“You could not remove a single grain of sand from its place without thereby… changing something throughout all parts of the immeasurable whole”.
— Johann Gottlieb Fichte, The Vocation of Man (1799)
Welcome to the second of the 12 themes which I’ll be writing about this year (and into next, quite obviously, given my exceptionally slow progress to date)—networks. We’ll entangle ourselves with the topic of networks in a moment; but again, I must reiterate some housekeeping from my previous piece: The overarching context within which these 12 themes will fit—in case you haven’t yet read it, I strongly encourage you to visit it here — is the application of natural and human sciences including (but not limited to) complexity, networks, evolution, ecology, anthropology and architecture, to guide system-level organisational change, via the enablement of agency (noting that agency will be the fourth topic I’ll write about).
Consider these writings as antithetical to the traditional leadership and change management fads that most people are familiar with. These pieces are as long and as technical* as they need to be, and because you’ve signed up for them there is an inherent assumption that you’re up for the challenge of thinking and learning about organisational strategy, management and change from an alternative and somewhat counter-cultural perspective. They are not easy pieces to consume.
I’m sequencing the 12 themes for discussion from levels of lower to higher order as follows:
I signed-off last time by suggesting that if you’d like to have agency in the world, you could do worse than taking the time to understand the basics of network science. Although a relatively new science of less than three decades maturity (which is why I refer to the science as contemporary), it’s important to remember that science itself is a human construct—just because we’ve only broadly started gaining an understanding of networks, their structures and their dynamics, it doesn’t mean they have not been at play for a very long time (at least 3.8 billion years, to be specific).
In my previous piece on distributions, I concluded by saying that it is often said that the Gaussian World privileges:
whilst the Paretian World privileges:
The difference between these two lists is very important. One set describes traits that are fixed and rigid, whilst the other describes states that are fluid and forever changing. The relevance for organisations (noting that the concept of an organisation itself is a Gaussian World construct) is that when we consider them in the Paretian context, recognising organisations as isolated and independent objects, or ‘things’ with fixed structure is mistaken: rather, we need to understand them as configurations of processes and dynamics of organising with emerging and evolving structure and topology (which is why Deleuze, Guattari and DeLanda’s Assemblage Theory is such a helpful and complimentary lens to understanding organisations).
Karl Weick was on to this many moons ago when, in 1974, he wrote:
“The word, organization, is a noun and it is also a myth. If one looks for an organization one will not find it. What will be found is that there are events, linked together, that transpire within concrete walls and these sequences, their pathways, their timing, are the forms we erroneously make into substances when we talk about an organization”.
The sharpness of Weick’s observation of more than 40 years ago becomes even more pertinent if we consider how connectivity, primarily enabled through the advent of the internet and communication technology, has shifted organisations even further into the Paretian World.
Viewed through a Paretian lens, it becomes an intriguing and indeed quite radical proposition that if trying to transform an organisation, what one must do is look to change the intangible, non-physical configurations of dynamics and processes that lead to the emergence and manifestation of the organisation, instead of (or in addition to) changing the more tangible physical things themselves.
This would, for example, involve identifying and then influencing (via amplification and dampening) patterns within the lower order micro-narratives shared between people (if one were trying to change the culture of the organisation—more on this when we look at narrative and culture next year), or identifying and then influencing (again via amplification and dampening, in addition to boundary-condition strengthening) the patterns of context-specific work tasks (eg. using best practice, good practice or emergent practice) within the organisation (if one were looking to restructure—more on this when we look at strategy and structure next year).
But how does any of this relate to network science?
The connection of all of this to network science is because the evolving structures* that emerge through the configurations of processes** of organising in the Paretian World are scale-free*** networks, meaning that if we acknowledge the realities of living and working in a Paretian World, we need to understand the basics of networks and the patterns and dynamics that they display. To illustrate the benefits of understanding the basics, let’s recycle a commonly-used network science example.
The Swiss-born mathematician Leonhard Euler is widely considered to be the ‘father’ of network science, and so it is with the infamous ‘Konisberg Problem’ that we begin. Cliched I refer to it and cliched it is—the Konisberg Problem is regularly cited as an introduction for anyone new to the field of networks, and rightly so: it teaches us a very important lesson about network science.
In 1736 Euler wrote a short paper on a problem that had long confounded the good people of Konisberg, a city in eastern Prussia, not far from St. Petersburg. The layout of Konisberg (now Kaliningrad) is such that it is bisected by the Pregel River, within which lie two islands—Kneiphof and Lomse—and seven brides connect the different parts of the city.
