This will involve the construction of a new typology of risks, one that I think could replace the typology developed by Nick Bostrom and Milan Ćirković for their 2008 co-edited book, Global Catastrophic Risks.

* * *

According to Bostrom and Ćirković (“B&C” from here on), an existential risk is any adverse outcome that (a) affects Earth-originating intelligent life, and (b) is transgenerational in scope and terminal in intensity. Ignoring (a) for now (the demarcation part of the definition), the terminology of (b) comes from B&C’s risk typology, shown in Figure A. This typology is based on a particular conceptual analysis according to which a risk is the probability of an event multiplied by its consequences. In his original 2002 paper, Bostrom further analyzes a risk's consequences into scope (how many are affected) and intensity (how severe the effects are). In Figure A, existential risks occupy the top right box, and global catastrophic risks (a broader category of serious outcomes) occupy the four shaded boxes.

Figure A. B&C’s typology of risks.

I think Figure A is problematic for several reasons. First, consider the y-axis. Here B&C distinguish between personal, local, global, and transgenerational categories of scope. The problem is that the first three concern a risk’s effects in space while the fourth concerns its effects in space and time. This makes the y-axis both inconsistent (conceptually) and inaccurate (empirically), since it mixes different types of categories and, in doing so, implies that personal and local risks can’t have transgenerational effects, which is false. A survivable germline mutation, for example, will be passed down from one generation to the next indefinitely. But where would it fit into Figure A? B&C’s typology makes no room for personal-endurable-transgenerational risks.

Figure A also has trouble categorizing the phenomenon of aging. I suspect that this is what prompted Bostrom to expand his 2002 typology into the schema above (which nonetheless includes a question mark in the “ageing” box), since aging is a global-terminal risk yet it doesn't result in the end of our lineage.ⁱ Thus, the category of “transgenerational” was added to enable B&C to talk about the human population itself, rather than just individuals within the population. But adding this category the way they did makes the y-axis of Figure A incoherent.

The solution to this problem is simple: analyze the scope of a risk’s consequences into its distinct spatial and temporal components. The first corresponds to the question “In a given time slice, how many people are affected?,” whereas the second corresponds to the question “In a given time segment, how many people are affected?” This leads to the analysis of Figure B.

Figure B. A modified analysis of “risk.”

Now consider the x-axis. Here one finds multiple problems. To begin, the way B&C define some of their terms makes their definition of an existential risk tautological. For example, a “terminal” event is one that “[causes] death or permanently and drastically [reduces] quality of life.” The problem is that a “transgenerational” event is one that affects “not only the current world population but all generations that could come to exist in the future.” Thus, “transgenerational” essentially means “global and permanent.” Because of this, B&C's conception of an existential risk as a transgenerational-terminal event can be reworded as follows: an existential risk is any event that causes the permanent extinction of all intelligent life on Earth; but it could also be an adverse outcome that permanently affects all intelligent life on Earth with the permanent curtailing of its potential (on the added condition that this curtailing is also drastic). The second “permanent” is clearly unnecessary, since the concept of permanence is built-in to the concept of transgenerational.

This definition also entails, peculiarly, that some endurable catastrophes can count as being “terminal.” Consequently, although “terminal” and “endurable” appear to be alternatives in Figure A, they aren't in any strict sense: some terminal events are actually survivable. This counterintuitive result is entirely avoidable, though, given the analysis of Figure B. Thus, in the typology below, B&C's idiosyncratic definition will be replaced by the standard dictionary definition of “terminal,” meaning “situated at or forming the end or extremity of something” (as the OED puts it), for the sake of both conceptual and terminological clarity.

Note, incidentally, that this makes the category of “terminal” binary, i.e., something is either dead or alive / extinct or extant, while the category of “endurable” is spectral, i.e., survivability is a matter of degree, a fact that will become important below. Adopting the standard dictionary definition also solves the redundancy problem above, since it eliminates the second “permanent” in B&C's definition.

