Smarter Intervention in Complex Systems

7/30/2019 Smarter Intervention in Complex Systems

1/32

SMARTER INTERVENTION

IN COMPLEX SYSTEMS

Maniesto or

STOP?

OBSE

RV

EO

RIE

NTDEC

IDE

ACT

INTERVENTIONMECHANISM

INTERVENTION

PRINCIPLE

INTERVENTION

MINDSET

Mark Fell, Managing Director, Carr & Strauss

with an Introduction byHanne Melin, Policy Strategy Counsel, eBay Inc.


2/32

Mark Fell is the Managing Director o Carr & Strauss, a rm t hat specialises in decision support

systems or proessional strategists. During the course o a career in strategy and governancehe has advised a wide range o organisations, including Arup, BSkyB, Chiquita, Diageo, eBay,

Este Lauder, Europes blank recording-media industry, the European Cancer Organisation, theEuropean Cooperative Movement, Hydro, Knau Insulation, LVMH Group, Mastercard, Nutricia,

PayPal, Richemont Group, Shecco, UPS and Vodaone. Mark holds a Masters Degree in Euro-

pean Politics and Administration rom the College o Europe, Bruges, a First Class HonoursDegree in Politics rom Edinburgh University and has studied at the Institut dEtudes Politiques,

Strasbourg. In his ree time he enjoys testing himsel through endurance running, having com-pleted races at the North Pole, in Antarctica, the Sahara, Amazon, Himalaya and Alps, as well as

an Ironman Triathlon.

About the Author

This maniesto draws extensively on the work o David Cli, Mary Cummings, Len Fisher, TimHarord, Jaron Lanier, Donella Meadows, Melanie Mitchell, Jason Pontin, John Sterman, Steven

Strogatz and Duncan Watts, as well as the labour o countless others too numerous to mentionhere. It is also the ruit o my interaction over the years with clients and colleagues rom a wide

array o sectors and geographies. My requent collaborations with Hanne Melin o eBay are oparticular note in this regard. As ever, any errors and omissions are solely o my own making.

Acknowledgements

Copyright 2013 Carr & Strauss. All rights reserved.


3/32

EXECUTIVE SUMMARY ...... 6

INTRODUCTION BY HANNE MELIN ...... 8

LIGHTS - ILLUMINATING THE CHALLENGE ...... 10

A maniesto or change ...... 10

CAMERA - BRINGING THE CHALLENGE INTO FOCUS WITH SYSTEMS DYNAMICS ...... 12

The global system dened ...... 12

Systems Dynamics introduced ...... 14

The example o product awareness as a system ...... 18

Leverage points in systems identied - Twelve Heuristics ...... 20

ACTION - SETTING OUT A METHODOLOGY TO MEET THE CHALLENGE ...... 24

A methodology or Smarter Intervention in Complex Systems:

- The Mindset ...... 26

- The Mechanism . ..... 34

- The Principle ...... 46

CONCLUSION ...... 50

Fast Forward to the Future ...... 50

ENDNOTES ...... 56

Contents


4/32

LIGHTS | CAMERA | ACTION ! MANIFESTO FOR SMARTER INTERVENTION IN COMPLEX SYSTEMS

76

Executive Summary

This maniesto is a call or better decisionmaking in our social, technological, environ-

mental, economic and political systems.

It marshals a series o disparate conceptsinto a ramework. This ramework aims to

acilitate Smarter Intervention in Complex

Systems. I t is a methodology.

In doing so the maniesto draws extensivelyon the ideas, words and images o many

proessionals rom a wide array o disci-plines. Systems Dynamics and Complex-

ity Theory helps to guide us through thisprocess.

Smarter Intervention is based on three

tenets:

(1) MINDSET: The systems we

are seeking to shape are otencomplex. They are nonlinear . A

does not always lead to B to C toD to E. Instead they can be ull

o surprises. We cannot predict-and-control their behaviour.

Instead we need to pursue a

ar more iterative trial-and-error approach. One in which

the decisions that we take aresurvivable.

(2) MECHANISM: The OODA Loop(Observe-Orient-Decide-Act) is

a practical method or actioningthis trial-and-error approach.

We can cycle this loop by using

experts or non-experts. Theymay be individuals or groups

o individuals - i.e. crowds.Increasingly the loop is being

cycled by sotware algorithms.Each o these Intervention

Agents has its strengthsand weaknesses. These are

examined at length in thisManiesto. Agents can emanate

rom either the public or the

private sector. The key messageis that none o these agents

is the panacea to cycling theloop. In each specic case we

need to identiy the appropriatecombination o these agents.

We require an I ntervention Mix.

(3) PRINCIPLE: To achieve this mix

a new Intervention Principle isproposed. To the extent that

a system can successully sel-organise, it should be let to

do so. Where intervention is

necessary, then the case has

to be made or why a givenintervention agent should be

involved in the OODA loop - i.e.the best agent to solve this type

o problem. The aim is to get allo these dierent agents cycling

around the loop as a team, at the

right tempo and to be able tohold them to account.

Smarter Intervention in Complex Systems

means pursuing a new intervention

mindset, mechanism and principle.


5/32


98

Introduction by Hanne Melin

In the early 1980s, commercial attempts todevelop mobile phone services in Sweden

were resisted. The reason being the right-ening legal uncertainty when a new tech-

nology was not covered by existing rules.

Experience has surely taught us to bettercope with change but have we actually

ound the policy ramework or doing sowith condence?

The European Commission launched in

2005 the Better Regulation agenda aimed

at improving lawmaking in a ast movingEurope. In 2010, the Commission evolvedthis agenda into a Smart Regulation strate-

gy to more eectively embed the objectives

o Better Regulation into the whole policycycle.

These are the right steps but I am convinced

that more evolution in act transorma-tion now needs to happen in the area o

Hanne is Policy Strategy Counsel at eBay and

ormerly associate in the EU competition law

practice o Sidley Austin LLP based in Brussels

While all other sciences have advanced,

that o government is at a standstill - little

better understood, little better practiced

now than three or our thousand yearsago.

John AdamsSecond President o the United States

policymaking in order to not only step intothe 21st Century but keep moving in synch

with it.

Take robotic technologies, expected bymany to revolutionise our lives even more

prooundly than the mobile phone. How

should the law think about robots, asks a re-cent essay rom Washington University. The

authors warn that the law almost always

considers new technology as merely a new

orm o something else,1 which sounds a lotlike what happened in Sweden in 1981.

Indeed, two interrelated questions both at

the heart o good lawmaking in changing

times - have been discussed or years.

The rst is nding the right tools or prob-lem-solving. Regulation is one o the manyways o implementing public policy but it is

not necessarily the best way o solving a given

problem nor is it the only way, stressed the2001 Mandelkern Group report that ormed

the basis or the EU Better Regulationagenda.2

The second is involving the appropriate ac-tors in decision-making. A 2006 study com-

missioned by the European Commissionand carried out by the Hans-Bredow-Institut

stressed that in increasingly complex andrapidly changing societies knowledge is

held by dierent actors.3

The traditional answers to these two ques-tions have been co-regulation and consulta-

tions with the public. But are these the bestanswers when todays world requires us to

think less in orm and more about unc-tion?

Last summer, I was listening to Spotiy

ounder Daniel Ek on the radio. Why think

about the music album as something xedto 12 tracks, he asked. When artists can sim-

ply release songs once nalised, unctionrather than orm matters.

We could approach policymaking with a

similar mentality. I we reed ourselves othe constraints o traditional orm, how

would we go about releasing the outcome

wed like to see?

The Hans-Bredow-Institute study pointedtowards the need or a system governance

approach to work out better ways to achieve

the policy goals under changing conditions

in view o it being impossible to controlsocial systems.

Indeed, the European Parliament embarked

in 2011 on an exercise to identiy changes

needed to prepare or the complex environ-ment resulting rom a multi-polar worldwhere governance is more and more a multi -

level one, involving multiple actors in decision

making and implementation4 and wheretechnologies are accelerating change.

In parallel, the European Commission hasreorganized one o its Directorates-General

to better ace the challenges o the next ten

years in a digital world5 by inter alia setting

up a Complex Systems unit.

I see these actions as products o an evolu-tion in policy thinking on how to grapple

with ar more complexity than beore as

Chan Lai Fung, then Chairman o Singa-pores Smart Regulation Committee, dened

the challenge already in 2006.6

So where do we go rom here? EuropeanCommissioner Neelie Kroes is right when

she says that we need new legal rameworks

and a new way o thinking.7 We need the

ramework that pulls together the evolutionin thinking and importantly lets also poli-

cymaking detach rom orm, rigid division

between regulators and the public, and apreerence or control.

