMartinSchoolUniversityofOxford
RobertMcCall,ThePrologueandthePromise(1983)
Myinterest:howlong-termcanprojectsbe?
• Arethereany“physical”lawsthatpreventsocietiesfromexistingindefinitely?
Wecareaboutcivilization-likecomplexsocialorganisations
• Theycanallowthegrowthofhumancapitalandwellbeing• Economiesofscale• Long-termprojects• Existentialriskreduction?
• Theirfallinducessignificantlossesofwellbeing,knowledge,populationandmaynowinduceexistentialrisks.
PieterBruegeltheElder:TheTowerofBabel(c.1563)
Macrohistorical theoriesofcivilizationlongevity• Cyclictheories(ibnKhaldun,Vico,Turchin)• Decadence(Gibbons)• Organicageing(Mommsen,Spengler)• Entropy(Adams,Georgescu-Roegen)• Spiritualdecline/internalpowerfailure(Toynbee)• Unsustainablecomplexity(Tainter)• Risingenergycosts(Homer-Dixon,Rifkin,Garrett)• Environmentaldegradation(Diamond)• Outsideinvasion• Disasters,environmentalshifts(Diamond)
Increasingriskovertime
Istheregrowingrisk?
• Let’stestwithhistoricaldata!• Nullmodel:constantrisk.
• Survivalcurve:exponentialdeclineovertime.
• Weibulldistributionfit:riskgrowingas𝑡"#$ overtime.
• PDF𝑓 𝑡; 𝑘, 𝜆 = "+
,+
"#$𝑒#
./
0
• Survivalcurve:morerectangular(𝑘 >1,“ageing”)orsteeperearlydecline(𝑘 < 1,“childhoodmortality”)
• (Needtoconsidercensoringissues)
Empires
• Civilizationsarebadlydefined(andprobablytoofewtobestatisticallysignificant).Whatwecareforislargetransgenerationalsocialorganisations
• Solution:useempiresandstatesinstead
ThomasCole:TheArchitect’sDream(1840)
Taagepera empiredata
• ReinTaagepera:• Empire:“anyrelativelylargesovereignpoliticalentitywhosecomponentsarenotsovereign”.
• 5,000yearsofempires.
• SamuelArbesman(2011):exponentialsurvivalcurvewithhalf-life220years.
Taagepera empiredata
• Fromstarttofinish:• 𝜆 = 348.3815and𝑘 = 0.9205with95%confidenceinterval[0.7869,1.0768].• Hence,cannotrejectnullhypothesisofnoageing!
• From“adulthood”(80%ofmaxsize)tofinaldate:• 𝜆 = 188.1137 yearsand𝑘 = 1.0217.Noageing!
• Notrendovertime(!)
Otherempiresets
• Wikipedia'slistofempires:𝜆 = 270.9173 yearsand𝑘 =0.9986with95%CI[0.8951,1.1141].Noageing.
• EstimatesbyLarryFreemanonhisblogofthedurationsofancientcivilizations/empires:𝜆 = 381.8091yearsand𝑘 = 1.2500witha95%CI[1.0294,1.5177].Mildageing,nearlynonsignificant.
StatesEurope1000-1850
• DatacourtesyofSchönholzerandWeese,basedonBairoch.• 𝜆 = 136.6277 yearsand𝑘 = 0.8604with95%CI[0.7950,0.9312]:abitofchildhoodmortality!
Worldsystems
• Worldsystemstheory/perspective(Wallerstein,Chase-Dunn)• 9pastcivilizations/worldsystemslistedbyWilkinson.• 𝜆 = 1080.7years(theypersistfarlongerthantheirindividualcomponentempires)and𝑘 = 0.8083,CI[0.4611,1.4169].• Noclearageing,butlittledata.
Humanbigsocialorganisations
• Atleastinthisdata,noclearsignofsenescence:riskisconstantovertime.
• Caveat:thishasbeenlifespans ratherthantimetocollapse.
• Maybecollapseshavedifferentdistribution.
• ButIwouldneedadefinitionofwhichcaseisacollapsetotest.
Complexadaptivesystems
• States,societies,civilizationsareallCAS.
• Backgroundassumption:maybetherearejointrulesforlargeclassesofCAS.• (Backgroundworry:maybeeverythingisjustindividualcasesbecauseof“wild”properties.)
• WhatkindsofCASshowlimitedlifespans?
Firms
• Exponentialsurvivalcurve• Independentofbusinesssector• Sizeatbirthhaspositivecorrelationlongevity
• Liabilityofnewnessandadolescence• 𝜆 ≈ 10 yearsforcurrentcorporations,𝜆 ≈ 170 yearsforextremetail
• Liabilitiesatbankruptcypower-laws,butlongevityexponential(Fujiwara2004)
• Salesandcostscloselybalanced?(West2017)
Daepp,M.I.,Hamilton,M.J.,West,G.B.,&Bettencourt,L.M.(2015).Themortalityofcompanies.JournalofTheRoyalSocietyInterface,12(106),20150120.
