Working on mechanical bees
The evolution of complex behaviour emerging from systems and its relationship with technology and innovation
Consider 4 types of organizations or in other words autonomous systems. For all 4 systems the core defines what type of actions the edges will be able to take.
Type 1: Organizations where humans and technology work together as one. There is no difference between the core and the edges. It’s a soup of technology and humans. The soup is making the product. An example is classic old school companies that have been around for ever.
Type 2: An organization where the core is formed by this soup of technology and humans and the edges consist purely of automation. Something automated is making the product. An example is car manufacturing companies where the construction process is fully automated or TikTok where the algorithms essentially create the final product by learning over time. For this organization to exist the core needs to build/use edges which execute defined tasks on command.
Type 3: An organization where the core consists of automation and the edges consist of the soup of humans and technology. The soup is making the product. Some examples are Bitcoin or fully decentralized DeFi protocols.
For the core to exist, we need to build/use a structure consisting of nodes, which together form a network of nodes, which when zoomed out look like an homogeneous entity. In reality each node consists of type 1, type 2 or type 3 systems, therefore the more nodes there are the more it makes sense to call the core of the organization to be an entity on its own.
For the core to become an automated entity, one needs to put in place a system which allows the different nodes of the network to communicate, change and agree on the different states of the network. In other words one needs to achieve distributed consensus. For that, there exist different types of consensus mechanisms.
For the organizations to exist, the edges need to interact with the core and produce certain outcomes.
Type 4: An organization where the core and the edges consist of automation. No other human or technology needs to interact with it for it to exist and keep existing for a certain amount of time. This self sustaining system is making the product. An example is full AGI (Artificial General Intelligence)
Think about these organizations in the form of systems made up of smaller elements forming the whole. Through the following examples see how these systems can behave in extremely complex ways.
Complexity within the system can occur in all 4 types. This can easily be demonstrated through negative externalities these systems caused in the past for type 1 & 2.
Type 1': Financial crises in 2008. On the most basic level, the financial system is an ensemble of different institutions (nodes) that transact (connectors between the nodes) between each other. The ensemble is a network that constantly updates itself over time. 2008 showed that complex financial products lead to extreme complexity within the whole system which resulted in its collapse. A soup of institutions corresponding to the first type of organizations managed to create such complexity from within the system that most observers or participants of the system were not able to predict what would ultimately happen.
Type 2': Polarization through algorithms used by social networks. Whether it is Instagram, Facebook or Twitter, the polarization of social network users is real. In the past years it was most noticeable during elections, Covid19 and other wide ranging topics. Who would have predicted when Facebook started that it would have such an effect on its users one day? Pretty much no one. The complexity of type 2 systems such as social networks is going through the roof and let’s see what TikTok has in store for us.
Type 3 & 4 haven’t shown negative externalities emerging from complexity within the system yet.
Type 3': Ethereum: Complexity is gently emerging not due to its underlying design or how blockchains work in general but the type of applications that are built on top of the Ethereum blockchain and the emerging ecosystem. I strongly believe that the complexity within such smart contract ecosystems will become more complex than any other type of organization we have seen so far.
Type 4': AGI. Once a fully self-sustaining AI exists which evolves and improves by itself, the complexity of its actions and behaviours will go through the roof. In the same way that we can not read the mind of another human being and anticipate its behaviour, we will not be able to predict this AI’s behaviour.
Complexity exists within the 4 types of systems. It seems however that the more autonomous and technologically advanced the system is, the higher and more frequent the amount of complexity we see in its resulting behaviour.
Why does this complexity actually matter?
It seems to me that the most powerful and interesting systems will become more complex in its overall behaviour. To illustrate, it is of bigger advantage for a company to build a system that fits “everyone’s” behaviour instead of convincing people to behave a certain way. In other words the behaviour of the system should not be repetitive, uniform or nested but instead should show complex behaviour.
Imagine selling apples where each produced apple turns out to have the buyer’s favourite colour. To make something like this happen, the behaviour of the apple producing system will need to be more complex than the behaviour of a system that only makes red apples (uniform) or makes red and green apples alternatively (repetitive) or a system that makes red apples and on randomly chosen days it makes yellow apples for 24 hours and then goes back to making red apples (nested). We like them apples.
Notice that we are talking about the behaviour of the apple producing system and not the underlying rules that make the system behave in a certain way. In fact, to achieve great complexity in the behaviour of a system, some argue that the underlying rules might just need to be over a certain complexity threshold. Once the rules are above that threshold, the same kind of behavioural complexity could emerge in any system. This threshold being so low that the underlying rules can be said to be simple.