To my knowledge, no one has yet proposed

that ramework. Now, with this Maniestosomeone nally has!


6/32

LIGHTS| CAMERA | ACTION !Illuminating the Challenge

MANIFESTO FOR SMARTER INTERVENTION IN COMPLEX SYSTEMS

10 11

A Maniesto or Change

Time and again we are ailing to managecomplex systems in an increasingly inter-

dependent world, be they complex social,technological, environmental, economic or

political systems. This is unsustainable.

This Maniesto thereore makes the case orSmarter Intervention in Complex Systems.

Not because Smarter Intervention is the

panacea or a more sustainable uture, but

because without it, the chances o successare even slimmer.

To do so this Maniesto:

(1) Begins by raming the bigpicture, namely dening the

global system in which weoperate and rom which our

challenges emanate

(2) Proceeds to introduce Systems

Dynamics as a lense throughwhich to understand any system,

be it social, technological,

environmental, economic orpolitical

(3) Identies 12 Leverage Points

that have suraced over the

decades as experts have studiedall kinds o dierent systems

(4) Sets out a methodologyor practically applying this

systems knowledge, namely therequisite intervention Mindset,

Mechanism and Principle

the challenges we ace are real. Finance

is in tumult, and while worries about a

banking crisis may have ebbed, ears o a

crisis in public nances are running high.Even without additional nancial reversals,

overall economic prospects look bleaker

than just a ew years ago. The global order

also seems more uncertain. Prosperity

and power are shiting to new places and

peoples. Old political doctrines and divi-

sions no longer seem viable. Technology

and media are changing beore our eyes.

So, apparently is the natural environment

itsel.

Pankaj Ghemawat, World 3.01


7/32

LIGHTS | CAMERA | ACTION !Bringing the Challenge into Focus with Systems Dynamics


12 13

The Global System Dened

Planetary

Sources

Our Society

A Subsystem

Planetary

Sinks

high-

grade

energy

low-

grade

energy

materials

& fossil fuels

wastes

&

pollution

solar

energy

heat

loss

GLOBALECOSYSTEM

Political

Systems

Economic

Systems

Technology

Systems

The set of systems through which

society seeks to satisfy its needs

INCREA

SING

INTE

RPLAY

OFALLT

HESE

SYSTE

MS

ACRO

SSBORDER

S

GLOB

ALISATION

THE BIG PICTURE

To make the case or Smarter Interventionin Complex Systems we rst need a concep-

tual ramework - a lense through which tobring the challenge into ocus.

This begins with a view o how social,

technological, environmental, economicand political (STEEP) systems relate to one

another.

The standard way to dene the economy is

as a system o production, distribution and

consumption o goods and services. This is

where economic value is created or de-stroyed.

However this Maniesto adopts a more ho-

listic view o the economy, seeing it as beingnot just about production, distribution andconsumption, but rather as the set o sys-tems through which a society seeks to satisy

its needs. This expanded denition there-

ore also includes our political and technol-ogy systems.

Our society is an expression o how all o

these systems interact - the resulting cultur-

al relations and institutional arrangements.It is here that we can evaluate whether

social value is being created or destroyed.

In turn, our society is nested in a biggerglobal ecosystem that will or will not ulti-

mately sustain it. At this level we need toconcentrate our eorts on conserving and

nurturing environmental value.

Globalisation entails the increasing inter-

action o all these systems regardless ogeography.

Intervention means any initiative that seeks

to inuence a systems behaviour. This maytake many orms, including government

regulation, subsidies and scal incentives. Itmay be a decision made by business or an

action by civil society. Increasingly it mani-ests itsel as algorithmic intervention - thinko speed cameras, driverless trains and High

Frequency Trading.

What thereore is this recurrent theme,namely a system?

FIGURE 1 (opposite page) - THE GLOBAL SYSTEM

How Everything Fits Together

Source: Adapted rom The Limits o Growth by Meadows, Randers and Meadows1


8/32



14 15

Systems Dynamics Introduced

WHY SYSTEMS DYNAMICS

Pioneered at MIT, Systems Dynamics is anapproach to understanding the behaviour

o complex systems over time.

Donella Meadows was a leading proponent

o systems dynamics and a key voice inwhat has become known as the sustain-

ability movement, the international eortto reverse damaging trends in the social,

technological, environmental, economic orpolitical systems.

Her work is widely recognized as a ormative

inuence on hundreds o other academic

studies, government policy initiatives, andinternational agreements.

This Maniesto will draw extensively on

Meadows work (particularly her bookThinking in Systems)1 to set out the un-

damentals o systems and their leveragepoints.

What ollows in this section at rst sight ap-

pears abstract. However consider the act

that on 6 May 2010 The Flash Crash wipednearly $1,000bn o the value o US shares

or a period o several minutes. Had thelosses not been recovered, the Dow would

have suered one o its biggest one-dayalls in history.

And that the UK governments subsequent

investigation into the High Frequency Trad-

ing system that caused this behaviour has

recently discovered that:

... in specic circumstances, a key typeo mechanism can lead to signicant

instability in nancial markets with

computer based trading (CBT): sel-

reinorcing eedback loops (the eect o

a small change looping back on itsel

and triggering a bigger change, which

again loops back and so on) within well-

intentioned management and control

processes can ampliy internal risks

and lead to undesired interactions and

outcomes.

The eedback loops can involve risk-

management systems, and can be

driven by changes in market volume or

volatility, by market news, and by delays

in distributing reerence data.2

This is the language o Systems Dynamics -

one o eedback loops and delays.

Equipped with the lense aorded by thisdiscipline Meadows oresaw this result al-

ready back in the early 1990s stating that:

...the great push to reduce inormation

and money-transer delays in nancial

markets is just asking or wild gyra-

tions.3

She observed that:

The industrial society is just beginning

to have and use words or systems, be-

cause it is only beginning to pay atten-

tion to and use complexity.4

A act brought home by the September

2011 edition o the Harvard Business Review

whose cover story headlined with, Embrac-ing Complexity - You Cant Avoid It, But Your

Business Can Prot From It.

Think o systemic risk, a term that policy-makers are just starting to grapple with as

they seek to tackle the nancial crisis.

To properly understand complex systemswe need to look at them through the Sys-

tems Dynamics lense.

THE BASICS

With this said and using Meadows explana-

tion (see her paper Leverage Points - Places

to Intervene in a System)5 let us start withthe basic diagram on the next page:

The state o the system is what-

ever standing stock is o importance:

amount o water behind the dam,

amount o harvestable wood in a orest,

number o people in the population,

amount o money in the bank, what-

ever. System states are usually physical

stocks, but they could be nonmaterial

ones as well: sel-condence, degree o

trust in public ocials, perceived saety

o a neighbourhood.


9/32



16 17

There are usually infows that increase

the stock and outfows that decrease

it. Deposits increase the money in the

bank; withdrawals decrease it. River

infow and rain raise the water behind

the dam; evaporation and discharge

through spillway lower it. Births and

immigrations increase the population,

deaths and emmigration reduce it.

Political corruption decreases trust inpublic ocials; experience o well unc-

tioning government increases it.

In soar as this part o the system con-

sists o physical stocks and fows - and

they are bedrock o any system - it obeys

the laws o conservation and accumula-

tion. You can understand its dynamics

readily, i you can understand a bathtub

with some water in it (the stock, the

state o the system) and an infowing

tap and outfowing drain. I the infow

rate is higher than the outfow rate, the

water gradually rises. I the outfow

rate is higher than the infow, the water

gradually goes down. The sluggish re-

sponse o the water level to what could

be sudden twists in input and output

valves is typical; it takes time or fows o

water to accumulate in stocks, just as it

take time or water to ll up or drain outo the tub. Policy changes take time to

accumulate their eects.

The rest o diagram shows the inorma-

tion that causes the fows to change,

which then cause the stock to change. I

youre about to take a bath, you have a

desired water level in mind (your goal).

You plug the drain, turn on the aucet,

and watch until the water rises to your

chosen level (until the discrepancy be-

FIGURE 2 - THE IMPACT OF THE 6 MAY 2010 FLASH CRASH ON THE DOW

Not understanding complexity and systems dynami cs can have potentially serious consequences

Source: Bloomberg

FIGURE 3 - HARVARD BUSINESS REVIEW - EMBRACING COMPLEXITY

We are just starting to wake up and pay attention to complexity and systems dynamicsSource: Harvard Business Review

FIGURE 4 - THE FUNDAMENTALS OF A SYSTEM

Systems consist o stocks, ows, eedback loops and goals

Source: Adapted rom Leverage Points - Places to I ntervene in a System by Donella Meadows

tween the goal and the perceived state

is zero). Then you turn the water o.