Species
• VanValen’s lawofconstantextinction• Appearstrueinmanybutnotallcases• Complicationsduetofossilrecord,clades• ExplainedbyoneofthevariantRedQueenHypotheses?
• Žliobaitė,Fortelius&Stenseth:extinctionduetoabioticfactors,bioticcompetitionforspatialspread.
Pearson,P.N.(1995).Investigatingage-dependencyofspeciesextinctionratesusingdynamicsurvivorshipanalysis.HistoricalBiology,10(2),119-136.
Jaeger,J.J.(1994).TheevolutionofbiodiversityamongtheSouthwestEuropeanNeogenerodent(Mammalia,Rodentia)communities:patternandprocessofdiversificationandextinction.Palaeogeography,Palaeoclimatology,Palaeoecology,111(3-4),305-336.
Unclearcases
• Institutions?• Multi-centuryuniversities.• Multi-millenniareligiousorders.• Toolittledataforthelong-livedones.
• Cities?• Jericho9,000BC,fortificationsfrom6,800BC.• GeoffreyWest:renewingeconomiesofscale?
Biologicalageing
• Senecense commoninmulticellularorganisms.• Butexceptionsinsomespecies:Perennialangiosperms,Blanding'sturtle,Olm,easternboxturtle,redseaurchin,rougheye rockfish,oceanquahogclam,bristleconepine...
Software
• Japanesecorporatesystems𝜆 ≈ 11.4 years,𝑘 ≈ 1.7• MeirLehman'slawsofsoftwareevolution:unlessworkhardtocounterit,maintainabilitywillgodownduetogrowingentropy/complexity.• “Lawofcontinuingchange.Asystemthatisusedundergoescontinuingchangeuntilitisjudgedmorecosteffectivetofreezeandrecreateit.“
• “Lawofincreasingentropy.Theentropyofasystem(itsunstructuredness)increaseswithtime,unlessspecificworkisexecutedtomaintainorreduceit.“
Belady,L.A.,&Lehman,M.M.(1976).Amodeloflargeprogramdevelopment.IBMSystemsjournal,15(3),225-252.
CASageing
• Causes:• Evolutioncanproduceageingbecauseofgeneticdrift
• Growthandunmitigatedcomplexity
• Notatalluniversal:manynon-ageingCAS• Competitiondoesnotcauseageing
• Butwhyshouldlack ofageingbecommon,either?
Asimplemodelofconstantrisk• Randomchallengeshappensatrate𝜆 withseverityX ∼ 𝑓 𝑥 .• Xhascomplementarycumulativedistributionfunction𝐹D 𝑥 = Pr[𝑋 > 𝑥].
• Systemhasresistancelevel𝜃, willsurviveif𝑋 < 𝜃.• Systemsurvivesageometricallydistributednumberofchallengeswithparameter𝐹D 𝜃 .
• Expectedtimeofsurvival:𝜏 = $+LD M
.
• Markovinequality𝐹D 𝑥 ≤ 𝐸[𝑋]/𝑥 givesthebound𝜏 ≥ M+R S
if𝐸[𝑋]exists.
• One-sidedChebyshevinequality𝐹D 𝑥 − 𝐸 𝑋 ≤ UV
UVWXVgivesthe
bound𝜏 ≥ $
+1 + MWR S V
UV∝ 𝜃[ ifvariance𝜎[ exists.
• If𝑓 𝑥 isheavy-tailedallbetsareoff.Weakdependenceon𝜃!
Heavytaildistributedrisksarecommon
• Naturaldisasterfatalities(earthquakes,volcanos,floods,tsunamis…)• Richardsonlawofwarfatalities.• Pandemicoutbreaksizes?• Ifcombiningseveral,mostextremetaildominates.
• ForextremeParetodistribution,survivaltimeindependent of𝜃
102 104 106 10810-3
10-2
10-1
100
Pr(X
³ x
)
x
War dead
100 101 102 103 104 10510-3
10-2
10-1
100
Pr(X
³ x
)
x
Flood dead
Synthesis:collapseascombined badluck
• Thedisastersaremixturesofveryextremeeventsorcombinationsoffactors.
• Hardtoavoid(unprecedentedorrareconfluences,“blackswans”)
• Competitivepressures(cost,alternativecosts,competition)leadtofinite𝜃.Butevenlarge𝜃 insufficientagainsttailrisks.
Centralsystemidea• Wilkinson:a“centralsystem”ofmergingpolitical-militarynetworkshasgrownbymergersandexpansionover7,500years
• Shklovskii &Sagan:“ThepresenttechnicalcivilizationoftheplanetEarthcanbetracedfromMesopotamiatoSoutheasternEurope,toWesternandCentralEurope,andthentoEasternEuropeandNorthAmerica.”
Whatdoesthistellusaboutourfragility?
• Maybelessthanwewouldwish:wearethefirstglobal,high-techcivilization.Wehavegoodreasonstobeconcerned.
• Butthereisapparentlynounavoidablelawforcingdecline onus.
• TheCentralSystemhasbeenaroundfor7,500years.Wehaverebuilteverhigher.Buthavingbackupssoundslikeagoodidea. RobertMcCall,ThePrologueandthePromise(1983)