Systems Thinking - When the human is in the machine

Consider any connected product or application that you may have – a fitness device, a mapping application, a dating app, a payment app, or a smart thermostat. On the face of it, these are superficially quite simple products. A fitness app tracks your activity, a mapping app helps you get from A to B, and a dating app matches you with potential dates.

However, beneath that veneer of simplicity lies something else. All these products are systems that connecting people and things. As the number of people and things connected to each other grows (let’s call them nodes), something rather magical happens. The collection of nodes as a whole becomes akin to a living organism. Each node, often a person, is in effect, acting as though it is a cell of a larger system, acting and reacting to its connections with other nodes.

This ‘organism’ develops its own rhythms, both seasonal and daily. Mobile networks will perform better in winter as the leaves fall, and the amount of data will track the rise and fall of the sun, spiking when there is a big sporting or political event. Social networks will reflect, reverberate, amplify whatever is happening in the real world, which in turn reacts to what takes place on social networks. In the meantime, our Smartphones and various connected devices are constantly reporting back to the ‘cloud’ on everything we do, who and what we interact with.

*Echo Chamber Polarisation in Twitter – retweets of liberal (blue) and conservative (red) messages*

This poses a significant challenge for anyone designing these systems, because people are key parts of the system, often following their daily routines, at times being unpredictable, and occasionally interacting vigorously with other people in the system. In short, users are an integral part of the product infrastructure. Your tech product / smart product / social network no longer consists solely of the technology components, (apps, cloud, networks etc), but also of the people who have chosen to use your systems.

This poses a serious engineering & technology problem. How do you design a product where people are effectively part of the product itself? This is where Systems Thinking comes in

Systems Thinking is not Systems Engineering

One of the first questions to be asked when faced with Systems Thinking is “isn’t it the same as Systems Engineering?” The answer is “not really.” Systems Engineering provides a way of dealing with complexity by breaking down complicated systems (see later on for a definition) into smaller subsystems. Descriptions of systems such as a space vehicle are decomposed into progressively smaller systems and subsystems until they can be described by their atomic constituent parts.

Systems engineering provides the tools for designing a system as the sum of these parts, describing them to a high level of detail, including how each system interacts with each other and all the ways in which systems may fail. This is effectively a blueprint for your entire system, be it a spacecraft, a plane or a car. Then, the process is reversed, the components are built, designed, and subsystems are progressively put together until the full working system is assembled, tested, and ready for operation. In a way, it is like creating a Lego kit – the systems engineering process produces the constituent blocks and assembly manual.

Systems engineering is a methodology that aims to drive full predictability in the creation and operation of their system. In doing so, its approach is to eliminate any vestiges of uncertainty and unpredictability. As such it is very popular in industries where failure comes at a high economic, or indeed, a moral cost such as aerospace, aviation and automotive.

Let’s consider the premise of predictability. It means that a problem, once solved, is eminently repeatable, and the same approach will provide the same outcome. For example, you can disassemble a Ferrari, and two days later reassemble it. In doing so, you will pull together the exact, same car. No matter how many times you repeat this process, as long as you don’t break anything, you will reassemble the same vehicle. The key point is that as the problem being solved is predictable, so is the solution.

Complex Systems

However, many systems are not repeatable. Certainly, the examples we started with in this post, such as social networks or large distributed cloud systems are very poorly suited to be solved purely by Systems Engineering. In these examples, very difficult to characterise exactly what the problem is, and how the system will or behave. Particularly, where the environment or people are part of the systems, or where a system consists of many interacting parts, such predictability is impossible.

Consider for example the traffic management systems of a large city, aimed at keeping road traffic flowing. There are a number of aspects that make this a complex system.

Britain from above: Aerial shots show off the beautiful colours of the UK from the sky at dusk | London, Sky london, London night — *A city’s transport network – The complex system par excellence*

Multiple Interacting Components. A city’s road system does not exist in isolation but is part of a mass transit system that includes a combination of buses, trams, cycle lanes, pedestrian pathways and underground systems. We can consider these as subsystems, all of which are dedicated to transporting people across the city and interact with each other.

Network Connectivities. Not only do the systems interact, but so do all the participants in the system. All vehicles seek to avoid each other, vehicular traffic gives way to pedestrians, cars make way for cyclists. These interconnectivities create a high degree of complexity.