The story goes that, in a time long before Netflix & Instagram was a thing, Konisberg folk used to pass their free time by trying to solve a most perplexing riddle: could one walk across the seven bridges of Konsibgerg and never cross the same bridge twice?*
It would seem that much free time was exhausted, and it was Euler who eventually proved mathematically that there was no solution, because a route crossing each bridge only once does not exist. Euler’s coup de grace was to view the problem not as one that required vast intellect and persistence—as one might need when trying to solve a riddle or a maze—but rather as a graph, consisting of a collection of nodes connected by links. He simply viewed the four portions of land as nodes and the seven bridges which connected them as links.
Euler’s proof was as follows:
The key point to grasp here is this:
Finding the existence of a path does not depend on a person’s intellect, their leadership skills, their emotional intelligence, or whether or not they’ve completed an MBA.
Rather, it is simply a property of the graph.
Just as physical landscapes have properties that constrain or enable our ability to do things with them (e.g. walk up a valley, locate a city on the edge of a lake, build a tunnel underneath a mountain range etc), so too do network landscapes.
We just need to be able to see these properties.
The topology of networks is one of the keys to understanding the complexity around us in today’s world. Being able to see network structures can reveal previously unseen elements, and give us insight into new emergent possibilities. But we can’t see these new possibilities if we’re operating blind to our landscapes. We need to have a situational awareness that enables us to know when and when not to act, and how to act when we do act. And yet this type of situational awareness is something very few people within organisations have (which is why I am writing this series, in an attempt to change this current reality).
Gaussian (random) and Paretian (scale-free) networks
Let’s continue our exploration of networks by using our two examples of networks from last time: the road network of Australia (i.e. the physical), and the Qantas domestic flight network (i.e. the informational). Both of these are networks, yes, but they’re also fundamentally different types of networks.
Firstly, the Australian road network:
As we discussed last time, the Australian road network mirrors the properties of the Gaussian World and Taleb’s Mediocristan: it is based on physical things, allows for predictability and planning and projection, and is eminently knowable.
The road network is a great example of a random network—with the term ‘random’ referring to the isolated and independent nature of each of the nodes (in this case, the ‘populated places’ shown on the map i.e. the cities and towns)—with each node having approximately the same number of links (in this case, roads).
The nodes are described as isolated and independent because what happens at one node is unable to influence what happens at any other node.
For example, if all roads into both Brisbane and Sydney were to be suddenly closed, and traffic diverted to Maroochydore or the Gold Coast and Wollongong or the Central Coast respectively, the overall structure of the road network across Australia would not change significantly. Locally (i.e. in the vicinity of Brisbane and Sydney) there would be changes, but on the whole the structure would remain the same (which is what makes random networks so robust, as per the findings of Paul Baran at RAND Corporation in the 1960’s).
Now, let’s look at the Qantas domestic flight network:
The Qantas domestic flight network mirrors the properties of the Paretian World and Taleb’s Extremistan: it is based on non-physical information, and due to its multiplicative nature a nearly infinite number of future scenarios are possible, therefore making prediction very difficult, as we will see below.
The flight network is an example not of a random network, but rather a scale-free network—with the term ‘scale-free’ referring to the connected and interdependent nature of the nodes (in this case, the ‘ports’ shown on the map, which are free of the predictable scale (or proportion) of node-connectivity that occurs with a random network).
The nodes are described as interconnected and multiplicative because what happens at one node can significantly impact what happens at other nodes, which in turn can dramatically alter the entire network topology (as we will see below when we experiment with the concurrent closures of Brisbane and Sydney airports).
There are some key distinctions here between these two network types, with the critical one to understand being the distinction between the physical and the informational.
Whereas the random physical road network is comprised of atoms, which together make up the matter (i.e. sand, rocks, dirt etc) upon which the roads are built and the roads themselves (i.e. tar, aggregate, bitumen, concrete, paint, markings etc), the informational flight network is comprised only of bits (i.e. 1’s and 0’s), which are not physical, but instead informational. Whereas you can literally see, feel and touch the Australian road network, you cannot see, feel and touch the Qantas domestic flight network. One network is comprised of tangible objects, or things, whereas the other is comprised of intangible information relating to the movement (i.e. dynamics) of physical things (in this instance, planes).
A good way to illustrate how these informational networks are not physical is to look at one of my favourite time-wasting websites: Flightradar24.
Flightradar24 shows the movement of real-time commercial aircraft across the globe. Although it shows individual aircraft (i.e. things), of equal interest is the collective pattern that emerges from the movement of all of the aircraft.
