Engineers and mathematicians (and CEOs) want to reduce the world to simple models. The pursuit of simplicity and elegance is a noble one. I love refactoring code to improve abstractions and remove repetition. I get the same warm and fuzzy feeling when simplifying a business process. The enemy of simplicity is complexity. I think understanding intelligence requires a deep understanding of the difference between complex and complicated.

Complicated Systems

Complicated systems include machines like airplanes, computer programs, and other engineered systems. These systems, while intricate and detailed, can be understood through analysis. Complicated systems typically have a finite number of states with well understood state-transitions. That makes them predictable and deterministic. Engineers love that! Limited feedback loops, separation of concern, linear relationship between input and output. Complicated systems can be analyzed and understood. That makes us feel we can trust them. Obviously, the simpler the system the easier it is to trust, but complication does not generally prevent predictability.

Most of our academic training as children is focused on learning to solve complicated problems. Problems that are deterministic and linear. "Read this book and then answer these questions". Tasks do not generally have dependencies or unexpected feedback loops. This favours kids that have reliable storage (memory) and powerful compute (raw intelligence). Fortunately, school also includes a social dimension through which kids learn about complexity. A gang of teenagers learning to socialize exhibit all the markers of an interconnected, sensitive and non-linear system.

Complex Systems

Complex systems have a number of characteristics that make them different from complicated systems:

These are obviously terrible properties of a system if you like control and simplicity. Yet, anyone who has managed a team or built a company or run a political campaign knows that society is complex, not complicated. Groups of humans form complex systems. To make matters worse, a group of humans is a complex system made up of complex systems. And each human body is a complex system with countless interconnected components, feedback loops and emerging abilities.

People drastically underestimate how complex our bodies and brains are. I've recently taken up studying neuroanatomy and the first thing you realize is that talking about the brain gives you the impression that we process information in a single place. We don't. There is the peripheral nervous system, the spinal cord, the thalamus, the hypothalamus, the cerebral cortex and many more components. Each of these systems are complex chemical and biological systems in and of themselves. Together, they form a dizzyingly complex web of dependencies. There is layer after layer of fragile interconnectedness. Humans are easily killed because of all these dependencies. Small errors in our genetic programming can lead to catastrophic disabilities. On the other hand, when it works, it's marvelous.

Complex systems are usually not engineered. They are evolved through experimentation. Genetic mutation is the Original Gangster (OG) of experimentation. Ultraviolet radiation is the reason humans are intelligent 😉. "Errors" are introduced but turn out to be a better solution than the previously dominating solution. Startups are the genetic mutations of the complex system that is our free market. Founders like to think their skill is the main source of success but as with all mutation there is a huge amount of randomness. Equilibrium emerges temporarily, but as subsystems change equilibrium moves. And complex systems can operate far from their equilibrium for a long time, making it hard to tell if the situation is stable or not.

The universe does not lend itself to prediction to such a degree that we can engineer highly complex systems. We over-estimate our ability to predict and engineer the future, and under-estimate the amount of randomness. The alternative is that we are not in control, which is uncomfortable to most humans. The computational cost of modelling enough details to predict the future likely exceeds our available energy.

Anti-Fragility

I'm a big fan of Nassim Taleb and his concept of anti-fragility. He describes Black Swan events, i.e. rare, unexpected events with massive consequences. Such events are possible because complex systems are non-linear. If a complex system is shocked by unexpected information it can lead to cascading failures and huge impact. This is true for the human body, our financial markets and ecosystems.

Taleb suggests we should design anti-fragile systems, i.e. systems that get stronger if shocked. Anti-fragility requires redundancy and optionality as strategies to navigate complex systems. Decision-makers should bear the consequences of their actions to align incentives and encourages more responsible and prudent behavior. Humans are anti-fragile in the sense that we have strong incentives to look out for ourselves, we have many redundant sources of food and we are adaptable to changes in our environment. You can throw humans into completely new environments with different temperature, humidity and food sources, and a surprisingly large number will survive.

Intelligence

How is all of this connected to intelligence? Well, I think intelligence is a form of anti-fragility that has emerged through evolution as a defence against complexity. The human body, with its brain and nervous system, has increased its biological market share since it is able to withstand shocks and get stronger from them. We have evolved the ability to zero-shot learn complicated things while simultaneously fitting observations and experiences into models for vast, complex systems. We are resilient, adaptive and fast-learning.

Our current form of machine learning is better equipped to solve complicated problems. Problems that require storage and compute. The last few hundred years have promoted humans with good storage and compute. But those days are over now. We have invented tools that can amplify our storage and compute, or delegate it all together. What's left to humans now is handling complexity. Solving linear, deterministic tasks that require an analytical approach is rapidly becoming a commodity. This commoditization of storage and compute has been ongoing since the invention of the computer, and is exponentially accelerating. At one point, the ability to do arithmetics was highly sought after. Royal courts had scholars that memorized large amounts of text. Harvard had human computers. Today we all have a calculator in our pocket. Or our watch. All analytical work is going in that direction. Crunching numbers, matching patterns, extrapolating trends - all of that is becoming a commodity. If your job is to make decisions based on data, you are about to be automated. Fortunately, that is not at all a threat to humanity. We did not build a dominant position in our ecosystem on memorization. We built our position on the ability to navigate complexity. Lean into that.

What is my point? My point is that intelligence is not primarily about solving complicated problems. It's actually about solving complex problems. Decision-making under uncertainty in unpredictable environment with a lot of dependencies. Humans remain the uncontested champions of handling such challenges.