Databases of existential risk estimates

Key concepts

This tool is a database of existential risk estimates that have been produced by different researchers for different purposes.

As originally conceived, this focused on the estimates for the likelihood of existential hazards, where hazards are events or purposes with the potential to cause (in this case existential quantities of) harm. In order to actually cause this harm, hazards must be combined with at least one vulnerability (an element susceptible of being harmed by this hazard) and exposure (an interface through which the hazard and vulnerability connect). The focus on existential hazards was driven in part by the fact that these have been the major focus of Existential Risk Studies in the past and in part by the fact that this focus allowed for a clearer conceptualization of the probabilities involved in each estimate.

The database has subsequently been expanded to include a wider range of existential risk estimates, including estimates of conditional probability (such as the “chance that ‘a full-scale nuclear war in the next century’ would ‘be the end of human potential’”) which capture a wider range of considerations including elements of vulnerability and exposure.

Purpose of the tool

The tool was originally developed by SJ Beard, Thomas Rowe, and James Fox in order to produce an analysis and evaluation of methodologies and use of evidence in the field of Existential Risk Studies. In this context, each estimate was presented together with the methodology and evidence that had been used to produce it and was evaluated both in terms of the fitness of these for the intended purpose and also the quality of their implementation in each study.

In its expanded form the database lives on as a repository for collective judgement across the Existential Risk Studies community and to help individual researchers think about how they would produce their own estimates.

History and background

The original database was published as a supporting literature review alongside an academic paper on methodologies in Existential Risk Studies that appeared in the journal Futures in 2020 having been compiled by the three authors between 2017 and 2020 and with an initial presentation at CSER’s 2018 Cambridge Conference on Catastrophic Risk.

Shortly after the publication of this paper Michael Aird, then a researcher for Convergent Analysis, created a spreadsheet combining the studies reviewed for this paper with estimates from elsewhere, as well as a separate spreadsheet to help individuals come up with their own estimates, and posted about this on the Effective Altruism Forum. Anyone can add to this spreadsheet but Michael remains its reviewer and curator. Michael later produced a similar database of estimates relating to nuclear war, which is entirely independent of CSER.

Where to get started

The original literature review and analysis and evaluation of methodologies is available at https://www.sciencedirect.com/science/article/abs/pii/S0016328719303313. An open-access version of this paper will shortly be made available as part of our forthcoming book An Anthology of Global Risk.

Michael Aird’s post about his enlarged database, together with a link to the spreadsheet, can be found here: https://forum.effectivealtruism.org/posts/JQQAQrunyGGhzE23a/database-of-existential-risk-estimates.

CSER's 3 top tips for using the databases

  1. These tools have been developed to both help people learn about the available research, methodologies, and evidence base in Existential Risk Studies and to encourage new researchers to see how they can contribute to improving it. However, both the original paper authors and Michael Aird are clear that they believe that quantified estimates like this are only a small part of the field and that making them has both advantages and disadvantages. Researchers should, therefore, avoid the temptation to blindly repeat these estimates as facts or to make their own estimates without thinking critically about their intended purpose and potential negative consequences. Deeper discussion of how to do this is provided in the paper and post as well as subsequent academic correspondence between Seth Baum and the original paper authors, which can be read at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3671042 and https://www.sciencedirect.com/science/article/abs/pii/S0016328720300963.
  2. In particular, we would advise caution when using estimates like this in a highly public context; not because we think ordinary people cannot engage productively with them but because of how easily they can be sensationalized for extra attention. Even the most responsible coverage of such estimates can easily be hijacked by less scrupulous communicators and should thus be handled with a high degree of sensitivity and caution.
  3. Given the difficulty of differentiating between different aspects of risk, such as hazard, vulnerability, and exposure, and the fact that different estimates are produced by different methods, with different evidence, and for different purposes, it is usually best to be as precise as possible when making a particular estimate and not to combine estimates without carefully reviewing all the ways in which they might interact.