And so while the site does not display a real-time network of commercial aircraft traffic, if you focus on the whole (i.e. the collective patterns) rather than the individual parts (i.e. each individual plane), what you start to get a sense of is a dynamic, shifting, pulsing network structure—these are the dynamics I keep referring to. Although you may not immediately perceive the patterns—instead just seeing a mass of aircraft—if you change your temporal scale (i.e. the timeframe through which you are watching it) you will see pulses around the world as plane landings and takeoffs peak during the daylight hours and then recedes during the night time (due to night-time curfews). The below animation does this pretty well:
But here’s the catch: because the network is informational and not physical, we don’t necessarily need the underlying map of Australia to help us understand the network’s properties. In fact, it probably helps to remove the Australian landmass from the illustration in order to explore the network.
This is what Euler infamously did back in 1736 when he solved the Konnisberg Problem, and so following in his footsteps that’s what I’ve done below:
A quick scan reveals how Sydney, Brisbane, Perth, Melbourne and Adelaide are the key ‘hubs’ of the network, and have much higher number of connections than the other nodes. Recall our power law distribution from last time:
The average number of flights to which a node on the flight network is connected is 4.4, with a standard deviation of 7.3. In layman terms, what this is saying is that: (a) on average, every city or town in Australia is connected to four other cities or towns; and that (b) we can’t be very confident in (a).
But of course the network itself—when viewed at a whole-of-system level, and not reduced into its constituent parts, as would traditionally be done when analysing things in the Gaussian World—reveals a different picture.
The average of 4.4 is revealed as meaningless and useless, or at the very least misleading. This is because these numbers do not sufficiently take into account the significant variation of some outliers i.e. Sydney, Brisbane, Perth, Melbourne and Adelaide (and a traditional analysis could easily lead these structurally-critical outliers to be dismissed or ignored entirely, as often happens with Black Swans).
Hence the need to understand that, in the Paretian World, complexity reduction is the approach of the idiot (but not the savant), and that comprehending, visualising, analysing, exploring, probing, sensing, intuiting and imagining at a whole-of-system level is instead the contextually appropriate response.
Now, to demonstrate the multiplicative and unpredictable nature of scale-free networks*, let’s consider for a moment what happens to the flight network when a small sequence of events lead to a dynamic known as trophic cascading—a term I’m borrowing from the ecological sciences—which results in significant structural change to the network.
Let’s assume that two interconnected events occur.
Firstly, a warm and moist airmass moving in a south-easterly direction over the eastern seaboard of Australia leads to a powerful thunderstorm cell forcing the closure of Brisbane airport, with all flights to Brisbane being redirected to Sydney.
Secondly, the south-easterly movement of the airmass pushes a large dust storm into New South Wales and over the Sydney basin, which forces the closure of Sydney airport. All flights into and out of Sydney are now grounded.
In a very short period of time, due to its dependence on stable weather conditions, the highly-connected nature of the flight network leads to trophic cascading across the network and a rapid breakdown of the network structure. Two interconnected events occur (arguably these are actually the same event), and the network structure changes to the extent that 23 nodes become isolated, and 63 links disappear, as I’ve illustrated below:
Now, let’s play around with one additional variable in this problem space.
Let’s assume that due to some idiosyncrasy all flights previously directed to Sydney (including the Brisbane flights which were redirected to Sydney) are now redirected to the small NSW town of Parkes, lying approximately 400 kilometres to the west of Sydney.
With a human population of about 11,000, you’re probably most familiar with Parkes for the early 2000’s movie The Dish (which told the story of the town’s radio telescope and its role in the 1969 moon landing) and the town’s annual Elvis Festival.
On any given day (other than the second week of January, when the Elvis Festival is on), Parkes is not particularly busy; nor is its small regional airport:
So can you imagine what would happen to Parkes if all of sudden it had 65 domestic flights—not to mention a similar number of inbound international ones, which we have not considered in this scenario—rerouted to it?
The flight network’s structure changes again—and although it retains its scale-free topology—it is an entirely different network, with Parkes as by far-and-away the most-central and most-connected node:
Consider how the connectivity of Parkes has changed in a short period of time:
A sequence of events such as this is not something that you could specifically plan for. The number of variations possible in the network is nearly infinite. And yet these are the unexpected and unpredictable type of Black Swan events that eventually manifest in the Paretian World.
This is exactly what happened to the small Canadian town of Gander on the 11th of September 2001. Due to the unprecedented nature of events that day, 38 wide-bodied airliners were redirected to Gander, where they landed and spent three days. The Gander population of approximately 10,000 people swelled to nearly 17,000 as they hosted the planes unexpected passengers as visitors.
