Introduction

Systems change over time.

Ideally, we want them to change in a way that benefits us.

To do this we need to understand how these systems change (evolve), and what tools we have available to influence the direction of this change.

Evolution isn’t just biological. Learning how to think about a system from an evolutionary perspective aligns decision making with long term success. This kind of thinking is what ultimately fuels large-scale transformations of virtually any system, and mastering it’s principles gives you the tools to enact exponential change.

“The world is wide, and I will not waste my life in friction when it could be turned into momentum.”

— Frances E. Willard

Mechanisms

Here’s a few handy definitions, for the purpose of this article;

• Entity – an individual unit in a system possessing certain traits.
• Trait – a measurable variable that positively, negatively or neutrally affects performance on a test.
• Mutation – a new trait.
• System – a collection of entities engaging in behaviours that drive a specific output for given inputs.
• Test – a way of determining whether a trait is above the system average.

For our purposes, a system is a collection of entities engaging in behaviours that drive a specific output. These entities have certain traits, some of which produce differences in their behaviour. Traits which improve the output of the system are referred to as “favourable”, while traits that do not improve output are “unfavourable”.

The thing about systems is that they are never stagnant. When a novel trait emerges in one of the system’s entities, the trait is an unknown variable. This is referred to as a mutation. Depending on the requirements of the system, the mutation will eventually be classed as either favourable or unfavourable.

Following that logic, a testing mechanism much exist to sort the wheat from the chaff, or the good from the bad. Traits that pass a selected test are classed as favourable. As soon as there is some way for favourable traits to be spread to other entities faster than unfavourable traits, the system begins to evolve in the direction of the tests.

Therefore, there are 3 mechanisms that fundamentally drive evolution of a system;

1. Continual randomised novelty
2. Tests that favour certain types of novelty
3. Spread of the favoured novelty

Continual randomised novelty

Systems do not exist in a vacuum. There are external unknowns constantly impacting the system in a positive, negative or neutral manner. Chaos, in its truest form. A property of chaos is that it is shrouded in the fog of the unknown. If your direct manager at work calls you into their office “to talk about something important”, the lack of certainty is bound to stir up all kinds of anxiety-inducing visions.

• Is my performance not acceptable?
• Have I been slacking off too much?
• Am I going to be fired?

But also, optimistic visions of a better future may dance across your internal monologue too.

• Is my work on the last project about to be acknowledged?
• Will I get a promotion? A raise?

Depending on which state is actually present in the unknown (a firing or a promotion) different behavioural traits take on differing levels of importance. If you are receiving good news, adherence to previous behaviour patterns may be a virtue. If you are being fired, the ability to change direction seamlessly and rebound effectively may be what is needed.

The point is that all systems contain both unknown problems and unknown opportunity. This is essentially the definition of chaos, or unpredictability. We have no way of knowing exactly what traits will be most useful for future tests thrown at the system. This means that lots of novel traits must be tested, and the tests imposed on the system must allow the most favourable novelties to succeed, thus ultimately enriching the system.

Biological evolution provides the clearest demonstration of continual randomised novelty in action. Each organism, a mosaic of genetic quirks and traits, stands as a testament to the relentless trial and error of life’s grand experiment. In the realm of ideas, creativity flourishes under this randomness. Thoughts and concepts are the entities, each bearing the potential for paradigm-shifting innovation.

Shifting our focus to the corporate world, companies that embrace mechanisms of continual randomised novelty often find themselves at the vanguard of their industries. By fostering an environment where risk-taking is encouraged and failure is not vilified, these organisations create a fertile ground for breakthroughs. A sort of ‘mutation-rich environment’ where the next revolutionary product or strategy might just be a fortuitous anomaly away.

Clearly novelty is essential to all innovation and forwards movement, but pure creative chaos doesn’t seem sustainable either. Not all new ideas work out, and pursuing all of them sounds like a recipe for disaster.

This begs the question – how do we figure out which forms of novelty are good?

Tests that favour certain types of novelty

Novelty provides us with unique soldiers to send through the gates of chaos, a select few of which will return as heroes and win the favour of the system. But who decides which forms of novelty are favourable?

Traditionally, trait favouritism depended on which tests emerge from chaos. Biological evolution through natural selection is a good example of this.For instance, before the Industrial Revolution the peppered moth predominantly had a light-coloured phenotype. This light coloration provided camouflage against the light-coloured lichen and trees where they rested, protecting them from predators. Dark-coloured moths, being more visible against the light background, were more likely to be eaten.

Then, everything changed.

During the Industrial Revolution, heavy pollution caused by the burning of coal in factories darkened the trees by covering them in soot. In this darkened environment, the light-coloured moths became highly visible to predators, while the dark-coloured moths – which were previously at a disadvantage – now had better camouflage.

As a result, the dark-coloured moths had a higher survival rate and were more likely to reproduce. The exact same test (survivability) favoured different traits depending on the environment, but ultimately drove higher survival rates no matter the externalities.

