Some thoughts on the complexities of our world

In the last few days, I have finished three books:

• The Fabric of Reality by David Deutsch
• Komplexe neue Welt by Marco Wehr
• The Maniac by Benjamín Labatut

All of them admired Kurt G?del, but they were making different cases: Marco Wehr was saying that our world is getting more complex, while David Deutsch is saying that our understanding of the universe is growing more coherent through unifying theories. We can understand more with less theories.

The Maniac is an outstanding biography of John von Neumann who is often considered to be the smartest person of the 20th century. For example, in 1948 he showed (theoretically) that a self-reproducing machine has to have a mechanism of copying the machine but also of copying the information that specifies the machine. This is what DNA and RNA do, and it was only discovered much later. (There is this nice talk about the history of ideas by Sydney Brenner who was awarded the 2002 Nobel Prize in Physiology or Medicine for his work on genetic regulation.)

Analog and digital scripts are running the world

The digitalization started some 50 or 60 years ago and there is a reason it won’t finish any time soon: Collectively, we run trillions of scripts every day.

I understand a script as a sequence of steps to execute a certain task or operation, such as:

• Delivering newspaper to apartment number 47.
• Organizing transplant organs.
• Ordering a new logo from a designer based in Vietnam.
• Writing the shareholder agreement for a new company.
• Paying for the bread in the bakery.
• Invoicing a client for the services delivered in the last month.
• Sending a damage report to the insurance company.
• Preparing an ESG report for an investment company.

Many of the scripts are crystal clear and we can create complete contracts. That is the crypto world where “code is the new law”. (Katharina Pistor wrote an interesting book on “The Code of Capital” which explains the coding of assets in the real world). A typical case would be a loan secured by a crypto collateral which will be liquidated in case of a defined default case. That can all be done in a fully pre-determined fashion.

The other extreme is best captured by Helmut Qualtinger describing my native Austria:

Austria is a labyrinth where everyone knows their way around.

There will always be a need for humans to navigate these labyrinths.

There are limits for online dispute resolution mechanisms. Ast and Deffains have outlined the history of the online dispute resolutions industry and found the first examples in the 1990s with iCourthouse. eBay tried a crowd-based model to resolve user disputes in the 2000s. Obviously, one of the key challenges is that rulings were hard to enforce in private settings, as only public courts can use the police to enforces rules.

The structure of scripts

Let us assume that there are really trillions of scripts which are executed every day. A script typically starts with a push or pull request (“a trigger”): “Do task A” or “Send me information B”. The individual then needs to extract data from somewhere to perform the specific task and load it in some form of database, which can also be his or her memory.

For example, if you order at your local pizzeria, you read the menu, memorize the drink and the pizza until the waiter takes your order. If you want to take a vacation, you will start researching potential destinations and start putting them together in a document.

Using computer science terminology, it might look like the illustration below. I am not saying that it makes perfect sense to use ELT (Extract, Load, Transform) to describe the structure of scripts, but it is a nice approximation.

How to measure complexity?

I have kept asking myself how we can measure complexity?

Shannon Entropy is an interesting concept. It basically says: The more unpredictable or random the outcomes, the higher the entropy. Conversely, the more predictable the outcomes, the lower the entropy. A coin toss is random, while the scheduled times of public transportation services are highly predictable and have low entropy.

It is not really applicable, but I have the feeling that our lives are becoming much more predictable.

I also like the idea behind computational complexity (or algorithmic complexity) which looks at the time and space complexity of algorithms: How long does an algorithm need to solve a certain problem and what is the needed space to solve this problem?

Complexity today

Let us now apply the general idea of computational complexity to the random list of tasks from above. All except the newspaper delivery are much less time-consuming than 10, 20 or 50 years. Some would not even have been possible a few decades ago:

• Delivering newspaper to apartment number 14 (it has not changed much in the last decades)
• Organizing transplant organs (using market exchanges as outlined by Alvin Roth)
• Ordering a new logo from a designer based in Vietnam (using online platforms)
• Writing the shareholder agreement for a new company (using generative AI)
• Paying for the bread in the bakery (using digital payments)
• Invoicing a client for the services delivered in the last month (using e-mails and digital signatures)
• Sending a damage report to the insurance company (using apps and e-mails)
• Preparing an ESG report for an investment company (using tools based on symbolic AI)

Increasing complexity

I do not want to sound too simplistic as there are obviously areas which are very different than in the previous century. I can think of the following areas and there are surely much more:

• Industry cadence
• Push and pull requests
• Probabilistic black boxes

Industry cadence

?Byrne Hobart described the following observation in a newsletter:

It's popular to say that every company is becoming a tech company, and to some extent that's true. But companies need to be judicious about evolving towards sectors of the economy that operate on a faster cadence than they're used to.

Some industries are operating on a very slow schedule. A typical fashion retailer needs to order stock for the fall season probably 6-8 months earlier. This makes sense: textiles need to be ordered, factories scheduled, and shipping containers booked.

Some fashion companies are now producing on much shorter schedules. Zara has cut the production time to 2-3 weeks, while the total production time from concept to final product is 2-3 weeks for Shein.

There are many areas where the instantaneous availability of data has led to a sharp reduction in the time to complete a cycle (however it is defined). However, it is not clear that an increase in the cadence of an industry increases the complexity (especially when tasks are becoming more digital at the same time).

Push and pull requests (triggers)

The tasks described above are also much stronger linked than before. Scripts often need to pull data from other sources to complete their tasks.

This might be relatively straightforward for many topics, but it becomes incredibly complex for other areas.

I have been involved in a study on “Impact investing in the framework of business and human rights” for the European Parliament and find it difficult to assess the role of businesses when it comes to human rights.

One of my most interesting research trips was a visit to Bangladesh to see how things have developed since the collapse of Rana Plaza which was one of the greatest tragedies of the last decade (see picture below).

I had the opportunity to visit a few factories and meet the owners. Some of the factories were urban paradises (nobody ever believes that). There were little parks, free education and school uniforms for the kids, air conditioning and the products were very high quality.

I also walked in some backyard factories. Obviously, the factory is not following Western standards, but the people seemed happy and the guard not too frightening. How can you robustly ensure that you are not violating human rights in the supply chain when you have factories with a few workers in a backyard?

That is one of the tasks I find complex, as you need a lot of computing time and space to complete it.

Probabilistic black boxes

Much of our world is already a black box and it will get more "black-boxy" over time as a significant part of all tasks will be completed by neural networks in the near future. Neural networks are the opposite of the blockchain or distributed ledger technology as the output is probabilistic in nature and hard to explain.

It is driven by the nature of the neural networks which follow a different approach. You provide input data and output data and let the neural network approximate the function connecting input and output.

Looking at the computation paths of a simple neural network it becomes obvious that the output will never as deterministic as a traditional algorithm.

What does that mean in general?

1. We can execute much more tasks than before. Think of fraud detection, automatic sorting of handwritten addresses, the writing of meeting summaries or chatbots. That greatly increases the total output.
2. However, we do not understand how this output is calculated as the underlying neural networks are a black box for outsiders. By the way, that is also a part of the story in “The Maniac”, as we do not understand how AlphaGo really plays beyond a rough intuititon. Obviously, this is a source of complexity, as someone needs to take it into account.

Outlook

In some ways, entrepreneurs look for complexity and develop business model to tackle these complexities and make them manageable. It could be an interesting idea for a VC fund to identify the sources of complexity and look at the emerging business models, which are surely interesting investment cases.