I you start to get into the bath and

discover that youve underestimated

your volume and are about to produce

an overfow, you can open the drain or

awhile, until the water goes down to the

desired level.

Those are two negative eedback loops,

or correcting [balancing] loops, one

controlling the infow, one controlling

the outfow, either or both o which you

can use to bring the water level to your

goal.

... Very simple so ar. Now lets take

into account that you have two taps,

a hot and a cold, and that youre also

adjusting or another system state:

temperature. Suppose the hot infow is

connected to a boiler way down in the

basement, our foors below, so it doesnt

respond quickly. An d youre making

aces in the mirror and not paying close

attention to the water level. The system

begins to get complex, and realistic, and

interesting.

There is also a second type o eedbackloop, namely a positive or reinorcing one.

These are sel-enhancing, leading to expo-

nential growth or to runaway collapses overtime - a snowballing avalanch eect. Think

o the escalating noise when a microphonecomes too close to a loudspeaker.

Goal

STATE OF

THE SYSTEM

Inow Outow

Perceived State

Discrepancy


10/32



18 19

Example o the Product Awareness System

Donella Meadows explanation gives us aclear picture o the basics o a system. Let

us now bring these systems concepts moreto lie. Take the example o how consumers

become aware o a sellers products. MITsProessor John D. Sterman examines this

system in his book Business Dynamics. He

reasons:

How do potential customers become

aware o a rms products? There are 4

principal channels: advertising, direct

sales eort, word o mouth, and media

attention. Each o these channels cre-

ates positive eedback [loops].

In most rms their advertising budget ...

grows roughly as the company revenue

grow. Larger advertising budgets have

two eects: (1) more potential consum-

ers are made aware o the item and

choose to enter the market (loop R1); (2)

to the extent the advertising is eective,

more o those who are aware and in

the market are likely to buy the product

oered by the company (R2). Similarly,

the larger the revenue o the rm, the

greater the sales budget. The more sales

representatives, and the greater their

skill and experience, the more calls they

can make, the more time they can spend

with customers, and the more eective

their calls will be, increasing both total

industry demand (R3) and the share o

total demand won by the rm (R4).

While a rm controls its advertising

and sales budgets, word o mouth and

media attention are largely outside

o their direct control ... As sales boost

the installed base and the number ocustomers who have experience o the

product, avourable word o mouth

increases awareness, increasing total

demand (R5) and also persuading more

people to purchase the products o the

seller (R6). A hot product or company

will also attract media attention, which,

i avourable, stimulates additional

awareness and boosts market share still

more (R7-9).1

IndustryDemand

Favourable

Word ofMouth

Sales or Firm

Growth Rate

Product

Attractiveness

Market

Share

Perception of

Hot ProductMedia

Reports

Sales

Installed

CustomerBase

Share fromWOM

R6Awareness

from WOM

R5

Awareness from

Media Reports

R7

Share from

Media Reports

R8

Hot Product

Eect

R9

+

+

+

+ + ++ +

+

+

+

+

+

+

Awareness

from Ads

R1 Share fromAdvertising

R2

Awareness

from SalesEort

R3

Share from

Sales Eort

R4Advertising

Revenue

Sales

Market

ShareIndustry

Demand

Product

Attractiveness

Direct

Sales

Capability

+

+ +

+

++

+ +

+

+

These are two views of the same product awareness system

There are 4 channels to consumers - Advertising & Direct Sales Capability (depicted above)And Favourable Word of Mouth & Media Reports (shown below)

FIGURE 5 (opposite page) - TWO VIEWS

OF THE SAME SYSTEM

One view capturing the advertising & direct

sales channels, the other view the avour-

able word o mouth and media reports

channels

Source: Both gures adapted rom BusinessDynamics by John Sterman


11/32



20 21

Leverage Points in Systems

Donella Meadows identies 12 places to

intervene in a system.2 In increasing ordero eectiveness they are:

(1) NUMBERS: Constants and

parameters such as subsidies,taxes and standards

(2) BUFFERS: The size o stabilizingstocks relative to their ows

(3) STOCK AND FLOW

STRUCTURES: Physical systems

and their nodes o intersection

(4) DELAYS:The length o time

relative to the rate o systemchanges

(5) BALANCING FEEDBACK

LOOPS:The strength o the

eedbacks relative to the impactsthey are trying to correct

(6) REINFORCING FEEDBACK

LOOPS:The strength o the gain

o driving loops

(7) INFORMATION FLOWS:Thestructure o who does and doesnot have access to inormation

(8) RULES: Incentives, punishments,constraints

(9) SELF ORGANIZATION:Thepower to add, change or evolve

system structure

(10) GOALS:The purpose o the

system

(11) PARADIGMS:The mindset

out o which the system itsgoals, structures, rules, delays,

parameters arises

(12) TRANDSCENDINGPARADIGMS:The ability to getat a gut level that no paradigm

is true that every one o theseparadigms is a tremendously

limited understanding o reality

Figure 6 illustrates the hierarchical nature

o these leverage points. How changing anouter leverage point - one higher in the list -

can condition all those within it.

MOST

EFFECTIVELEAST EFFECTIVE

MOST

EFFECTIVE

INFORMATION &

CONTROL PARTS

OF SYSTEM

INFORMATION &

CONTROL PARTS

OF SYSTEM

PHYSICAL PARTS

OF SYSTEM

DELAYS

STOCK

&FLO

WSTRUCTURES

BUFFERS

NUMBERS

REIN

FORCI

NGFEEDBACKLOOPS

INFORM

ATIONFLOWS

BALA

NCING

FEEDBACKLOOPS

RULESSELF

ORGANISATION

GOALS

PARADIGMSTRANDS

CENDINGPARADIGMS

LEVERAGE POINTS12 Places to

Intervene in a System

FIGURE 6 - TWELVE LEVERAGE POINTS IN SYSTEMS IN ORDER OF EFFECTIVENESS

Paradigms are the most eective, buers and numbers the least

Source: Carr & Strauss drawn up rom Thinking in Systems by Donella Meadows

Numbers, the size o fows, are dead last

on my list o powerul interventions. Did-

dling with the details, arranging the deck

chairs on the Titanic. Probably 90 - no 95,no 99 percent - o our attention goes to

parameters, but theres not a lot o leverage

in them.1

Donella Meadows


12/32



22 23

Meadows states that:

Paradigms are the sources o systems.

From them, rom shared social agree-

ments about the nature o r eality, come

system goals and inormation fows,

eedbacks, stocks, fows, and everything

else about systems.3

Change the paradigm - the mindset rom

which the system emanates - and every-thing else changes.

Meadows observes that the higher the le-

verage point, the more the system will resist

changing it.

She is also careul to caveat this list notingthat it is tentative and that there are excep-

tions to every item that can move it up ordown in order o leverage:

I oer this list to you with much humil-

ity and wanting to leave room or its

evolution... [it is] distilled rom decades

o rigorous analysis o many dierent

kinds o systems done by many smart

people. But complex systems are, well

complex. Its dangerous to generalize

The test o a rst class mind is the ability

to hold two opposing views in the head at

the same time and still retain the ability to

unction.

F. Scott Fitzgerald on paradigms

about them. What you read here is still

work in progress; its not a r ecipe or

nding leverage points. Rather, its an

invitation to think more broadly about

system change.4

Elaborating on this she continues:

I have come up with no quick or easy

ormulas or nding leverage points incomplex and dynamic systems. Give me

a ew months or years and Ill gure it

out. And I know rom bitter experience

that, because they are so counterintui-

tive, when I do discover a systems lever-

age points, hardly anyone will believe

me.5

Take the example o amines. Jason Pon-tin, editor, journalist and publisher at MIT

explains:

Until recently, amines were understood

to be caused by ailures in ood sup-

ply (and thereore seemed addressable

by increasing the size and reliability o

the supply, potentially through new

agricultural or industrial technologies.

But Amartya Sen, a Nobel laureate

economist, has shown that aminesare political crises that catastrophi-

cally aect ood distribution. (Sen was

infuenced by his own experiences. As a

child he witnessed the Bengali amine

o 1943: three million displaced armers

and poor urban dwellers died unneces-

sarily when wartime hoarding, price

gouging, and the colonial governments

price-controlled acquisitions or the

British army made ood too expensive.

Sen demonstrated that ood production

was actuallyhigherin the amine years.)

