Where this fits

How this tool relates to the report, Rethink Priorities’ later models, The Unjournal’s other evaluations, and other work on the value of reducing extinction risk

This site is an executable version of one family of arguments: what a small, temporary cut in extinction risk is worth, given beliefs about how risk evolves and how much future years are worth. It isn’t a donation optimizer or a forecast. Other tools do those jobs, and several of them build on the same model. This page maps the neighbours, what each adds, and how this tool differs.

Links were checked on 13 October 2026. Descriptions are our summaries, not the authors’.

The lineage in brief

2020. Toby Ord’s The Precipice popularizes the “risk is unusually high right now” framing and gives per-century risk estimates.

2022-23. David Thorstad shows that in the simple constant-risk model, higher background risk lowers the value of mitigation, and that only a sharp, lasting drop in risk (a “time of perils”) restores very large values.

October 2023. Arvo Muñoz Morán’s report for Rethink Priorities (RP), part of its CURVE sequence, extends that model with four risk paths, five value-growth paths and variable persistence. This tool re-implements it.

November 2023-July 2024. RP builds tools that use these ideas as inputs: the Cross-Cause Cost-Effectiveness Model, then the Portfolio Builder and Moral Parliament tools.

2026. RP launches its Cross-Cause Fund and Donor Compass. The write-up for its global catastrophic risk funds lists the report as an influence (see Departures). The Unjournal publishes its evaluation of the report, and this tool adds the evaluation’s alternatives as switches.

The report and its companions

• The report. Arvo Muñoz Morán, “How bad would human extinction be?” (RP Worldview Investigations, October 2023): RP page, EA Forum post, and the technical report PDF this site cites by page. Companion notebook: github.com/arvomm/verm.
• The CURVE sequence. Causes and Uncertainty: Rethinking Value in Expectation (Bob Fischer and others, RP, October 2023), ten posts on cross-cause prioritization. Two are close siblings of the report:
  • David Bernard, “Charting the precipice: The time of perils and prioritizing x-risk”. It breaks the time-of-perils case into a chain of premises and argues that their conjunction deserves fairly low credence. This is the qualitative counterpart to the “Risk scenario” menu here.
  • Laura Duffy, “Is x-risk the most cost-effective if we count only the next few generations?”, with open code. It asks the opposite question to the report: what x-risk work is worth on short horizons, with risk aversion.

Rethink Priorities’ later tools

These take the report’s kind of reasoning and put it to work in allocation decisions. They aggregate across causes, so the mechanics of any one model are harder to see. This tool does the reverse: one model, every mechanism exposed.

Tool	What it does	How it relates here
Cross-Cause Cost-Effectiveness Model (2023, live)	Monte Carlo cost-effectiveness for projects across causes, including x-risk projects (biorisk, AI, nuclear and others)	Its x-risk module has the same structure as the report (proportional cut in yearly risk for a limited time); its overview thanks the report’s author
Portfolio Builder (2024, live; post)	Splits a budget across causes, with cost curves and a chance that projects are futile or backfire	Adds the decision-theory layer: backfire odds and risk aversion shift money away from x-risk work. Compare switch 2 here
Moral Parliament (2024, live; post)	Allocates across causes when you’re unsure which moral view is right	Handles normative uncertainty; this tool handles empirical assumptions
Cross-Cause Fund and Donor Compass (2026, live; announcement, methodology, data viewer)	Recommends a split across funds, including three global catastrophic risk funds, from worldview questions	Its GCR fund model lists this report as an influence. Our “RP 2026” preset approximates its central values, and switch 4 applies its time-window weights. Details on the Departures page

RP’s write-up says it plans “extensive internal and external review” of the GCR fund model.

The Unjournal’s evaluations of related work

The Unjournal commissions public expert evaluations of research that matters for decisions. Besides our evaluation of this report (the evaluation itself), these are the closest:

• Karger and others (Forecasting Research Institute), “Forecasting Existential Risk: Evidence from a Long-Run Forecasting Tournament”. Evaluation summary; paper, since published in the International Journal of Forecasting. The best public source of expert and superforecaster risk estimates to plug into the risk settings here.
• Trammell and Aschenbrenner, “Existential Risk and Growth”. Evaluation summary; paper (May 2024 version evaluated). Makes risk depend on choices: richer societies spend more on safety, which can produce a rise-then-fall path like the time of perils. The report takes that path as given; switch 3 here is a crude version of risk that responds to actions.
• Clancy, “The Returns to Science in the Presence of Technological Risks”. Evaluation summary; paper. Weighs the benefits of science against forecast catastrophe risk; its answer turns on the weight given to the long run and on which forecasts you trust.
• Denkenberger, Sandberg, Tieman and Pearce, “Long term cost-effectiveness of resilient foods for global catastrophes compared to artificial general intelligence safety”. Evaluation summary, authors’ response; paper. A Monte Carlo comparison of two kinds of catastrophe work, the “which x-risk work” question this tool abstracts away.