This is great for producing hard-to-see moths, but what about achieving a more useful outcome?

The average height of basketball players in the NBA is 6″6 as of 2023(1). Given that less than 1% of Americans are over 6″4, it is clear that the “tests” imposed by the NBA system naturally favour height. The outcome for us takes the form of athletic giants slam-dunking on each other for 48 minutes – not bad.

If we want to act in place of chaos and ensure that the traits favoured by tests actually serve the system’s goals (and ultimately, our goals), we must be extremely specific about what tests are imposed on the system. For example, a selection of different potential advertisement campaigns for a company can be first tested with focus groups. The test could be based on how accurate each of the focus group participants are at remembering the central messaging of the advertisement, 10 minutes after being shown it. If the ad that is eventually shown to the public is the one with the best remembrance from focus groups, and the loop is repeated with each new marketing campaign, then the company marketing “system” will evolve to become more and more memorable.

Essentially, the tests imposed on a system determine the ideals it evolves towards. Whatever novelty is required to pass the test is the novelty the system will move towards.

To complete the loop, one final piece is needed.

Spread.

Favoured novelty is spread, unfavoured novelty is contained

The final mechanism necessary for system-level evolution is a distribution of the favoured trait to other entities in the system. The benefit of the novelty will not exceed the lifespan of the single entity it is connected to unless it can replicate itself by spreading to different entities within the system.

The most natural facilitator of this mechanism exists inside every human, including you, in the form of mirror neurons. These systems are thought to be located in various brain regions, including the premotor cortex, the supplementary motor area, the primary somatosensory cortex, and parts of the parietal lobe. They allow individuals to replicate actions they observe in others. This is a fundamental mechanism in learning – especially during early childhood.

The entire fashion industry is built around this mechanism. For instance, when a new trend emerges and is seen as desirable—perhaps worn by a celebrity or influencer—it quickly spreads among the public. The trend, or ‘favourable novelty’, enhances the entity’s output, which, in this context, is social approval, identity expression, and aesthetic appeal. Those who adopt the trend ‘pass the test’ set by the system’s standards of desirability. Interestingly, the temporary nature of fashion and the zero-sum mechanism of social status means that the fashion system can never truly evolve – only cycle. This is because the elevation of status of one individual must, by necessity, reduce the status of everyone else in comparison. Additionally, this cycle is driven more by social and cultural factors than by actual functionality or adaptability, which are central to classical evolution systems.

There are, however, instances where the wide-spread adoption of novelty does permanently raise the system standard. The Fosbury Flop, introduced by Dick Fosbury during the 1968 Olympics, completely revolutionised the high jump landscape. Before Fosbury, jumpers used techniques like the straddle method or the scissors jump. However, Fosbury’s technique, involving a back-first approach over the bar, allowed for greater clearance height. This “novel” approach quickly spread among jumpers who saw it and became the default technique at the Olympics to this day. As a result, fans get to watch continuously higher jumps than ever before.

In summary, produce novelty, test it to find the best forms, then spread the winning traits to the rest of the system’s entities. This is how the system grows.

Scalability

After each of the evolution mechanisms are in place, the next step is to increase the rate of evolution using 1 of 3 responses. These are;

• Improve novelty production
• Improve novelty testing
• Improve spread

The choice of which strategy to pursue at any given time will depend on where the primary limitations are, and thus where the biggest performance gains can be found.

Improve novelty production

All innovation (and thus all improvement) is encased within novelty. This means that the only way to increase the amount of innovation in a system is to a) increase total novelty production, or b) increase the ratio of positive to negative mutations.

A good example of the “more volume” approach comes from Toyota and their production philosophy, which emphasises the generation of ideas (mutations) from anywhere within the organisation. Since there is minimal cost in discarding bad ideas, drastically increasing the total idea production ensures an increase in good ideas – even if most of the ideas aren’t up to scratch. Employees on the factory floor, for instance, are urged to suggest changes that could improve efficiency, reduce waste, or enhance quality. These suggestions are taken seriously, and the best ones are often implemented. This leads to on-going improvements in Toyota’s production processes and an ever-growing company culture.

Of course, volume isn’t the only approach. If volume stays the same but the ratio of favourable-to-unfavourable mutations improves, total system innovation still increases. This usually occurs through logically aligning the types of mutations to an end goal, rather than relying on randomness.

The company Genentech is a pioneer in personalised healthcare space, and a fantastic example of this approach. By focusing on the genetic makeup of individuals and the identification of certain biomarkers, Genentech’s precision medicine allows for the creation of new treatments (mutations) that have a higher chance of being effective. For instance, the company identified the HER2 protein in patients, and used this to test medicines that positively interact with this protein. This eventually led to the creation of Herceptin, a drug specifically effective against HER2-positive breast cancer.