This is yet another example of a Black Swan event which—although possible—was not entirely predictable and only made sense once viewed retrospectively. (As an aside, the novel events of that time in Gander have been enacted in the successful Broadway play called “From Far Away”).But again, so what? Road and flight networks? What does this have to do with organisations? Please bear with me: we’re almost there.
The types of network topologies that arise from social connection between humans—for a long time thought to be random in nature (where everybody has roughly the same number of connections, and where the mean is meaningful and useful)—are in fact scale-free* (where there are some individuals who have order of magnitude higher number of connections, and where the mean can be misleading). In short, social network structures are more like the flight network and much less like the road network. This is also means that emergent properties manifesting from the connections and relationships between people—such as culture—display scale-free properties, too.
* Or at least close enough to being scale-free. The primary drivers of this are the interplay of network dynamics including growth and preferential attachment, which is where an existing node in a network with more connections is more likely to receive yet more connections—this is multiplicativity at play.
But the problem is that the traditional approach to organisational management understands the typical organisation as akin to the random road network, comprised of its isolated and independent parts. The reductionist and additive (not multiplicative) sum-of-its-parts traditional approach to changing systems is flawed and often counter-productive when the system’s parts are interconnected and multiplicative. Thus, the entirely different Paretian World means than when things are connected, an entirely different approach is required. This is the key foundation underpinning Cooper Ramo’s heuristic that connection changes the nature of things.
So now let’s reconsider our flight network, but this time switch the names of cities and towns for the names of people, and the names of states for the names of departments within an organisation:
This network exhibits a fairly typical structure of the social network arising in communities of people, in which some people are very well connected, and other people are less so. The nature of the connection in a community can be any number of different things, such as friendship, influence, advice, sexual or financial dependence, but in this instance let’s assume it is influence amongst employees within an organisation. As such, we could assume that Sally, Barry, Melanie, Penelope and Adele are the most influential people within the network*.
* Which means that the source of your relationship data is important. How do you determine exactly who is and who isn’t influential? This is tricky, and means we must be realistic about the limitations of the dataset we use when mapping networks of influence and other similarly ambiguous metrics.
Now, if Sally, Barry, Melanie, Penelope and Adele have job role responsibility to exert their influence within the organisation—let’s say that they sit within the executive (and noting that their ability to influence the organisation may be a result of their job role)—then that’s great; it looks like they’re the right people for the job.
But what if the opposite were true? What if Hillary, Misty, Gupta and Tom comprise the executive and have job role responsibility to influence change in the organisation? They will have their work cut out for them—and the first piece of work to be done would probably involve understanding why it is that there are people in the organisation outside of the executive (i.e. Sally, Barry, Melanie, Penelope and Adele) who have far more influence than the executive.
And now let’s again imagine that a similar sequence of events were to occur to Barry, Sally and Parker, just as they did to Brisbane, Sydney and Parkes. Let’s imagine that Barry is fired for financial misappropriation, that Sally retires to live in South America, and that for whatever reason Parker emerges from obscurity to become the most influential node within the network. Only a few events have occurred but the structure of the network has changed dramatically, as has the potential for different nodes to play significant roles within the system.
Of course, the scenarios I have toyed with here are huge simplifications not only of the complexity of human systems and their dynamics, but also of how you go about deploying network visualisation and analysis in an organisation (it ain’t that easy, I can assure you). But hopefully you are now beginning to understand how valuable an ability to see and understand the changing network structure of an organisational system—at a whole-of-system level—might be.
Critically, it is the patterns of connection between a network’s nodes that determine network structure (and hence, behaviour) as much as the composition of the network’s nodes—in my experience, this is a point very few people grasp. This means it is critical to understand that network structures emerge out of the dynamics of connection, and that as these structures emerge they in-turn influence the network dynamics, which in-turn influence the network structure, and so on and so forth. This is feedback in action*.
Perhaps the most important thing to grasp—and indeed, every time I work with a group of people who are ‘new’ to the network space I see this—is that working with networks, be it visualising them, analysing their data, or mapping them to see where to intervene in the network, requires an entirely different way of thinking.
I refer to this type of thinking as ‘ecosystem thinking’, and its what I’m trying to help you as the reader of this series develop. For most of us, we are not brought-up to intuitively understand and ‘see’ the relationships between things; rather, we are encouraged to understand and see ‘things’ as reducible, discrete, isolated, neatly-defined objects.