Technology can improve crop yields or

systems or storing and transporting

ood; better responses by nations and

nongovernmental organizations to

emerging amines have reduced their

number and severity. But amines will

still occur because there will always be

bad governments.6


13/32

LIGHTS | CAMERA | ACTION !Setting out a Methodology to Meet the Challenge


2524

Smarter Intervention in Complex Systems

GETTING TO GRIPS WITH COMPLEXITY

Beore employing Meadows 12 leverage

points, we rst need to get a better

handle on complexity. Only then will wehave a better shot at achieving smarterintervention in complex systems.

Meadows draws attention to the actalthough people deeply involved in a

system oten know intuitively where to ndleverage points, more oten than not they

push the change in the wrong direction.

She observes that this is because complex

systems are inherently unpredictable.Nonlinearities in their architecture can

ip them rom one mode o behaviour to

another. Take as an example, an economicsystem ipping rom boom to bust.

Meadows explains that:

A linear relationship between two

elements in a system can be drawn

on a graph with a straight line. Its a

relationship with constant proportions.I I put 10 pounds o ertilizer on my eld,

my yield will go up 2 bushels. I I put

on 20 pounds, my yield will go up by 4

bushels. I I put on 30 pounds, Ill get an

increase in 6 bushels.

A nonlinear relationship is one in

which the cause does not produce a

proportional eect. The relationship

between cause and eect can only be

drawn with curves and wiggles, not

with a straight line. I I put 100 pounds

o ertilizer on, my yield will go up by 10

bushels; i I put on 200, my yield will not

go up at all; i I put on 300, my yield willgo down. Why? Ive damaged my soil

with too much o a good thing.1

Meadows also provides the example o

trafc ows:

As the fow o trac on a highway

increases, car speed is aected only

slightly over a large range o car

density. Eventually, however, small

urther increases in density produce a

rapid drop-o in speed. And when the

number o cars on the highway builds

up to a certain point, it can result in a

trac jam, and car speed drops to zero.2

Sotware can behave in complex nonlinear

ways. It is made up o lots o dierenti-then-else branches in t he code and

eedback loops. As stocks o data areprocessed changing combinations o these

branches and eedback loops come to

dominate the system at dierent moments.This can lead to unpredictable results.

That is why even ater exhaustive testingsotware or any signicant undertaking is

never entirely bug ree.

A good illustration o nonlinearity is the

Logistic Map - one o the most amousequations in the science o dynamical

systems.3

The equation states that:

xt+1=Rx

t(1-x

t)

The specics o the equation are

unimportant. What matters is that the

components o the equation (x, 1, R, etc.) arereadily comprehensible, but the resulting

behaviour o the equation is not. Itproduces erratic, chaotic and unpredictable

behaviour - see Figure 7.

Put dierently, the systems behaviour

is greater than the sum o its parts - it isemergent. To date no comprehensive

explanatory theory or complex systemsexists.

Melanie Mitchell, Proessor o ComputerScience at Portland State University,

explains:

The mathematician Steven Strogatz

has termed this goal the quest or a

calculus o complexity. The analogy

is apt in some ways: Newton, Liebniz

and others were searching or a general

theory o motion that could explain

and predict the dynamics o any object

or group o objects subject to physical

orce, whether it be earthly or celestial.

Beore Newtons time, there were

individual pieces o such a theory (e.g.

the n otions o innitesimal, derivative,

integral, etc. existed) but no one had

gured out how all these pieces t

together to produce a completely

general theory that explained a

huge range o phenomena that were

previously not unied. Similarly, today,

we have many dierent pieces o theory

related to complex systems, but no one

has yet determined how to put them

all together to create something more

general and uniying.4

Nevertheless, this has not preventedStrogatz rom oering up a top-level

ramework or classiying systems on the

basis o their dynamics - see Figure 8.

MANY SYSTEMS ARE NONLINEAR

Why are nonlinear systems so much harder to analyze than linear ones? The essential

dierence is that linear systems can be broken down into parts. Then each part can be

solved separately and nally recombined to get the answer. This i dea allows a antastic

simplication o complex problems ... In this sense, a linear system is precisely equal to

the sum o its parts. But many things in nature dont act this way. Whenever parts o a

system interere, or cooperate, or compete, there are nonlinear interactions going on.

Most o everyday lie is nonl inear, and the principle o superposit ion ails spectacularly.5

Steven Strogatz


14/32



2726

A NEW INTERVENTION MINDSET

In the absence o a calculus o complexityenabling us to perectly predict-and-control

reality we need some other mechanismor eectively deploying Meadows twelve

leverage points in a complex nonlinearsystems environment - or using her

heuristics (i.e. rules o thumb) as a point odeparture or systems intervention.

Tim Harord oers some insights in his bookAdapt - Why Success Always Starts With

Failure.

He sets out three Palchinsky principles,7

named ater the Russian engineer who

ormulated them:

(1) VARIATION: Seek out new ideas

and try new things

(2) SURVIVABILITY: When trying

something new, do it on a scalewhere ailure is survivable

(3) SELECTION: Seek out eedback

and learn rom your mistakes as

you go along

There is an important role or prediction in

this approach - some attempt at oresight -but with Meadows major caveat that:8

(A) MODELS: All our predictions aremodels o the world. None o

these models is the real world.

(B) CONGRUENCE: Our models may

well have a strong congruencewith the world and oten do.

(C) LIMITATIONS: Howeverconversely our models also all

short o representing the worldully.

This led statistician George Box to proclaim30 years ago:

All models are wrong, but some are

useul9

It is when we couple prediction with theillusion o perect control - that they are

reality - that we run into problems.

As Meadows puts it:

Systems cant be controlled, but they

can be designed and redesigned. We

cant surge orward with certainty into a

world o no surprises, but we can expectsurprises and learn rom them and even

prot rom them. We cant impose our

will on a system. We can listen to what

the system tells us, and discover h ow

its properties and our values can work

together to bring orth something much

better than can ever be produced by our

will alone.10

FIGURE 7 - THE LOGISTIC MAP

An illustration o how complex nonlinear behaviour can result rom a very simple starting point

Source: Melanie Mitchell, Complexity

R=2 R=4

Melanie Mitchell elaborates on the complex

nonlinear behaviour o The Logistic Map:

The logistic map is an extremely

simple equation and is completely

deterministic: every xt maps onto one

and only one value o xt+1. And yet the

chaotic trajectories obtained rom thismap, at certain values o R, look very

random - enough so that the logistic

map has been used as a basis or

generating pseudo-random numbers on

a computer. Thus apparent randomness

can arise rom very simple deterministic

systems.11

It is a simple illustration o how complex

systems conound our expectations about

the relationship between action andresponse - between input and output /

cause and eect.

Change the value o R in the logistic map

rom 2 to 4 and the systems behaviour

changes radically - it ips to a newbehavioural pattern (in act this happens

somewhere between R=3.4 and R=3.5). Thissystem is complex, nonlinear, unpredictable.

A uller simulation o its dynamics can beviewed at:

http://en.wikipedia.org/wiki/Logistic_map

You think that because you understand

one that you must thereore understand

two because one and one makes two. But

you orget that you must also understand

and6

Meadows recounting

a Su teaching story
http://en.wikipedia.org/wiki/Logistic_maphttp://en.wikipedia.org/wiki/Logistic_map


15/32



2928

Growth,decay,or

equilibrium

Oscillations

Exponentialgrowth

RCcircuit

Radioactivedecay

Linearoscillator

Mass

andspring

RLCcircuit

2-bodyproblem

(Kepler,Newton)

Fixedpoints

Bifurcations

Overdampedsystems,

relaxationaldynamics

Logisticequation

forsinglespecies

Pend

ulum

Anha

rmonicoscillators

Limitcycles

Biolo

gicaloscillators

(neurons,heartcells)

Predator-preycycles

Nonlinearelectronics

(van

derPol,Josephson)

Civilengineering,

structures

Electricalengineering

Strangeattractors

(Lorenz)

3-bodyproblem(Poincar)

Chemicalkinetics

Iteratedmaps(Feigenbaum)

Fractals

(Mandelbrot)

Forcednonlinearoscillators

(Levinson,Smale)

Chaos

Practicalusesofchaos

Quantumchaos?

