resources Article

Obtaining Wisdom
Through Simulation

Although wisdom sits at the apex of the knowledge hierarchy, it is often the aspect that is least discussed. Wisdom should be the ultimate goal of any analytical endeavor. Scrutinizing data lets us explore structures, spot patterns, and understand relationships. However, insight from data alone is insufficient to make wise decisions when we are confronted by new and unseen scenarios. Why is this the case? And how do we gear our analysis towards obtaining wisdom?

Computational simulation adds a layer of wisdom to our understanding of complex systems.

The knowledge hierarchy

There are four levels on the classic knowledge hierarchy, with each a necessary condition for progress up the pyramid.

Data is not information. Information is not knowledge. Knowledge is not understanding. Understanding is not wisdom.

The first and most familiar level is data — the raw facts of any system. Next comes information. At this level we give meaning to our data, we understand where it comes from and how it fits together. Third comes knowledge. We begin to understand the context of our information and relate it to the real-world processes which have generated it: we move from observation and organisation of facts, to understanding.

However, merely understanding our environment is not enough. To become better decision makers, be it in our day-to-day lives, or in the boardroom, we need to turn knowledge into wisdom.

Knowledge is knowing a tomato is a fruit; Wisdom is not putting it in a fruit salad.

Wisdom is what lets us take our knowledge of our system and apply it to previously unseen scenarios in order to make the right call:

Three routes to wisdom

1. Reflection

The bulk of modern data analytics focuses on this method. It hopes to uncover wisdom from historical data generated by a system. This can be a fruitful strategy: the clever folks at Deepmind were able to reduce Google’s data centre cooling bill by 40% by applying machine learning techniques to optimize efficiency.

2. Experience

Many of us learn life’s lessons through experience as children. It doesn’t stop here: experiential learning continues into the boardroom. The Global Financial Crisis taught finance executives painful lessons about the workings of the global financial system, while the Deepwater Horizon disaster taught oil executives hard lessons about safety and overconfidence. This is a costly way to learn: the cost to BP alone of Deepwater Horizon costed around $65 Billion. Meanwhile, the cost of the global financial crisis has been estimated at a staggering $22 Trillion!

3. Imitation

We train our fighter pilots in flight simulators. These accurate representations of real flight give learners room to fail cheaply and safe from danger. Having fighter pilots gain their wisdom through experience would be too costly, both in financial terms, and also in terms of human lives.

Simulators provide an easy and cost-effective way to train pilots against any number of scenarios. By changing the settings on the simulator, the pilot can train repeatedly against low-probability, high-impact scenarios — such as engine failure — that they might expect to encounter at most once in a career. Learning to fly a fighter jet is a costly domain. However, as the earlier examples earlier prove, so too is learning to run a corporate from the boardroom.

Simulation for enterprise

A flight simulator is simply an accurate model of the physics of a real aircraft: a recreation of thrust, lift, and aerodynamic properties. This is coupled with a visualization layer that mimics how pilots see the world when it comes to flying.

We can think about building ‘enterprise simulators’ in the same way. Firstly we need to build accurate models of our domain — to create a virtual environment. And then we need useful visualizations to explore that environment.

We need to ask ‘what-if ?’and to gain wisdom through exploration of scenarios.

Let’s explore these two steps in a bit more detail:

Step 1: Building accurate simulation models

Agent-based modeling (ABM) is a powerful and flexible tool for building domain-agnostic models of our world. An ABM is a bottom-up modeling approach, where the high-level properties of a system are generated from the low-level interactions of its constituent parts, or agents. An agent is an abstraction which can be used to represent just about anything: a financial contract, an individual, a bank, or even an entire country. ABMs are expressive enough to generate emergent phenomena of the type that characterize Complex Adaptive Systems (CAS). Cities and financial markets are archetypal examples of CAS. In reality, almost any large-scale organisation or system will behave in a similarly complex manner.

These models allow us to build rich and expressive models which are grounded in the micro foundations of a system. They also provide large and rich parameter spaces which we can explore to study what-if? scenarios. For a bank, we can study what happens if depositors withdraw their funds, if the central bank changes policy, or what happens if the bank were to change its pricing strategy.

These types of questions are very hard to answer for previously unobserved scenarios using traditional approaches. This is where we find need for coupling agent-based models with massive-scale simulation — to generate millions of realistic possible scenarios. This empowers decision-makers who are faced with an unpredictable future to move beyond knowledge of the past towards wise decisions.

Step 2: Exploring the virtual environment

The visualizations that boardroom executives and corporate decision-makers have to hand don’t need to be as rich and immersive as they do for fighter pilots. Most of the time, standard statistical visualizations convey the information required to make informed decisions. The goal is to couple high-fidelity simulation with a rich and informative visualization layer.

Agent based models allow decision makers to pick the scenarios that they want to learn against. Just as the fighter pilot wants to practice dealing with engine failure, a banker will want to know how best to respond when savers start queuing at high-street branches to withdraw funds. They can tweak the input parameters to the model of their domain in an attempt to avert disaster. And based on experience gained in the simulation, exploit double-loop learning. What is so crucial is that all of this can be done in a controlled virtual environment. By simulating shocks, financiers can learn how to make radically better decisions, and become wiser through this process.

Advances in complexity science and computational power have made this type of advanced simulator possible. The technology to build these tools already exists. Today’s tech allows you to explore every possible scenario, simulate every possible future, and empower every possible decision.

Chloe Hibbert