You've successfully subscribed to Safety is Dangerous
Great! Next, complete checkout for full access to Safety is Dangerous
Welcome back! You've successfully signed in.
Success! Your account is fully activated, you now have access to all content.
Success! Your billing info is updated.
Billing info update failed.

All Models are Wrong, but Some Models are Useful. Sometimes.

Our perception works by creating and interacting with models. They're used a lot in business too, but all models are inherently wrong.

Simon Haighton-Williams
Simon Haighton-Williams

At a fundamental level, models are part of the fabric of cognition. Reality is entirely subjective. Our perception works by creating and interacting with models. They're used a lot in business too, but all models are inherently wrong. Or at least imperfect. They aren't reality; they're a proxy for it. But sometimes we forget that, which can create problems.

Models are inherently wrong because they don't involve the same level of detail or variance as reality. They are a way to understand what's going on, plan our response to it and examine the results of our actions. Using and accepting models can be helpful when we don't want to burden ourselves – or others – with the task of needing to know everything about a scenario.

Models aren't reality

The problem with models is that they aren't reality. They don't possess the same fidelity as reality, and they are limited to a particular scope. They can't cover every externality, and some things aren't deterministic enough to be modelled anyway. They're imperfect, too, when it comes to predicting the future. They can only operate based on what we already know, whether it's behaviours, interactions, costs, capabilities, processes and so forth. The gap between the model and reality is where the potential problems arise.

Models in reality: Flipping a coin

A good way to think about the difference between a model and reality is to think about what would happen if you flipped a coin ten times. The model in your head tells you that you'll see five heads and five tails. But what do you think will happen? If we set up a wager that said I'd pay you £10 if there's an equal number of heads and tails, but you pay me £20 if there aren't, would you take the bet?

What about the same bet over 100 flips? It's pretty easy to see that, whilst the model seems sensible, it doesn't take much for reality to vary from it. Statistician George Box coined the phrase 'All models are wrong, but some are useful'. I'd go a step further and say that some models are useful, but only sometimes.  

So why do we use models

If you've studied any of the sciences from primary school to degree level or beyond, you'll surely appreciate that the same phenomena are explained – or modelled – in strikingly different ways at different stages. The consequences of expecting primary school children to understand particle physics, for example, are pretty obvious. However, even advanced models in physics are only that – models.

Newton's laws, for example, are reliable and accurate at human scale. This means that we're happy to fly in aeroplanes. However, they don't hold true at a larger scale (astronomical) or a smaller scale (particle level). Einstein's models need to be observed instead. However, even Einstein's theories are still models, ways of making sense of reality. Scientists even treat the same phenomena with different models – there are both particle theories and wave theories for explaining the behaviour of light.

When they're useful

If we look at a more everyday analogy, we can see how models and reality don't need to be the same. Harry Beck's Tube Map is regarded as a classic of information design. The London Underground stations it displays don't correspond to their geographical locations. In the context of helping people get around London, it doesn't matter because the model makes it easier to navigate. However, if you wanted to provide construction workers with a location for drilling a new access shaft, the Tube Map is woefully inadequate.

Before Harry Beck's feat of information design excellence, just navigating the map of the London Underground to work out how to travel between two points was cumbersome. Imagine if you didn't have a model (map), though. The task of just getting to the right neighbourhood would be lengthy, complicated and frustrating. Models can, of course, be useful, though they aren't always so good at detail and edge cases.

Context, purpose and value

It seems obvious to say it, but context and purpose are crucial when using models. Models can help individuals, teams and even organisations communicate and share understanding. They can save time and effort, but thought needs to be given to how and when they're used. Their most valuable role is to make it easier to formulate plans and effect change.

The value of a model is a function of several factors. How accurate it is – or needs to be – will usually determine the cost of building it and maintaining it. How often it will be used, by how many people, and how big the challenge or opportunity is, likely to determine the benefits of using it.

Effectively, a shared model is like a contract. It says that we're agreeing that the model is how we're going to think about a scenario because we want to interact with it and make a change. We're accepting that we expect particular actions to result in specific outcomes. Like with all contracts, we should be conscious that we're entering into one.

Because reality

The problems with models arise if we forget that the model isn't reality. When we forget that reality – or our understanding of it – can change without the model changing, there are consequences. Of course, they may be insignificant, but they could be costly.

Models can be very convincing, and we often like to think that having a model means that we have understanding. Remembering that all models are wrong reminds us to look at what's happened unexpectedly and decide whether our model can be improved or whether the outcome has been acceptable

Models should change too

We need to be conscious and aware of what we're doing when we create and use models. The model won't necessarily last forever, so we need to test the model, interrogate the underlying assumptions and give it an expiry date.

Don't stick to the model when it's no longer helpful. A broken, simplistic, or simply outdated model can drag you off course as fast as anything else. We shouldn't be afraid to dispense with existing models and shouldn't follow them slavishly. Not following the model shouldn't necessarily be seen as a transgression. Not testing, challenging and improving the model with real-world learning should, though.

Decision Making

Simon Haighton-Williams

A veteran of multiple tech and tech-enabled businesses, Simon has led Adaptavist Group since 2010. He can usually be found where organisations, people and technology intersect.