Not all uncertainty is created equal. In fact, there’s an important distinction between types of uncertainty—one that originated in the medical field but has since become widely used in data science, artificial intelligence, and deep learning. This is the distinction between **aleatoric** and **epistemic** uncertainty. The names might sound intimidating, but the idea is simple: some uncertainty comes from the fundamental randomness of the world, while some comes from our ignorance. ## Aleatoric Uncertainty: The Uncertainty We Can’t Eliminate The first type, **aleatoric** uncertainty, comes from the Latin *alea*, meaning dice—which gives away its core meaning. This is uncertainty baked into the system itself, the kind you simply can’t get rid of, no matter how much knowledge or computing power you throw at it. It’s the reason why even the best weather models can’t tell you exactly what the temperature will be three weeks from now. The system is just too chaotic, too sensitive to initial conditions—hence the famous **butterfly effect**, where a tiny change at the micro-scale can trigger massive consequences down the line. This kind of uncertainty is **irreducible in practice** because our measuring instruments have physical limitations—we can’t track every particle, every gust of wind, every molecular interaction. But more profoundly, it is **irreducible in principle** because nature itself imposes limits on predictability. Michael Berry illustrated this beautifully in his 1978 paper Regular and Irregular Motion. He calculated that even something as seemingly straightforward as the movement of billiard balls would become impossible to determine after just **eight collisions**—not because of any failure in our equipment, but because of the minuscule gravitational influence of the billiard player. And if you take it further, the trajectory of an oxygen atom in a gas becomes indeterminate after about 55 collisions due to the gravitational pull of a single electron at the edge of the observable universe. [^1] In other words, some things will always be unpredictable—not because we’re not trying hard enough, but because reality itself too entangled and indeterminate. ## Epistemic Uncertainty: The Uncertainty We Can Reduce On the other hand, **epistemic** uncertainty—derived from the Greek *episteme*, meaning knowledge—refers to uncertainty caused by a lack of information. This is the kind of uncertainty that can be reduced by learning more, collecting better data, and refining our models. Take the example of a disease outbreak. At the beginning, we don’t know how fast it’s spreading, how deadly it is, or who is most at risk. This is epistemic uncertainty—we are simply missing information. But we can reduce it through large-scale testing, data analysis, and epidemiological modelling. However, even after we’ve gathered as much data as possible, some aleatoric uncertainty remains. Testing equipment has limitations—false positives and false negatives are inevitable. Doctors may misdiagnose patients. The virus itself may mutate unpredictably. No amount of knowledge will remove these sources of uncertainty completely. ## Why This Distinction Matters The difference between aleatoric and epistemic uncertainty isn’t just an academic curiosity—it has profound real-world implications. These two types of uncertainty behave in fundamentally different ways, and if we don’t distinguish between them, we risk making serious errors in reasoning. - **Epistemic** uncertainty can be reduced through research, measurement, and better decision-making. - **Aleatoric** uncertainty is part of the fabric of reality, something we have to learn to live with. In many situations, both types coexist. A company trying to predict market fluctuations must deal with epistemic uncertainty (e.g., unknown consumer behaviour, missing economic data) and aleatoric uncertainty (e.g., random geopolitical shocks, black swan events). A self-driving car navigating a busy intersection faces epistemic uncertainty (e.g., incomplete sensor data, unknown driver intentions) and aleatoric uncertainty (e.g., the inherently unpredictable behaviour of pedestrians or sudden weather changes). Recognising which type of uncertainty we are dealing with is crucial. If we mistake aleatoric uncertainty for epistemic, we waste resources trying to eliminate something that can’t be eliminated. If we mistake epistemic uncertainty for aleatoric, we resign ourselves to ignorance when, in fact, better information could have given us an edge. Uncertainty comes in different flavours. And understanding which kind we’re dealing with is the first step in navigating it wisely. [[Ignorance|Next section]] <hr> [^1]: MacKay, R.S., & Meiss, J.D. (Eds.). (1987). Hamiltonian Dynamical Systems: A REPRINT SELECTION (1st ed.). CRC Press. https://doi.org/10.1201/9781003069515