Epistemic vs Aleatory Uncertainty
Some uncertainty comes from what we do not yet know and shrinks with study; some is inherent randomness that no amount of study removes.
Two Kinds, Two Remedies
Epistemic uncertainty is uncertainty from lack of knowledge: an imperfectly known parameter, an approximate model. It can be reduced by more data, better measurement, or better theory. Aleatory uncertainty is inherent variability, genuine randomness in the system that no amount of knowledge removes. Confusing the two leads to false expectations about what more study can buy.
Telling Them Apart
- Epistemic: would this uncertainty shrink if we knew more or measured better? If yes, it is epistemic.
- Aleatory: is this irreducible variation in the process itself? If yes, it is aleatory.
- Epistemic uncertainty is about our state of knowledge; aleatory is about the world.
- The boundary can shift: what looks random may become predictable as understanding grows.
Why the Distinction Guides Action
The two demand different responses. Epistemic uncertainty is an argument for gathering more information, running the experiment, refining the model, before deciding. Aleatory uncertainty is an argument for designing to tolerate variability, since it will not go away. Spending effort to reduce uncertainty that is actually aleatory is wasted; ignoring reducible epistemic uncertainty is negligent.
Keeping Them Separate in Analysis
Sound UQ often propagates the two separately, so a result shows how much of its spread could be narrowed by more knowledge versus how much is irreducible. Collapsing them into a single number hides exactly the information a decision-maker needs: whether waiting for more data would help, or whether the variability must simply be accommodated.
In Physical Modeling
In a fusion design, an uncertain material property is epistemic and shrinks with measurement, while turbulent fluctuation has an aleatory character that persists. Kronos UQ distinguishes reducible from irreducible uncertainty where it can, so effort is directed at what more study will actually improve and the rest is designed around.