Also notice that showing the same amount of behavioural complexity doesn’t mean that the system actually does the same thing. My apple producing system is not going to start making peaches suddenly for instance. We are just talking about the amount of complexity in a system. The underlying rules can in fact contain specificities and details, defining certain types of behaviours (such as peaches vs apples), while remaining simple.
Complexity in the behaviour of a system is not necessarily related to the complexity of the system’s underlying rules.
The complexity emerging from successful internet companies: Zooming in on physical processes & technological advancement
At the origin of complexity are physical processes which are “the natural forces that change Earth’s physical features” and I believe that they are the underlying forces that define our tendencies to display certain behaviours. The way we act based on these underlying physical processes is what some call freedom of choice.
Examples of physical processes: belonging, eating, exchanging, understanding, feeling, communicating etc
These are physical processes that seem to exist wherever we look in our universe. No matter how broad or narrow we look at the definition of these words we will realize that these things are present everywhere in some form or another. For instance “exchanging”: me and you might exchange money for an apple while bees exchange the work of pollinating for food, which they are getting from the flowers.
Innovation and technology allows us to go into greater and greater detail of these physical processes, identify the underlying elements that form them and then apply certain technologies to create new types of systems around those more detailed and more precise elements. Going into greater detail is essentially zooming in on systems to discover other systems and so on. All of these systems consist of nodes and connections between the nodes.
The big winners of the last generation of internet companies were at the time of their inceptions, innovative and new systems. Their founders zoomed in on the existing systems, which were based on physical processes, to discover smaller elements that they could connect through new technological progress that had been made, namely at that time the internet and the possibility to create global networks.
Table 1: Read the table from left to right
This table was made from the top of my head so don’t take this as facts. It’s content is to explain to you my train of thought and to illustrate my intuition behind it.
I will now show what type of systems I believe will have a large impact in the world we (western middle class people) live in, in the near future.
A mental model for the near future
At this moment I believe that AGI (type 4) is not something we should spend time on and waiting for. I am not talking about the dozens of subfields of AI which are more than relevant and super exciting, but that is a completely different subject. In fact, whether it is language processing, robotics, machine learning or others, the type of organizations involved in those fields are type 2 systems. What I am saying is that if we think in the form of systems, what we want to look at is type 3 systems that I described at the beginning of this post. Also, it is quite probable that to get type 4 systems right one day, we need to explore type 3 systems first.
In other words, we have been exploring type 2 systems for a long time already. From an investors perspective that area is overcrowded. Type 3 systems are just getting started and to get type 4 systems right we need to take care of the type 2 and type 3 systems first.
Another argument in favour of type 3 systems becomes obvious when looking at the macro environment we are in right now. With extremely complex problems such as geopolitical tensions, polarization within democratic countries, Covid19 and disagreements on regulations, climate change, inequalities and rising wealth discrepancies, systems that can produce extremely complex behaviour could be a good starting point to look at for lasting solutions. We will for sure not be able to solve highly complex problems with simple and naïve approaches. Whether our goal is to change the world simply by building cool new things that people want or to tackle some of the issues we discussed here, systems that can produce more complex behaviour seem to be the way.
Finally, I want to present to you a small framework that might help to understand what we need to create a system showing complex behaviour and hopefully to have a significant and large scale impact in this world. I will not go into detail but I believe the following bullet points are good to start when thinking about what we should be aiming and looking for.
- Define an initial purpose based on cool things you want to be a part of or important issues you want to help solve.
- Use existing physical processes to zoom in and define smaller and smaller elements that can be used thanks to the past advancements in technology.
- Build network systems with clear and simple initial rules. As an example, the last column in Table 1 shows the initial and main tools that the corresponding companies used to define and create their rules.
- The constraints of the network are defined by technology, so don’t add unnecessary constraints.
- Simple rules and few constraints are a good setup for complex behavior to emerge over time.
- Achieve distributed consensus within the network by setting up one of the existing or some new type of consensus mechanism.
- The network is now autonomous and will evolve over time through some type of work coming from within the network itself. If done right, which differs from case to case, complexity will emerge.
This is it for now. Thanks for taking the time and don’t hesitate to reach out for feedback and/or comments.
If you are still wondering why the title of this post is “Working on mechanical bees”: Yes it’s a complex problem but I also think this video is funny.
Recommended readings
https://blog.ethereum.org/2014/05/06/daos-dacs-das-and-more-an-incomplete-terminology-guide/
A new kind of science — Stephen Wolfram
