Worrying facts from 'The Precipice'

Published 24 March 2020

Back to writing

I recently finished reading Toby Ord’s new book ‘The Precipice: Existential Risk and the Future of Humanity’. The book discusses the risks of catastrophic events that destroy all or nearly all of humanity’s potential. There are many of them, including but not limited to the Hollywood scenarios that enter most people’s minds: asteroids, supervolcanoes, pandemics (natural and human-engineered), dystopian political ‘lock-in’, runaway climate scenarios, and unaligned artificial general intelligence. The overall risk of an existential catastrophe this century? Roughly one in six, this author guesses: Russian roulette. Clearly, mitigating existential risk is not nearly treated like the overwhelmingly important global priority it is: not in our political institutions, nor in popular consciousness. Anyway, it’s excellent– highly recommended.

It was also full of some fairly alarming and/or surprising facts. So in place of a full review, here are some highlights:

Not so much a fact, but an interesting thought: consider a toy model of existential risk where each century is exposed to an equal amount of risk rr (a constant hazard rate), and each century prior to the catastrophe has some constant value vv. Then the expected value of the future would be: EV=i=0(1r)iv=vrEV=\sum_{i=0}^{\infty}(1-r)^i v=\frac{v}{r}. So the expected disvalue of this century’s existential risk is rvr=vr\frac{v}{r}=v. Notice, perhaps surprisingly, how it follows that the value of reducing this century’s existential risk by some proportion is independent of the initial risk, of the hazard rate. This does also make sense intuitively: the higher the risk each century (the hazard rate), the shorter the expected length of our future; so the stakes are lower while lowering this century’s absolute risk is ‘easier’. On the other hand, the lower the hazard rate, the longer and larger our expected future; so the stakes are higher while the risk this century ‘harder’ to reduce. The value of halving all future risk is vr/2vr=vr\frac{v}{r/2}-\frac{v}{r}=\frac{v}{r}. More generally, the value of reducing all future risk by some proportion xx is equal to xvr(1x)\frac{xv}{r(1-x)}. Notice that the value of reducing risk over all centuries by some fixed proportion is higher the lower the risk per century (hazard rate). Further:

Suppose you thought it equally likely the value of reducing risk was ten times as important as the basic model suggests, or a tenth as important. The average of these is not one times as important: it is 5.05 times as important.

Again, strongly recommended; but not at all comforting.

The Precipice and the Land Beyond - Hilary Paynter

The Precipice and the Land Beyond - Hilary Paynter

More resources:



Back to writing