Natural phenomena are to a smaller or greater extent governed by uncertainty. Less pronounced but sometimes more evident, behaviours and markets are too. In its basic expression, risk is the product of probability and consequence, and arguably both factors in our business depend on natural processes, human performance, and – usually complex – industrial and financial synergies.
It is of great importance in uncertainty-focused engineering models to distinguish between randomness and knowledge-related uncertainty, or in other terms, aleatory and epistemic uncertainty. On one hand, the term aleatory uncertainty stands for randomness inherent in the given input to the problem. Thus it cannot be handled (reduced or modified) otherwise than statistically. On the other hand, epistemic uncertainty stands for lack of knowledge on the exact nature of an input to the problem. This type of uncertainty can indeed be reduced through an expansion of relevant knowledge; for example this can be scientific and technical development or accumulation of experience through expertise, monitoring and investigations, or data processing in available records. As knowledge on a system increases the nature of uncertainty flows from the cloud of the aleatory (irreducible) to that of the epistemic (possible to reduce, or even to eliminate).
Based on the characteristics of the problem, the following further classification can be made:
- Parameter uncertainty: epistemic in general
- Model inadequacy: unequivocally epistemic
- Residual variability: both aleatory and epistemic
But, whilst fundamental, the distinction between aleatory and epistemic uncertainties is not definitive. It only makes unambiguous sense in quantifiable problem-solving, when we ourselves categorize our resources and decide consciously on how to attribute the uncertainty characteristics of each input magnitude. For one problem an addressed uncertainty may be epistemic; in another one it may be aleatory.
It is, in the end a pragmatic choice; it depends on our resources, our means, our scope, and our purpose.
Some literature:
Ang, A.H.S.; Tang, W.H.:Modeling and analysis of uncertainties for risk-informed decisions in infrastructures engineering. Structure and Infrastructure Engineering. 1 (2005), pp. 19-31, 2005
Der Kiureghian, A.; Ditlevsen, O.: Aleatory or epistemic? Does it matter? Structural Safety, 31, (2009), pp.105-112, 2009
O’Hagan, A.; Oakley, J.: Probability is perfect, but we can’t elicit it perfectly . Reliability Engineering & System Safety, 85 (2004) pp.239-248, 2004
