A critical look at risk assessments for global catastrophes.

Recent articles by Busza et al. (BJSW) and Dar et al. (DDH) argue that astrophysical data can be used to establish small bounds on the risk of a "killer strangelet" catastrophe scenario in the RHIC and ALICE collider experiments. The case for the safety of the experiments set out by BJSW does not rely solely on these bounds, but on theoretical arguments, which BJSW find sufficiently compelling to firmly exclude any possibility of catastrophe. Nonetheless, DDH and other commentators (initially including BJSW) suggested that these empirical bounds alone do give sufficient reassurance. This seems unsupportable when the bounds are expressed in terms of expectation value-a good measure, according to standard risk analysis arguments. For example, DDH's main bound, p(catastrophe) < 2 x 10(-8), implies only that the expectation value of the number of deaths is bounded by 120; BJSW's most conservative bound implies the expectation value of the number of deaths is bounded by 60,000. This article reappraises the DDH and BJSW risk bounds by comparing risk policy in other areas. For example, it is noted that, even if highly risk-tolerant assumptions are made and no value is placed on the lives of future generations, a catastrophe risk no higher than approximately 10(-15) per year would be required for consistency with established policy for radiation hazard risk minimization. Allowing for risk aversion and for future lives, a respectable case can be made for requiring a bound many orders of magnitude smaller. In summary, the costs of small risks of catastrophe have been significantly underestimated by BJSW (initially), by DDH, and by other commentators. Future policy on catastrophe risks would be more rational, and more deserving of public trust, if acceptable risk bounds were generally agreed upon ahead of time and if serious research on whether those bounds could indeed be guaranteed was carried out well in advance of any hypothetically risky experiment, with the relevant debates involving experts with no stake in the experiments under consideration.