Image for post — *Comparison between Complicated and Complex Systems – Source: General Stanley McChrystal, Teams of Teams*

Feedback Loops. In addition to the interaction between participants, there are often strong feedback loops between the systems. For example, disturbances to road traffic will create stress on the underground system and vice-versa.

Susceptibility to Disruption by Predictable and Unpredictable Events. We have seen how participants and sub-systems interact with each other on a day-to-day basis. On top of all this, disruptive events. A large sporting event will cause stress to both road and mass transit systems, though these can be planned for. More difficult to deal with is the unexpected – freak weather events, unauthorised demonstrations, strikes by public transport workers. All these have complex knock-on effects that ripple throughout the system.

Taken together, all these factors create systems that cannot be predicted and cannot be resolved by traditional systems engineering techniques. The issue of unpredictability is not limited to very large systems. We are surrounded by complex systems. Anybody who has worked in a large office block will wonder why the temperature control is often so poor has experienced the failings of traditional systems to deal with a complex problem. A large building is subject to interactions between different floors, different working patterns, unpredictable weather, which make the simple act of maintaining a comfortable temperature, a complex task.

Social networks such as Facebook and Twitter would have struggled to predict the impact of their algorithms in creating the highly polarised echo-chambers of political discourse. The viewpoint of a self-driving car in an ordinary urban street poses a highly complex and unpredictable problem and each journey will be distinct from every single simulation which the algorithms will be trained on.

Dealing with Complex Systems

Having established that complex and-difficult-to-predict systems exist and that existing management and design approaches aren’t quite appropriate, the obvious question is how do we deal with them? One of the first frameworks to describe these different types of problems was proposed by David Snowden and Mary Boone, in what they called the Cynefin Framework (based on a Welsh word meaning a habit where one feels most at home in). Problems are defined as being Simple, Complicated, Complex and Chaotic.

Simple problems are well-defined, and lend themselves to being codified in manuals, and to top-down management. Complicated problems typically require groups of experts to come together, and break down into constituent parts and subsystems that can be dealt with by experts. This is the realm of Systems Engineering as described above. There is no single established ‘correct solution’, but a number of different good solutions that need to be honed down to the chosen ‘best’ solution.

*Adapted from Snowden and Boone, A Leader’s Framework for Decision Making, Harvard Business Review, 2007*

So how do we deal with Complex problems? Much of the most advanced literature into systems thinking can be found in areas such as socio-environmental studies and public health. These fields are full of complex systems of intrinsic complexity and inter-relationships between multiple complex organisms (be they people or otherwise) together with self-correcting mechanisms and loops found in these fields.

These fields have been very much the driving force behind modern Systems Thinking. Here are some of the key approaches, and their broader applicability into tech and product development.

1. Deal with the Whole, not with the Parts

The cardinal rule of Systems Thinking as applied to complex problems is to consider the problem as a whole and avoid the pitfalls of optimising a solution for part of the problem. In fairness, this is also the approach adopted by Systems Engineering. The larger and more complex the problem, the greater a challenge this is. While there isn’t a single playbook that can be applied systematically, here are some pointers.

Understand the Interconnections. Most complicated systems have linear relationships. You apply the brakes. The car stops. Complex systems are characterised by circular relationships. This is well understood in environmental and biological systems, but is equally true in modern connected systems.

Look for feedback loops. Some of the circular relationships will self-stabilise. Many processes in the Earth’s atmosphere are self-correcting, and these are known as balancing loops. For example, as the concentrations of carbon dioxide in the atmosphere increases, so does its absorption by plants through photosynthesis. Reinforcing loops, on the other hand, create run-away effects and instability. For example, an increase in house prices results in greater demand, which in turn drives up prices further. This loop is self-reinforcing until there is a correction of some form, such as a price crash.

Consider the public health response to the Covid pandemic and its possible outcomes. A paper by Bradley et al presents a causal loop diagram showing a number of interactions between policies, environmental transmission, person-to-person transmission, public sentiment, the number of susceptible people etc. What this clearly shows is that there isn’t a clear linear cause-and-effect. In other words, you cannot simply state, we will enact this policy, and this will be the outcome. The number of interactions between actors in the system create a highly unpredictable system. Nevertheless, understanding the linkages, loops and interactions is the first step towards finding a solution.

