What is it about?
When dealing with low-probability, high impact risks estimating probabilities becomes tricky. The reason is that there is always a chance that our theory of the world is wrong, that our model of the risk is mistaken, or that we do a calculation mistake. For unlikely risks this chance is much larger than the risk probability itself. This makes the risk estimate less effective in telling us the real risk than it looks. We demonstrate this problem with the debate about risks from particle accelerators and suggest that the way forward is to use multiple independent risk assessments.
Featured Image
Why is it important?
Low-probability, high impact risks are important because despite their low chance of happening they might do too much damage if they happen to be acceptable. Consider the risk of a physics experiment accidentally destroying the world, killing all 7.5 billion current people *plus* all future generations. Reasoning well about such risks, especially in domains where we do not know all the rules, is important for mitigating them.
Perspectives
We are often intellectually overconfident, believing our arguments to be perfect because *we* cannot see a problem with them. Unfortunately we are only human, and make mistakes all the time. When dealing with big risks the sensible thing to do is to take the chance of mistakes into account. While the paper is focusing on physics risks as a case study, it is applicable to a lot of other risks too. One example is finance: even if one's model predicts very low risk with an investment, one needs to take the risk that the model is wrong into account. It is nearly always a good idea to use independent approaches to check the answer to any problem: the chance of them all being wrong in the same way is lower.
Dr Anders Sandberg
University of Oxford
Read the Original
This page is a summary of: Probing the improbable: methodological challenges for risks with low probabilities and high stakes, Journal of Risk Research, March 2010, Taylor & Francis,
DOI: 10.1080/13669870903126267.
You can read the full text:
Contributors
The following have contributed to this page