Coupledharmonic

oscillators

Solid-statephysics

Moleculardynamic

s

Equilibriumstatistical

mechanics

Collectivephenom

ena

Continuum

Wavesandpatterns

Elasticity

Waveequations

Electromagnetism

(Maxwell)

Quantummechanics

(Schdinger,Heisenberg,Dirac)

Heatanddiusion

Acoustics

Viscousuids

Couplednonlinear

oscillators

Lasers,nonlinearoptics

Nonequilibriumsta

tistical

mechanics

Nonlinearsolid-statephysics

(semiconductors)

Josephsonarrays

Heartcellsynchron

ization

Neuralnetworks

Immunesystem

Ecosystems

Economics

Spatio-temporalcomplexity

Nonlinearwaves(shocks,solitons)

Plasmas

Earthquakes

Generalrelativity(Einstein)

Quantum

eldtheory

Reaction-diusion,

biologicalandchemicalwaves

Fibrillation

Epilepsy

Turbulentuids(Navier-Stokes)

Life

n=1

n=2

n3

n>>1

Numberofvariables

Linear

Nonlinear

Nonlinearity

Thefrontier

FIGURE 8 - A FRAMEWORK FOR CLASSIFYING SYSTEMS

One axis tells us the number o variables needed to characterize the state o the system.

The other axis tells us whether the system is li near or nonlinear.

Source: Steven H. Strogatz, Nonlinear Dynamics and Chaos

Steven Strogatz proposes a ramework or

classiying systems:

Admittedly, some aspects o the picture

are debatable. You might think that

some topics should be added, or placed

dierently, or even that more axes are

needed - the point is to think about

classiying systems on the basis o their

dynamics.

... Youll notice that the ramework also

contains a region orbiddingly marked

The rontier. Its like in those old maps

o the world, where the mapmakers

wrote ,Here be dragons on the

unexplored parts o the globe. These

topics are not completely unexplored o

course, but it is air to say that they lie

at the limits o current understanding.

The problems are very hard, becausethey are both large and nonlinear.

The resulting behaviour is typically

complicated in both space and time,

as in the motion o a turbulent fuid or

the patterns o electrical activity in a

brillating heart .12

Note that systems such as economics andecosystems all beyond this rontier.


16/32



3130

In other words, we need to embrace a ar

more iterative approach to problem solving

i we are to learn to dance with increasinglycomplex systems in the 21st Century.

This means taking small steps, predicting,

but monitoring reality constantly and being

willing to change the course as we nd outmore about where it is leading.

We have to abandon the illusion o predict-and-control and adopt a new trial-and-

error approach.

It is a dierent approach to problem solving.

Meadows again:

The trick, as with all the behavioural

possibilities o complex systems, is to

recognize what structures contain whichlatent behaviors, and what conditions

release those behaviors - and, where

possible, to arrange the structure and

conditions to reduce the probability o

destructive behaviors and to encourage

the possibility o benecial ones.13

In other words, try to coax the system intoproviding the solution - try to evolve the

solution, not determine it.

Frame the question. Provide the right tools

or it to yield the answer. Try to releasebenecial system behaviours. Correctly

structured, a system will arrive at thesolution.

Put simply, we need a undamentally newparadigm or problem solving. We need

to adopt a new Complex Systems mindset

based on three tenets:

(1) EMBRACE UNCERTAINTY

(2) USE TRIAL-AND-ERROR

(3) PROVIDE TOOLS

Recall that changing the paradigm is thesecond most eective place to intervene in

a system - point eleven in Meadows twelve

point list. For this is the mindset out owhich the entire system arises and the basis

or regulating it.

It has concrete implications. For example,

individuals and organisations are held liablein courts o law or system behaviours based

on the presumption that they can predict-and-control these complex nonlinear

systems. This can result in multi-millioneuro nes or those accused and ound

guilty. Think o the litigation surroundingcountereit items.

Similarly, Inormation Society ServiceProviders (ISSPs) may use probabilistic

techniques, such as rankings, to try

and help consumers with oresight in acomplex retail environment. They are not

providing assurances that these are perectpredictions. Yet we are seeing lawsuits

against ISSPs based on claims to the latter.

I can live with doubt, and uncertainty, and

not knowing. I think its much more inter-

esting to live not knowing than to have

answers which might be wrong.

Richard Feynman

Author Robert Pirsig captures the

importance o prevailing paradigms when

he states:

I a actory is torn down but the

rationality which produced it is let

standing, then that rationality will

simply produce another actory. I a

revolution destroys a government, but

the systematic patterns o thought

that produced that government are let

intact, then those patterns o thought

will repeat t hemselves... Theres so much

talk about the system. And so little

understanding.15

The longer we cling to a 19th Century

reductionist view o a clockworkNewtonian universe that is ticking along

a perectly predictable path, the more

REALITYMODEL

Everything we think

we know about the

world is a model. Our

models have a strong

congruence with the

world.

... BUT NOT ALWAYS which is

why we encounter surprises.

For our models fall far short

of representing the real

world fully. Think of the

recent failure to understand

risk modeling in nance.

Congruent

REALITY

MODEL

Incongruent

FIGURE 9 - UNDERSTANDING MODELS

Models oten have a strong congruence with reality but not always

Source: Carr & Strauss drawn up rom Thinking in Systems by Donella Meadows14


17/32



3332

trouble we are going to get into - be that

in our environmental, economic, politicaland social systems or our system o

globalisation.

Disciplines such as physics have already had

to undergo this change in mindset - albeitthat they began 100 years ago.

It was in Parc Leopold adjacent to theEuropean Parliament that Einstein and

Bohr debated this change in mindset at theamous Solvay Physics Conerence o 1927.

Playing out at the beginning o the 20th

Century and known as the Einstein-Bohr

debates, the arguments centred overwhether the universe evolves in an arbitrary

or deterministic way.

Einstein argued that God does not playdice, Bohr the opposite, on one occassionanswering, Einstein, stop telling God what to

do.

Todays prevailing scientic consensus

is that God (i.e. nature) does play dice -perect prediction la Laplace is impossible.

And they act in this knowledge daily.

Chris Pollard, Member o the Royal College

o Veterinary Surgeons, cites the example oadministering anaesthetics:

With anaesthetic drugs we know what

they can do, what side eects t hey

may have in certain situations, which

situations they may be most eective

in. However, or the most part, the

most elusive element is determining

exactly how they really work. We do

not control the physiological system.

We do not dictate what happens. We

try to dance with the system. We add

inputs to the system. We monitor. We

make adjustments. But we do not have

the nal say over what happens. Its

a nonlinear way o dealing with the

system. Not a linear one. I can look at

two identical cases. Treat them exactly

the same and have two dierentresults.16

This approach ts with Strogatzs rameworkor classiying systems. Recall that he

places lie beyond the rontier o current

understanding in gure 8.

Indeed the entire modern scientic method

is a constant process o alsication - o trial-and-error.

The social sciences mindset needs to catch

up with the natural sciences mindset.

It is a case o C.P. Snows Two Cultures.

In his book The Two Cultures and the

Scientic Revolution, Snow contends thatthe breakdown o communication between

the Two Cultures o modern society thesciences and the humanities is a major

hindrance to solving the worlds problems.

As Meadows puts it, our challenge is one o :

Linear minds in a nonlinear world17

FIGURE 10 - A CENTURY AGO PHYSICISTS WERE ALREADY CHANGING THEIR MINDSET

The prevailing scientifc consensus is that perect prediction turns out to be impossible

Source: Institut International de Physique de Solvay, Parc Leopold

Note:

1) For those interested, the physicist Stephen Hawking expands on the debate at:

http://www.hawking.org.uk/index.php/lectures/64

2) The picture above was taken on the steps o what is today the Lyce Emile Jacqmain in Parc

Leopold, Brussels
http://www.hawking.org.uk/index.php/lectures/64http://www.hawking.org.uk/index.php/lectures/64


18/32



3534

A NEW INTERVENTION MECHANISM

How then do interventionists put this new

Complex Systems mindset into concrete

action in a system?

Here it is instructive to draw on the work oJohn Boyd, a military strategist, US Airorce

ghter pilot and Colonel. Boyd helped

design and champion the F-16 ghter jet.He ormulated the OODA Loop18 to

enable ghter pilots to quickly assess andadapt to complex and rapidly changing

environments - ones that cannot becontrolled. In which poor decision-making

can be a matter o lie and death. Today theloop is written into US Air Force doctrine.

Taken at its simplest level, Boyd contendsthat to prevail up in the air a ghter

pilot needs to be more eective than hisopponent at continually cycling through

our processes:

(1) OBSERVE: Gather sensory inputs

rom the environment - monitorand intelligence gather.

Radio chatter inorms me that

there is an unidentied aircrat in

my airspace.

(2) ORIENT: Make sense o this

observational data by creating amental picture - a model - o the

situational reality.

....The aircrat is not yet in range

but it would already be useul to

try to predict the likely identity o

the pilot, his nationality, level o

training, etc.

