We believe we describe the world; in fact, we predict it. Every judgment we make – about a colleague, a customer, an organization – is an implicit bet on what will happen next. This mechanism, invented by Sapiens to convert uncertainty into risk, explains both our extraordinary capacity for action and our current disarray: our prediction machines are still running, but they predict less and less well.
You are at the office and you pass Paul in the hallway. You find him friendly – you have always found him friendly. Without even thinking about it, you expect him to greet you warmly, as usual. But this morning, Paul walks past you without a word, his face closed. You are surprised. And that surprise, trivial as it seems, reveals something fundamental about how we work: you were not observing Paul, you were predicting him. Or rather, your brain was predicting him for you.
Every model is a prediction machine
We tend to think of our judgments as descriptions: “Paul is friendly” looks like a statement about Paul. In reality, it is a prediction about him. Saying that Paul is friendly means expecting him to behave accordingly with me – to greet me, to help me if I ask, not to criticize me behind my back. The judgment is phrased in the present, but it works on the future. It tells me what is going to happen.
This is true of all our mental models, the individual and collective beliefs through which we read the world. “A good manager masters all the data in his scope” is not just a belief: it is a prediction about what will happen to the one who shows up at a meeting without having checked everything. “Our customers are loyal” is not an observation: it is an anticipation of their buying behavior. The mental model is not a photograph of reality, nor even merely a representation of it; it is a simulator. It runs constantly, silently, producing expectations about what will happen – allowing us to act without having to recompute everything at every moment. It is a prediction machine, remarkably effective because it is implicit and largely unconscious.
Sapiens’ invention
Why does our brain work this way? Go back to the savanna, tens of thousands of years ago. Something moves a hundred meters away. As long as you do not know what it is, you face pure uncertainty: the unprecedented, the unclassifiable, something against which no action stands out. Your survival depends on your ability to connect “what moves” to a known category in your model. If it is an antelope, I know what to do: I hunt. If it is a lion, I know what to do: I stay away. Once the category is found, everything changes. You are no longer facing the unknown; you are facing known scenarios, with their implicit probabilities and their associated responses.
This is exactly the distinction economist Frank Knight formalized a century ago: risk is what can be assigned probabilities – the possible cases can be listed and acted upon because they are known and catalogued; uncertainty is what cannot – we do not even know the list of cases; it is the unprecedented, the new situation. Risk can be managed; uncertainty cannot. But assigning probabilities requires a precondition: you first need a list of possible cases. And that is precisely what the mental model provides. It closes the world. It turns the infinity of possibilities into a finite repertoire of recognizable situations. It focuses us on the situations it deems essential, at the expense of those it deems secondary.
The mental model is thus the tool Sapiens invented to convert uncertainty, which cannot be managed, into risk, which can. It is his secret weapon in the war of species. Other animals have reflexes; Sapiens has models, and above all shared models, which multiplies their power: if we believe the same things, I can predict your behavior and you can predict mine. Thousands of individuals who do not know each other can cooperate, because the collective model makes each one predictable in the eyes of the others. The model does not only reduce uncertainty in our heads; it reduces it in the world, by coordinating our actions.
The conversion is a bet, not a guarantee
But there is a downside, and it is essential to see it. The conversion performed by the model is not a definitive operation: it is a bet. When I classify the silhouette as an antelope, I bet that my list of cases is complete and that my category is the right one. When I hold Paul to be friendly, I bet that his past behavior predicts his future behavior. The model does not make uncertainty disappear; it puts it in brackets. It acts as if the world were known; and as long as the world cooperates – that is, as long as my model matches reality – the bet works. We then live in the comfort of the obvious, without even noticing that we are betting.
It is when the world stops cooperating that the bet fails and reveals itself. Paul walks past without greeting me: my model is in default. Surprise is nothing but that: the signal that reality has just stepped outside my list of cases. This is why surprise is such valuable material for anyone who wants to understand an organization: it reveals, in negative, what was taken for granted. Tell me what surprises you, and I will tell you what you believe. I use this a lot in seminars.
Faced with a surprise, two paths open up. The productive path: adjust the model. Perhaps Paul is not “friendly” in general, but friendly under certain conditions; perhaps he is going through a hard time; perhaps my model of Paul was too simple. The sterile path: protect the model against reality – “Paul must have slept badly,” and we think no more of it. Rationalization is the mechanism by which we prefer to save the prediction – that is, our model – rather than learn and modify it, because modification is costly. In small doses, rationalization and ignoring surprises are economical; in large doses, they cut us off from reality. The limiting case is the one Karl Weick studied with the Mann Gulch fire: an event so far from any existing category that the firefighters’ collective model collapsed at once – and with it their very capacity to act. Without a model, no prediction; without prediction, no action.
Uncertainty is not in the world; it lies between the world and our models
This understanding reverses the usual way of talking about uncertainty. It is presented as a property of the world: “the world has become uncertain.” That is not quite right. The world has always been changing. What varies is the ability of our models to absorb that change. Uncertainty is not a property of the world; it is a property of the relationship between our models and the world. It is low when our models correctly convert the flow of events into recognizable situations; it explodes when they become obsolete, because the filter no longer works, or even becomes counterproductive. They must then be modified, or even reinvented, and that requires a creative phase.
This is exactly what we are living through today: historical models becoming outdated – about work, geopolitics, technology, climate – and new models emerging. The discomfort we call “uncertainty” is that of the in-between: our prediction machines are still running, but they work less and less well. We are all, in a way, standing in front of a growing number of friends who no longer greet us, and anxiety takes hold because the world no longer makes sense.
The good news is that if uncertainty arises from our models, it is also through them that it can be worked on. Not by seeking ready-made new certainties – the first reflex, and the riskiest – but by making our models visible, by taking our surprises seriously instead of rationalizing them, and by accepting that every judgment is a revisable bet rather than an established truth. The model will always be our prediction machine; the point is not to do without it, which is impossible, but to know when it predicts and when it is merely betting without telling us.
🇫🇷 A version in French of this article is available here.