Academic work the model draws on or speaks to

Work	Main point	Where it shows up here
Toby Ord, The Precipice (2020), book site	Risk is unusually high this century; per-century estimates	The report’s 20% per-century high-risk default; Ord’s simple model is the starting point
David Thorstad, “Existential risk pessimism and the time of perils” (GPI working paper, 2022, PDF); published as “High Risk, Low Reward”, Philosophy & Public Affairs (2023), open access	Higher background risk lowers the value of mitigation unless risk later drops sharply; the arguments for such a drop are weak	The risk scenarios, the low-risk setting, and switch 1 (a floor on risk)
David Thorstad, “Mistakes in the Moral Mathematics of Existential Risk”, Ethics (2024), DOI, working paper	Common errors (cumulative versus per-period risk, ignoring background risk, ignoring population dynamics) inflate estimates	Why the page converts per-century risk to per-year and shows what the horizon leaves out
Christian Tarsney, “The epistemic challenge to longtermism”, Synthese (2023), DOI, working paper	Random future events can wipe out an intervention’s effect; the case for long-run work then rests on small chances of huge payoffs	Persistence \(P\); RP’s GCR model also cites it as a basis
Charles I. Jones, “The AI Dilemma: Growth versus Existential Risk”, AER: Insights (2024), DOI, free copy	How much extinction risk is worth accepting for growth depends on how fast extra consumption loses value	The value-growth menu: how much a richer future year is worth
Millett and Snyder-Beattie, “Existential Risk and Cost-Effective Biosecurity”, Health Security (2017), open access	An early cost-per-life-year estimate for x-risk spending, very sensitive to horizon and discounting	The horizon diagnostics and switch 4
Alexandrie and Eden, “Is Existential Risk Mitigation Uniquely Cost-Effective? Not in Standard Population Models” (GPI working paper 2023, PDF)	Some non-x-risk interventions also have permanent effects, so x-risk work isn’t uniquely persistent	A limit of this model: only extinction changes the long run
Brauner and Grosse-Holz, “The expected value of extinction risk reduction is positive” (2018), and DiGiovanni’s critique (2021)	Whether the future being protected has positive value at all	The model assumes it does; the value paths are all positive
Hayden Wilkinson, “The unexpected value of the future” (GPI working paper 2022, PDF)	Expected value can fail to exist when outcomes are heavy-tailed	The uncapped exponential option, where the page says no finite bound exists

The report also builds on an unpublished manuscript by Tom Adamczewski on the expected value of the long-term future, which we couldn’t find online. The Global Priorities Institute closed in 2025; its working-paper PDFs are still online but its paper pages aren’t, so we link the PDFs.

Other public models

• Gregory Lewis, “X-risk reduction for the person affecting view” (Guesstimate, 2018; post). Counts only people alive now: the opposite end from this model’s long horizons.
• Laura Duffy’s public risk model (RP, 2023). Open code for short-horizon x-risk cost-effectiveness with risk aversion.
• Arepo, “Two tools for rethinking existential risk” (2024), with model code. A model of civilizational states with setbacks and recovery, where the report has a single, permanent extinction state.
• MIT AI Risk Repository, mitigations map. A taxonomy of concrete AI risk mitigations, without effect sizes. It’s the complement to this tool, which values an abstract cut in risk \(f\) for \(P\) years without saying what achieves it.

Other Unjournal interactive models

• Cultured chicken production cost model: the same approach (an explicit model, every assumption adjustable, sources tagged) applied to the cost of cultivated meat.

What this tool is for, among all these

Its advantage is that it can be inspected. Every assumption can be changed. The original report is reproduced, not just summarized (Replication). Horizon effects are shown, not hidden. Each extension can be switched on alone. Each setting in use is tagged by where its value comes from. It doesn’t try to be the most complete x-risk model, and it shouldn’t be read as an estimate of any real intervention’s value.

Know of something we’ve missed, or think we’ve described a piece of work unfairly? Tell us.