... Now that the aircrat is in radar

contact the speed, size, and

manoeuvrability o the enemy

plane has become available. I

have some real inormation to go

on and unolding circumstances

can take priority over my earlier

predictions.

... The aircrats intentions appear

hostile.

(3) DECIDE: Use this newknowledge as the basis or

decisions.

... Time or evasive action.

... I am going to get into the sun

so that my opponent cannot get a

clear visual on me.

(4) ACT:Translate this into action.

... Pull back on the j oystick and let

my aircrat cl imb.

Back to the process o observation:

Is the other plane reacting to my

change o altitude?

Then to orient:

Is he reacting characteristically

- predictably - or perhaps instead

acting like a noncombatant?

Is his plane exhibiting better-than-

expected perormance?

Do I need to revisit my

assumptions?

Are they congruent with the

complex reality that I am

observing?

And so the cycle continues.

We nd similar decision making loops in

other disciplines. For example, the Sense-Model-Plan-Act Loop in the eld o Ar ticial

Intelligence.

Indeed in sotware development it is akin toAgile Programming. Agile Programming is

a methodology or quickly and eectivelycycling through sotware development.

It supersedes an earlier linear design

approach called the Waterall Model.

Importantly the 3 Palchinsky Principles oVariation, Survivability and Selection can

OBSE

RVE

ORIENTDE

CID

EACT

INTERVENTION

MINDSET

Gather sensory

inputs from the

environment

Make sense

of this

situational

reality

Translate

these decisions

into action

Use this new

knowledge as

the basis for

decisions

FIGURE 11 - CYCLE THE OODA LOOP

A practical intervention mechanism or complex nonlinear systems

Source: Carr & Strauss expanding on John Boyds concept o the OODA Loop


19/32



3736

be integrated into the OODA Loop.

As we cycle through the loop we can

seek out new ideas and try new things(Variation).

Providing we cycle through the loop

iteratively, then when we do attempt

something new we can act on a scale whereailure is survivable (Survivability).

And as we act the OODA Loop is designedto seek out eedback and learn rom our

mistakes as we go along (Selection).

In other words the OODA Loop is a practical

trial-and-error mechanism or dancing witha complex systems.

Thereore i we wish to optimisesystems intervention in or example, the

consumption system, we need to know how

to cycle most eectively through the OODALoop.

INTERVENTION AGENTS & THE

INTERVENTION MIX

To successully address this questionwe need to examine the strengths and

weaknesses o those that have the potentialto cycle through the loop, namely the

Inter vention Agents. Then we can decidewho should intervene.

Should individuals be the interventionist?

The wisdom o the crowd?

Experts or non-experts?

Computer algorithms?

A combination o all the above - an

Intervention Mix?

In act we live with intervention mixes

everyday. Think o when you apply thebrakes whilst driving your car. It is your oot

pressing the brake pedal, but it is also thecars anti-lock braking system at work - a

computer. It is the combination o thesetwo agents that brings a ast moving car to

a smooth stop at the trafc lights. A mix o

man and machine.Again these are not abstract issues.

At this very moment we are trying to

determine how ar we can automate HighFrequency Trading? Can we leave the

sotware algorithms to their own devices ordo we need human intervention to avoid

events such as the May2010 Flash Crash?

Similarly a debate ensues over whether

Non-Expert Expert

Indivi

dual

Crowd

People have common sense- in more formal terms,sensitivity to context -

computers do not. This is alsothe domain of personal choice

Experts come into theirown in the area betweenrote rule following and

probabilistic prediction- an area in which a

combination of knowledgeand initiative is required

Groups of experts willoutperform most, if not all,

individual experts

Algorithms

People can be replaced bycomputers when it comes tomaking certain types of

rule-based decisions if therequisite hardware, softwareand data exists and it makeseconomic sense to do so

AGENT TYPE

AGENTCONFIGURATION

EACH OF THESE

SQUARES IS A

POTENTIAL

INTERVENTION AGENT

To gure out the value ofsomething its best to take

an average of a groupsanswers. For choosing theright answer from a small

number of possiblealternatives majority opinionserves us better. The rightconditions need to be in place

to take advantage of eithermethod

FIGURE 12 (opposite page) - THE SIMPLIFIED INTERVENTION MIX IN 2D

Interaction can emanate rom average people, experts or computers and can be confgured in

dierent orms, either in the orm o individuals or o wisdom o crowds

Source: Carr & Strauss - the human decision making components o this diagram are assem-

bled principally rom Len Fishers explanation in The Perect Swarm19


20/32



3938

ISSPs can completely automate countereit

items o o their platorms?

As we will see, no intervention agent is the

panacea - expert, wisdom o crowds or

computer algorithm. Each has its strengthsand weaknesses depending on the specic

circumstance. For example, we are all tooamiliar with the madness o crowds and

with mob mentality in certain situations.

THE WISDOM OF CROWDS

What ollows on the wisdom o crowds

versus that o experts is in large partdrawn rom the explanation provided by

Len Fisher, a Physicist at the University oBristol, in his book The Perect Swarm - The

Science o Complexity in Everyday Lie.21

For problems that involve guring out the

value o something (like a compass bearing

or the classic case o the number o beans in

a jar) the best way to tackle the problem is

to [employ group decision making and to]

take an average o all the answers. Sci entists

call these state estimation problems. For

problems that involve choosing the right

answer rom among a small number o

possible alternatives, a groups majority

opinion serves us better. To take the best

advantage o either, we need to ull just th ree

conditions:

(1) DIVERSITY: The people in the

group must be willing and able

to think or themselves and reach

diverse, independent conclusions.

(2) VERIFICATION:The question

must have a denite answer that

can ultimately be checked against

reality.

(3) ALIGNMENT: Everyone in the

group must be answering the

same question. ( This may seem

obvious, but it is oten possible

or people to interpret the same

question in very dierent ways.)

When these 3 conditions are ullled, the

mathematics o complexity leads to three

astounding conclusions:

(1) STATE ESTIMATION:When

answering a state estimation

question, the group as a whole

will always outperorm most o its

individual members.23 Universityo Michigan complexity theorist,

Scott Page, proves this with histheorem that demonstrates that

our collective error as a group has

to be smaller than our average

individual error because o the

diversity o our answers.24 AsPage puts it, being dierent is asimportant as being good.25

(2) MAJORITY OPINION:Imost o the group members

are moderately well-inormed

about the acts surrounding

a question to which there are

several possible answers (but only

one correct one), the majority

opinion is almost always bound

to be right. I each member o a

group o one hundred people has

a 60 percent chance o getting

the right answer, or example,

then a rigorous mathematical

ormulation [Condorcets Jury

Theorem] proves that the answer

o the majority has a better than

99 percent chance o being the

correct one.

(3) VALIDITY:Even when only a

ew people in the group are well-

inormed, this is usually sucient

or the majority opinion to be the

right one.26

EXPERTS AND GROUPS OF EXPERTS

Fisher cautions that, There is just one caveat,which is to remember th at Pages theorem

proves only that the group outperorms

most o its members when it comes to state

estimation problems. It does not necessarily

outperorm all o them. I there is an

identiable expert in the group, it may be that

they will do better than the group average ...

Thats not to say that experts always do better

than the average.

PAGES THEOREM ON STATE ESTIMATION

collective error = average individual error - prediction diversity

The calculations are slightly tricky because statisticians use squared values o errors.

But the message is simple ... the theorem shows that our collective error as a group has

to be smaller than our average individual error because o the diversity o answers. The

crowd perorms better than most o the people in i t.20

Len Fisher, The Perect Swarm

CONDORCETS JURY THEOREM ON MAJORITY OPINION

The theorem in its simplest orm says that i each member o a group has a better than

50:50 chance o getting the right answer to a question that has just two possible answers,

then the chance o a majority verdict being correct rapidly becomes closer to 100 percent

as the size o the group increases. Even i each individual has only a 60 percent chance o

being right, the chance o the majority being right goes up to 80 percent or a group o 17

and to 90 percent or a group o orty-ve.22

Len Fisher, The Perect Swarm


21/32



4140

A collection o experts can also outperorm

most, i not, all o the individual experts.

Page gives the example o a group o ootballjournalists orecasting the top dozen picks

in the 2005 NHL drat. Not one o them

perormed nearly as well as the average o all

o them ...

According to Michael Mauboussin, a

Proessor o Finance at Columbia Business

School, experts come into their own in

the area between rote rule ollowing and

probabilistic prediction - an area in which a

combination o knowledge and initiative is

required.27

That is why we continue to have expertsin policymaking, experts in sotware

development, experts in the enorcement otrademark rights etc.