*Causal Loop Diagram showing some of the interactions in a society responding to the pandemic threat of Covid-19*

Emergent Behaviour. From a systems perspective, a seemingly coherent, ordered, larger behaviour can ’emerge’ from many things coming together. The Social Media echo chambers of people sharing a political persuasion, resulting in certain topics trending on Twitter and gaining a life of their own are examples of emergent behaviour.

2. Experimental Mode of Management

The first rule of dealing with Complex systems is to assume that you will unlikely to stumble across the first solution straight away. The Covid causal loop diagram above makes that pretty clear. Instead, product launches, business initiatives and product development programmes need to be designed as experiments, which can be used to navigate the uncertainty and iteratively converge to a good outcome.

Leaders who ‘know’ the right way or seek to impose top-down control are simply deluding themselves and are doomed to fail. The intent here is to go forward step-by-step, wait for patterns to emerge, and then ascertain the best way forward. Key to success is an acceptance, and indeed, a willingness to embrace failure as a necessary step towards success. This is why the lean development cycle of Build-Test-Learn, so beloved of the Lean Startup Movement, has become so entrenched in the culture of tech companies. Simply put, it works.

*Build – Test – Learn Cycle – Source: Lean Startup*

3. Decentralise Decision Making

In a previous post, we looked at General Stanley McChrystal, who used to lead the US Joint Special Operations Command in Iraq which captured Saddam Hussein and killed al Qaeda’s leader in Iraq. When faced with the task of dealing with the insurgency in Iraq, he realised that this was a rapidly-changing, fragmented and complex problem which traditional hierarchical military decision-making structures struggled to cope with.

McChrystal advocated “delegating until it hurts’. As long as the intent is well-understood, and everyone has a good grasp of the overall picture, a state he calls shared consciousness, then operational decisions can be made by those teams closer to where the problem is, allowing for a greater speed of response. For the military, this was pretty revolutionary. McChrystal sacrificed some sacred cows, prioritising delegation over tight command and control, as he felt that the delayed reaction are too great, and situational awareness over information security.

Team of Teams: A Leadership Model for a Complex World
By Dan Snelson The 21 st century is a time unlike any other. Modern technology allows instant global communication for everyone, making the world no longer just highly complicated, but increasingly complex. It is this complexity, argues General Stanley McChrystal in his 2015 book Team of Teams , that makes it vital that we take a fresh look at how we think about leadership, management, and teamwork.

A topical example of the cost of delayed action can be seen in the public health response to the current Covid-19 pandemic. A paper by Michael C Jackson, a respected systems thinking expert, persuasively makes the case that the UK response was hampered by its decision-making structure. The highly centralised National Health Service (NHS) and public health bodies struggled to get a grip of the fast-moving Covid pandemic.

At the beginning of the outbreak, the situation was chaotic – there were fundamental unknowns in how the public would react to restrictions, the mode of transmission was poorly understood, and the government was treading in uncharted political and psychological territory. It however decided to take an analytical, and deliberate decision-making approach, popularised by the phrase, “follow the science.”

However, there was no established science, just models with high degrees of variability. As Taleb, of Black Swan fame, put it, “someone watching an avalanche heading their way does not call for complicated statistical models to see if they need to get out of the way.”

*Mismatch between Pandemic Complexity and UK Government Structures*

Once the initial chaotic phases were navigated, and the government ramped up its Covid response, it was still stuck by an organisation that was ill-suited for the task. Public Health in England is centralised at a national level, meaning it was set-up as a command-and-control structure. There was therefore little diversity in approach, a disconnect with what was actually happening on the ground, and no understanding of the local context.

There was not a single pandemic, but clusters of infection around the country, which the decision-making structures were increasingly struggling to match. Even when a regional ‘tiered’ system was put in place, all decisions as to which Tiers each region should be placed in were taken centrally.

Conclusion

Unfortunately, the current Covid pandemic spells all too clearly the failures to adopt the decision-making and solution-design approaches that best match a problem. However, what is true for societal and environmental problems, is also true for complex tech and engineering problems. In this post, I have touched on superficially some of the approaches that can help deal with the unpredictable, the unknown unknowns.

Think of the whole system, be prepared to experiment, and delegate decision making. That will make for a good start.

Happy New Year!