Amazon leverages data analytics, machine learning algorithms, and artificial intelligence to analyse vast amounts of customer data. They observe browsing history, purchase patterns, and even time spent on particular products. This data is then used to create highly personalised product recommendations and targeted advertisements. Amazon’s method exemplifies the “quality” approach by focusing on the relevance of each customer interaction, rather than simply bombarding customers with a high volume of generic advertisements and seeing which one sticks.

Taking this “quality” approach does sacrifice some of the out-of-the-box thinking that occurs as a by-product of unfiltered randomness, but it also cuts down on negative/neutral mutations massively. Focusing on volume or quality are both valid strategies, and the perfect approach for you will likely rely on some combination of both.

Improve novelty testing

Novelty production is often the bottleneck, however sometimes an inefficient testing process can begin to slow down the evolution of the system. Inefficiencies in the testing process occur through two main limitations;

The first limitation is volume – i.e. can the testing system handle an increase in novelty production without significantly decreasing the rate of testing? As an example, pharmaceutical companies use high-throughput screening to test thousands of chemical compounds for potential drug efficacy. This process involves automated robotic systems and advanced data analysis software to rapidly assess the effects of numerous compounds.

Automation, while often ideal, isn’t even the only option for rapidly testing solutions.

Platforms like Kaggle enable the distribution of problem-solving tasks among a global community of data scientists. By allowing a large number of participants to work on the same problem simultaneously, diverse solutions (mutations) can be generated and evaluated independently by the contributing data scientists. This produces large amounts of novelty and equips it with user-initiated testing to complete the loop.

The second limitation is testing duration.

Once additional novelty generated can be handled without slowing down the system, reducing the gap between when a novelty is produced and the deliverance of it’s testing will speed up the evolution of the system. Once again, automation reigns king here but is not the only approach. Rapid prototyping is a transformative approach in product development, significantly reducing the timelines for testing and refining designs. It involves quickly fabricating a scale model of a physical part or assembly using three-dimensional computer-aided design (CAD) data, rather than building an entire new product to test. This massively cuts down the timeline between creation and testing, allowing for rapid feedback loops.

Improve spread

Spreading mechanisms have exploded over the past few decades, spurred on by the continuous growth of the internet. For instance, in a corporate environment, intranet systems, newsletters, and regular meetings can be used to spread information about successful innovations or practices. In an informal setting, social media and instant messaging apps play an enormous role at spreading novel behaviours and ideas (although not always favourably).

The more entities there are in a system, the greater the importance of spread. If favoured anomalies do not spread throughout a system rapidly, their utility is massively diminished. This spreading mechanism can either be innate or manual. Innate spread means that entities in the system that receive the novelty have the capability to pass the novelty to other entities. Manual spreading requires external intervention to deliberately pass the novelty to other entities, typically with various automations.

There can also be a hybrid approach. A centralised education curriculum that is manually spread to schools through dissemination, and mandated through legislation, can effectively distribute quality education at scale. Private tutors can then automatically spread the curriculum to students that did not effectively receive the original spread. By combining these approaches, nearly 80%(2) of all Australians pass all 12 years of schooling – and the ones that don’t typically choose not to.

Since spread does not exist to the exclusion of other mechanisms, external forces can enhance or hinder spread. For our purposes, we group all hindering forces as “spread friction”. This includes psychological inertia, replication issues, technological limitations, competition with other mutually exclusive novelty, and more. Understanding spread friction is essential, as improving spread is often a by-product of simply addressing the biggest counter-forces rather than adding anything else to the system.

There are times when spread friction is useful, too. Reducing the spread of unfavourable novelty provides a layer of insurance in the system, helping to reduce degeneration below the system mean. If spread is automatic, protecting against the spread of unfavourable novelty is of the utmost importance – even more so if the mutation is mutually exclusive with other more beneficial mutations.

Summary

There are certain mechanisms that drive the evolution of all systems – not just biological evolution. This includes evolutions in advertising, pharmaceuticals, sporting technique, and more.

These mechanisms can be deliberately introduced into a system to cause it to evolve. The “direction” it evolves in will depend on what tests are used. Whatever variable is tested is the one that will continually improve over time.

The mechanisms used to drive evolution are;

1. Continual randomised novelty
2. Tests that favour certain types of novelty
3. Spread of the favoured novelty

Once these mechanisms are in place, the speed of evolution can be increased through improving each mechanism.

• Mechanism 1: Continual randomised novelty
  • Improve novelty production
    • Increase mutation volume
    • Improve mutation quality (more good ones, less bad ones)
• Mechanism 2: Tests that favour certain types of novelty
  • Improve novelty testing
    • Can additional mutations be handled?
    • Can the speed of testing be shortened?
    • Can the tests be automated?
• Mechanism 3: Spread of the favoured novelty
  • Improve spread
    • Innate vs Manual
      • Innate – if entities can spread the novelty.
      • Manual – if intervention is required to spread the novelty.

Growth, above all else.