When we dont have an expert available, we

must all back on the di versity o the group.28

COMPUTERS VERSUS COMMON SENSE

Fisher observes that, Experts are also

being replaced in many areas by computer

algorithms when it comes to making rule-

based decisions.29 An algorithm is a

series o steps that transorms an input

into an output - in other words, a deniteprocedure or a recipe.30

Back to Melanie Mitchell who notes that:

Computer controlled vehicles cannow drive by themselves across

rugged desert terrain. Computer

programs can beat human doctors at

diagnosing certain diseases, human

mathematicians at solving complex

equations, and human grand masters

at chess. These are only a ew examples

o a surge o recent successes in articial

intelligence (AI) that have brought a

new sense o optimism in the eld.31

But she also draws attention to the act thatdespite this optimism we need to remain

grounded and not get ahead o ourselves:

There are a ew human-level things

computers still cant do, such as

understand human language, describe

the content o a photograph, and more

generally use common sense... Marvin

Minsky, a ounder in the eld o articial

intelligence, concisely described this

paradox o AI as, Easy things are hard.

Computers can do many things that

we humans consider to require high

level intelligence, but at the same time

they are unable to perorm tasks that

any three-year-old child could do with

ease.32

We requently experience the act thatcomputers lack common sense, or more

ormally in computer science terms,

sensitivity to context.33

Duncan Watts, Principal Research Scientist

at Yahoo, provides a denition o commonsense in his book Everything Is Obvious* -

*Once You Know The Answer:

Roughly speaking, it is the loosely

organized set o acts, observations,

experiences, insights, and pieceso received wisdom that each o us

accumulates over a lietime, in the

course o encountering, dealing

with, and learning rom, everyday

situations.34

He states that:

It is largely or this reason, in act, that

commonsense knowledge has proven so

hard to replicate in computers - because,

in contrast with theoretical knowledge,

it requires a relatively large number o

rules to deal with even a small number

o special cases.35

Mitchell again:

My computer supposedly has a state-

o-the-art spam lter, but sometimes

it cant gure out that a message witha word such as V!a&@ is likely to be

spam.36

Most people however see V!a&@ or what

it is instantly. That is why most spam ltersdraw extensively on pooling inormation

eedback rom real users. I 1,000 peopleclick on the This is Spam button when they

encounter V!a&@ then the computer cancatalogue it as likely to be spam.

FIGURE 13 (opposite page) - THE RISE OF THE MACHINES

Algorithmic trading has rapidly taken o

Source: Aite Group diagrams reproduced by The Economist, February 25th 2012


22/32



4342

This lack o sensitivity to context is also thereason why web users are on occasion asked

to retype a code - see Figure 14 below.This code cannot be read by a computer

because it lacks the common sense - theintuition - to tell that it says MY5N5. The

aim in this CAPTCHA code example is toprevent automated programmes (i.e. spam

in this instance) rom requesting access

to some piece o inormation - it is just orhuman consumption - those agents that

can discern the text.

It is or this reason, as well as or legal,ethical and moral considerations, that

military drones are remote controlled andare not ully autonomous. At least or now

humans still make the decision to re.

Johann Borenstein, head o the Mobile

Robotics Lab at the University o Michiganobserves that:

Robots dont have common sense and

wont have common sense in the next 5 0

years, or however long one might want

to guess.37

Consequently he believes that human skillswill remain critical in battle ar into the

uture.

And there are additional limits to

computing power, namely:

(1) THEORETICAL LIMITATIONS: In

the 1930s Kurt Gdel and Alan

Turing quashed the hope o theunlimited power o mathematics

and computing. Gdel provedthat mathematics is either

inconsistent or incomplete.Building on this discovery Turing

proved that the answer to theEntscheidungsproblem(decision

problem) is no - i.e. not everymathematical statement has

a denite procedure that can

decide its truth or alsity. Hedemonstrated this act via theHalting problem. In short, thereare certain classes o problem

that cannot be computed.39

Moreover even when problems

can be computed it may bevery time consuming to do

so. For instance, to date no

efcient algorithm is known oran important class o problems

known as NP-complete. Thesesearch problems include

scheduling, map colouring,protein olding, theorem

proving, packing, puzzles, thetravelling salesman and many

more such problems. In actthe Clay Mathematics Institute

classies the NP-complete

problem as one o the sevenmost important questions to

be answered this century and isoering $1,000,000 dollars to the

rst person that does so.

(2) HARDWARE AND SOFTWARE

LIMITATIONS: Theoretical

limitations aside, there arehardware and sotware

limitations to consider. Yes,computer processing power

is increasing exponentiallyand becoming less and

less expensive (a process

encapsulated by Moores Law).However as virtual reality

pioneer Jaron Lanier notes in hisbook You Are Not A Gadget,sotware development doesnt

necessarily speed up in sync with

improvements in hardware. Itoten instead slows down as

computers get bigger because

there are more opportunities

or errors in bigger programs.

Development becomes slower and

more conservative when there is

more at stake, and that is what is

happening.40

FIGURE 14 (below) - THE LIMITS OF COMPUTATION

Computers are unable to perorm some tasks that humans fnd trivial

Source: Drupal.org

THE NP-COMPLETE PROBLEM

I it is easy to check that a solution to a problem is correct, is it also easy to solve the

problem? This is the essence o the P vs NP question. Typical o the NP problems is that o

the Hamiltonian Path Problem: given N cities to visit, how can one do this without visiting

a city twice? I you give me a solution, I can easily check that it is correct. But I cannot so

easily nd a solution.38

Clay Mathematics Institute


23/32



4544

(3) DATA LIMITATIONS: Eric

Schmidt, Executive Chairman oGoogle, points out that, Fromthe dawn o civilization until

2003, humankind generated

ve exabytes o data. Now we

produce ve exabytes every

two days... and the pace is

accelerating.41 As a result

there is a lot excitementsurrounding the promise o

what is being termed, Big

Data - the ability to cycle theOODA loop in unprecedented

ways in almost every walko lie due to advances in

technology. Nevertheless thereare signicant challenges to

be addressed, not least thosepertaining to privacy. Simon

Szykman, Chie Inormation

Ofcer o the US CommerceDepartment, has outlined his

top nine Big Data challenges.42These are: data acquisition;

storage; processing; datatransport and dissemination;

data management and curation;archiving; security; workorce

with specialized skills, and; thecost o all o the above.

(4) ECONOMIC LIMITATIONS:Sometimes it simply makes

no nancial sense to try to

develop complex sotwarealgorithms to solve a problem

that a human can solve instantlyand eortlessly. Lanier notes

that until recently, computerscouldnt even recognize a

persons smile.43

As a result, leading articial intelligence

researchers, such as Proessor Hans Moravecrom the Robotics Institute at Carnegie

Mellon University, acknowledge thatComputers have ar to go to match human

strengths.44

Lanier recalls that Beore the crash, bankersbelieved in supposedly intelligent algorithms

that could calculate credit risks beore makingbad loans.45

Indeed Lanier is careul to draw the

distinction between ideal and realcomputers.46

Nevertheless this is not to say that commonsense approaches do not have their own

shortcomings.

Watts identies three limitations in common

sense reasoning:47

(1) INDIVIDUAL BEHAVIOUR: Our

mental models systematicallyall short o capturing the

complexity o what drivesindividual behaviour.

(2) COLLECTIVE BEHAVIOUR: Ourmental models o collective

behaviour ail to take account othe act that collective behaviouris greater than the sum o its

parts.

(3) LEARNING FROM HISTORY:

We learn less rom the pastthan we think we do and this

misperception in turn skews ourperception o the uture.

Non-ExpertExpert Individual

CrowdAlgorithms

Public

Sector

Private

Sector

AGENT

CONFIGURATION

AGENTORIGIN

AGENTTYPE

EACH OF THESE

CUBES IS A POTENTIAL

INTERVENTION AGENT

FIGURE 15 - THE FULL INTERVENTION MIX IN 3D

As well as there being dierent intervention t ypes and confgurations, intervention can also

originate rom the private and public sectors

Source: Carr & Strauss

Watts concludes that:48

Commonsense reasoning, thereore,does not suer rom a single

overriding limitation but rather rom

a combination o limitations, all o

which reinorce and even disguise one

another. The net result is that common

sense is wonderul atmaking sense

o the world, but not necessarily at

understanding it .

PUBLIC VERSUS PRIVATE SECTOR

INTERVENTION

We have covered the strengths and

weaknesses o the various InterventionAgents. We now need to introduce the third

dimension to the Intervention Mix, namely

the division between public and privatesector intervention.