And so when we are ‘ecosystem thinking’, the type of analysis that we do—the types of questions that we ask, and the answers that we seek—are of an entirely different nature. Rather than asking the individual, we ask the system. Rather than isolating and reducing, we look for connection and emergent possibilities. In short, we are interested in the whole-of-system structure and patterns, and not just the system’s individual parts*.
Until 1999 and the publishing of the work done by the Faloutsos brothers, the topology of the internet was assumed to be random in nature—that is, just like the Australian road network. But everyone was wrong, as the Faloutsos brothers showed.
The same thing has happened with organisations. Assumptions have been made that they are of the Gaussian World, and are random in nature, where every person is, for the most part, like every other person, and is treated as such (i.e. additive, isolated and independent). What I’ve tried to show you in this piece—in a rather abstract way, admittedly, given I’ve used transport networks to begin with—is that it might be wise to reconsider these assumptions.
In the linear and additive Gaussian World, projection- and probability-based approaches allow us to know with mathematical certainty the number of outcomes that can occur. This certainty enables us to do what organisations are very good at doing: categorising things. Whether it be organisational charts, risk matrices, financial models or cultural values, organisations love to categorise*. Everyone loves to fit their data into their categories**.
But in the non-linear and multiplicative Paretian World the limits of projection- and probability-based approaches are reached and then exceeded. The infinite number of possibilities means that you can’t predetermine the categories and then fit the data.
Rather, you need to explore the data and then see what patterns and structures emerge (that’s why the Cynefin Framework is meant to be a sense making, not categorisation, framework).
Remember back to this piece where I quote Professor Richard Feynman and advocate for science and exploration rather than philosophy and tourism? The need to make sense of rather than to categorise is what I was getting at. Categorisation goes hand in hand with exploitation, where you know the environment before-hand and you know what and how to take advantage of. Sense making is, on the other hand, all about exploration of the unknown environment. And yet what is the standard response for any organisation facing existential pressures borne from the complexity of the Paretian World? It is too cut the budget for any spending on exploration.
Think of the pre-eminent traditional approach to categorisation in an organisation: the ‘org chart’. That’s what the org chart is: a categorisation framework. It identifies different categories—be they different job roles, levels of reporting, or authority and power—and then places people into them. Org charts work great for… yep, you guessed it: the complicated, linear Gaussian World.
But… org charts don’t work so great for the complex, non-linear Paretian World. This is because they can overly constrain the movement of people, information, cognition and decision making (to name but a few) within the organisation, and reduce its requisite diversity and optionality to be able to respond-to, adapt-to, evolve-to, co-evolve-with and change its surrounding environment.
So give this some thought… what if, instead, you were to use the existing network topologies within your organisation to make sense of and see possible emergent structures, and then scaffold these emergent structures (where appropriate) to encourage context-specific responses which in-turn enable the appropriate configurations of dynamics to ensure work is done in the best way possible?*
Although in the past two decades there has been a recognition that organisational restructuring from hierarchy to networks is an appropriate response to increased environmental complexity, almost everyone advocating this still fails to grasp the difference between categorisation and sense making (even though such an understanding of this difference is a basic tenet of working with complexity).
Simply ripping up the old hierarchy and drawing up something newer and flatter or more networked is nothing more than talking about the birds with no deeper knowledge or understanding whatsoever about the birds. It’s commonly portrayed as something along the lines of this:
But the thing is, this naive approach is simply replacing one categorisation framework with another categorisation framework, regardless of the natural sciences language and metaphor. It’s like ripping an old bandaid off and replacing it with a new one, but not treating the cause of the symptoms. In the short-term it might work—and it will certainly alleviate tension, because everybody has something new and exciting to talk about—but as the environment continues to change, the new structure will inevitably become overly-constraining and no-longer relevant for its context, and… cue the next restructure based on the next management fad*.
You may recall this piece from a few years ago where I suggested three heuristics to guide organisational transformation:
Hopefully it will be clear to you that these heuristics are just as relevant today as when I previously wrote them (eg. the first two heuristics are all about sense making—starting from where you are and with what you have—rather than categorisation). Although I cover a lot of topics in my writing, the topics are all underpinned—or indeed, connected—by the same basic concepts which do not change.
These basic concepts are what I refer to as scale invariant, meaning that they are true regardless of scale and context. And so this brings us to a close for today, but gives us a tantalising lead-in to our next piece of writing, which will be everything about spaces, scales and fractals.
Finally, for those of you who take my writing seriously, and invest the necessary time to understand and comprehend, I say: thank you.