Observation selection effects and global catastrophic risks

Abstract

Different types of global catastrophic risks (GCRs) are studied in various chapters of this book by direct analysis. In doing so, researchers benefit from a detailed understanding of the interplay of the underlying causal factors. However, the causal network is often excessively complex and difficult or impossible to disentangle. Here, we would like to consider limitations and theoretical constraints on the risk assessments which are provided by the general properties of the world in which we live, as well as its contingent history. There are only a few of these constraints, but they are important because they do not rely on making a lot of guesses about the details of future technological and social developments. The most important of these are observation selection effects. Physicists, astronomers, and biologists have been familiar with the observational selection effect for a long time, some aspects of them (e.g., Malmquist bias in astronomy or Signor-Lipps effect in paleontology) being the subject of detailed mathematical modelling. In particular, cosmology is fundamentally incomplete without taking into account the necessary ‘anthropic bias’: the conditions we observe in fundamental physics, as well as in the universe at large, seem atypical when judged against what one would expect as ‘natural’ according to our best theories, and require an explanation compatible with our existence as intelligent observers at this particular epoch in the history of the universe. In contrast, the observation selection effects are still often overlooked in philosophy and epistemology, and practically completely ignored in risk analysis, since they usually do not apply to conventional categories of risk (such as those used in insurance modelling). Recently, Bostrom (2002a) laid foundations for a detailed theory of observation selection effects, which has applications for both philosophy and several scientific areas including cosmology, evolution theory, thermodynamics, traffic analysis, game theory problems involving imperfect recall, astrobiology, and quantum physics. The theory of observation selection effects can tell us what we should expect to observe, given some hypothesis about the distribution of observers in the world. By comparing such predictions to our actual observations, we get probabilistic evidence for or against various hypotheses.