For both sectors can deploy all o theseagents. Think o general elections and

reerendums (wisdom o crowds),


24/32



4746

expert committees and groups o wise

men (groups o experts), as well asrepresentatives elected to political ofce

(oten non-experts) and online tax orms orspeed cameras (computer algorithms).

Taken to its logical conclusion we can andshould extend this approach to include

other sectors, such as civil society (i.e. NGOs,

Trade Unions, Think Tanks, Academia, Media,Citizens, etc.). However or the sake o timeand simplicity we will stick here to just two

sectors.

So can we delineate between whether we

need:

(A) Public sector intervention

(B) Private sector intervention

(C) A mix o public and private

sector intervention

To even begin to make this call we need to

return to the OODA Loop and ask:

Knowing the strengths and weaknesses

o the intervention agents that we have

at our disposal - the various types and

congurations ...

And aware that we can deploy eitherpublic or private sector intervention or

both ...

How do we use intervention agents to

cycle most eectively through the OODA

Loop?

And we need to rame this question in the

context o a specic issue.

A generic debate over the merits o more

or less government intervention - oten inthe orm o regulations, subsidies or scal

incentives - will serve no purpose.

There is no o-the-shel answer.

So let us try to answer this question when

we seek to intervene in a specic system:

How do we use intervention agents tocycle most eectively through the OODA

Loop?

A NEW INTERVENTION PRINCIPLE

And when answering this question let ustry to invoke a new Intervention Principle

to guide us:

An intervention agent is to intervene

only i, and in so ar as, it is reasonably

oreseeable that the objectives o the

proposed intervention cannot better be

achieved by the system running itsel or

in deault o this by another agent49

The objective should be as littleintervention as possible. Reerring back to

Donella Meadows:

Aid and encourage the orces and

structures that help the system run

itsel. Notice how many o those orces

and structures are at the bottom o

the hierarchy. D ont be an unthinking

intervenor and destroy the systems own

sel-maintenance capabilities. Beore

you charge in to make things better, pay

attention to the value o whats already

there.50

FIGURE 16 - SMARTER INTERVENTION

A new intervention mindset, mechanism & principle - in short a new paradigm or intervention

Source: Carr & Strauss

STOP?

OBSE

RVE

O

RIENTDECIDE

ACT

INTERVENTIONMECHANISM

INTERVENTION

PRINCIPLE

SMARTER INTERVENTION

INTERVENTION

MINDSET


25/32



4948

As she points out:

Any system, biological, economic

or social that gets so encrusted that

it cannot sel-evolve, a system that

systematically scorns experimentation

and wipes out the raw material o

innovation, is doomed over the long

term on this highl y variable planet.51

For:

The ability to sel-organize is the

strongest orm o system resilience. A

system that can evolve can survive

almost any change, by changing itsel.52

Note the I ntervention Principles holisticapproach to intervention. I t applies to any

Intervention Agent, regardless o its type,conguration or the sector rom which it

originates.

Also observe that the Intervention Agent

has to be able to make a Reasonable caseor why they are intervening. Why are they

not leaving the system to sel-organise- to run itsel? And i intervention must

take place why not deer to the actions o

another Intervention Agent?

The principle employs the wordForeseeable because we have somepointers to help us to intervene - we

have the OODA Loop and we know thestrengths and weaknesses o the various

Intervention Agents, their dierent typesand congurations - but nothing is certain

in a complex nonlinear systems world.

Finally, when we do intervene we need to

take into consideration:

(1) TEMPO: Can the InterventionAgent cycle through the OODA

Loop at the right tempo?

Intervening too ast or too slowcan be detrimental, leading

to oscillations in the systemsbehaviour. To paraphrase the

words o Donella Meadows,intervention needs to t with

the beat o the system.53

(2) ACCOUNTABILITY: What will

ensure that the InterventionAgent can be held accountable

as it cycles through the OODA

Loop? Will it always be possibleto hold Agents accountable

in systems whose behavioursare greater than the sum o

their parts? This has signicantimplications or our current

liability regimes.

(3) CONSISTENCY: How do we

avoid silo approaches tointervention whereby one

Intervention Agents actionundermines those o another?

Instead all o these actions need

to reinorce one another. They

need to act in concert with eachother - to interace successullywith one another. In other

words, the various InterventionAgents need to cycle around the

loop as a team.

(4) BIAS: Mary Cummings,

Director o the MIT Humansand Automation Lab, draws our

attention to the phenomenon o

automation bias.54 She denes

this as the tendency to trustan automated system, in spite

o evidence that the systemis unreliable, or wrong in a

particular case.


26/32


27/32


5352

made by starting with the digital

code in the computer, building

the chromosome rom our

bottles o chemicals, assembling

that chromosome in yeast,

transplanting it into a recipient

bacterial cell and transorming

that cell into a new bacterial

species. So this is the rst sel-

replicating species that weve had

on the planet whose parent is a

computer. It also is the rst species

to have its own website encoded

in its genetic code.3 Venter has

likened the advance to making

new sotware or the cell. Hebelieves that I we can really get

cells to do the production that

we want, they could help wean

us o oil and reverse some o the

damage to the environment by

capturing carbon dioxide.4

(2) MORALITY: To what extent,are we going to be able encode

morality into our algorithms?Or are we still going to need to

mix human intervention intothe OODA loop to guarantee

ethical decisions and actions -

OTHER IDEAS TO WATCH

Avatar assistants; Biomimicry; Clean coal; Comort eating; Contextual deficit;

Diminishing use o email; Decline o voice communication; Electrification o transport;

Facial recognition on mobile phones; Gene hacking; Holographic telepresence; Increasing

complexity; Local living; Mobile money; Peak water; Peer-to-peer lendin g/giving;

Quantum computing; Reverse migration; Sel-tracking; Smart inrastructure; Slow

education; Shit rom products to experiences; Ultra-ecient solar; Value redefinition;

Voluntary simpli city.2

Richard Watson

ones instilled with empathy and

compassion? As military orcesaround the world increasingly

deploy drones they are havingto grapple with this issue. Can

or should armed robots be givenully autonomous capabilities

in combat? Various approacheshave been proposed, such

as Ronald Arkins EthicalGovernor.5 Arkin is a roboticist

at the Georgia Institute o

Technology and he suggestsa 2-step decision procedure

beore pulling the trigger- i.e.an algorithm. However a report

coauthored by Human RightsWatch and Harvard Law School

contends that, ully autonomousweapons would be incapable

o meeting international

humanitarian law standards,

including the rules o distinction,

proportionality, and military

necessity. These robots would

lack human qualities, such as the

ability to relate to other humans

and to apply human judgement

that are necessary to comply

FIGURE 18 (opposite page) - CRAIG VENTERS SYNTHETIC CELL

The synthetic cell l ooks identical to the wild ty pe and extends our concept o computation

Source: BBC Online News sourcing the picture rom Science, 20 May 2010

FIGURE 19 (above) - MORALS AND THE MACHINE

Do humans need to remain in the OODA Loop to ensure ethical decision making?

Source: The Economist, June 2nd-8th 2012


28/32


5554

with the law.6 As a result, SteveGoose, Director, Arms Division,

Human Rights Watch, argues

that, ... you must have meaningul

human control over key battleeld

decisions o who lives and who

dies. That should not be let up

to the weapons system itsel .7 Inother words, when it comes to

the question o armed drones,then humans need to remain in

the OODA loop. An InterventionMix is required.

(3) HUMANITY: Rapid advancesin technologies, such as

prosthetics, are increasingly

blurring the line betweenman and machine. Take the

example o Nigel Ackland and

his bebionic3 prosthetic hand.It operates by responding toNigels muscle twitches. He

congures its unctionality- grips, thresholds and radio

requency - by using sotware

that runs on Microsot Windows.Nigel states that, Having abebionic hand is like being human

again, psychologically I wouldnt

FIGURES 20 and 21 - THE INCREASING CONVERGENCE OF MAN AND MACHINE

Nigel Ackland using his bebionic3 prosthetic hand made by RSLSteeper (below). Ray Kurweils

controversial vision o The Singularity (bottom)

Sources: YouTube video o Nigel Ackland, RSLSteeper technical inormation manual or bebionic3,and The Singularity by Ray Kurzweil

We are made wise not by the recollection

o our past, but by the responsibility o our

uture.

George Bernard Shaw

be without it. I can hold the

phone, shake hands and wash

my let hand normally, which

I havent been able to or ve

years!8 Ray Kurzweil, Directo

Documents

Smarter Intervention in Complex Systems