This leads to a final problem with the x-axis: it's entirely unclear why B&C include the category of “imperceptible” next to “endurable” and “terminal.” (Bostrom even includes this category in his most recent and even more extravagant 2013 typology.) Whereas “terminal” and “endurable” are genuine alternatives (if standard definitions are applied), “imperceptible” and “endurable” aren't. Rather, “imperceptible” is a subcategory of “endurable,” denoting something like “an endurable risk that's barely noticeable.” Placing it next to “endurable” is like calling a kitchen a “small house.”

Since endurability constitutes a spectrum, one can divide it up at will to include any number of subcategories, such as “imperceptible, “somewhat perceptible,” “quite perceptible,” and so on. None of these should be on the x-axis, though, since this would commit a category error. The typology below thus removes this term altogether, leaving only “endurable” and “terminal” as the categories of intensity.

* * *

My typology of risks is shown in Figure C. It is based on the analysis of Figure B, and it appears to solve all the problems mentioned above. Although I don’t attempt it here, the opacity of the dots (specifying either risk types or tokens) could be made a function of the probabilities of the risks being represented: the more opaque, the more probable; the more transparent, the less probable. As a result, Figure C could, unlike Figure A, exhaustively represent every aspect of any kind of risk, as understood in Figure B. I can imagine this being useful in discussions about how we should prioritize big picture issues and properly allocate our resources.

Figure C. A typology of risks that distinguishes between intensity, spatial scope, and temporal scope.

Looking at specific examples: a simulation shut down, an engineered pandemic, an all-out nuclear war, and a physics disaster are all possible red dots. A totalitarian regime with global control that severely oppresses its citizens and a dysgenic scenario in which human cognitive capacities significantly decline are both possible black dots.

The phenomenon of aging, then, corresponds to the orange dot. It is a global-terminal-generational risk, since it affects all individuals within the population but not the population as a whole (i.e., the population doesn't grow old and die). Notice here that once we get to the “global” level of spatial scope, the difference between “generational” and “transgenerational” is the difference between members of the lineage and the lineage itself. (This was alluded to above, as the reason that Bostrom expanded his 2002 typology into Figure A, which introduced the scope category of “transgenerational.”)

Furthermore, the disappearance of Maldives due to rising sea levels corresponds to the green dot; the volcanic winter caused by the supereruption at Toba (lasting 6 to 10 years) corresponds to the purple dot; 2012’s Superstorm Sandy corresponds to the light grey dot; a survivable germline mutation corresponds to the blue dot; and a survivable case of skin cancer corresponds to the yellow dot.

* * *

​Figure C requires a slight reconceptualization of global catastrophic risks. In my typology, any event with global scope – including the red, black, purple, and orange dots – counts as a global catastrophic risk. As for existential risks, B&C's definition doesn't map perfectly onto Figure C. According to their account, all red dots are existential risks, but only black dots that are especially bad – i.e., that result in the “drastic curtailing of our potential” – will count as such. Although I initially suggested in my aforementioned article that we reserve the term “existential risk” for red dot events only, I think there are reasons for including some black dot events as well.

First, notice that red dots are the only type of risk that results in 100% of humanity dying out. In his book Reasons and Persons, Derek Parfit famously argues that the difference between 100% and 99% of humanity dying out is far greater than the difference between world peace and 99% of humanity being destroyed. The reason is that the 100% scenario closes off the possibility of any future civilization developing into something of value, whereas the 99% scenario doesn't.

But this isn't necessarily true. As B&C understand “endurable existential risks,” such events cannot be recovered from. Thus, a repressive totalitarian regime with a global and permanent (i.e., “transgenerational”) control over the population could close off the possibility of future lives worth living just as effectively as total annihilation. It follows that the difference between peace and the most unthinkable black dot events may be equally as great as the difference between 99% and 100% of humanity kicking the bucket.

Another similarity between red and black dots is that, because recovery is by definition impossible, neither can be learned from the way we can learn from every other type of risk, in order to avoid them in the future. Once a red or black dot event has occurred, the game is over – and it's been lost. These similarities may justify giving both the title of “existential risks.” After all, the concept of an existential risk is supposed to latch onto scenarios in the world that constitute the worst of all possible outcomes, and it's hard to tell which is more undesirable: total annihilation or survivable disasters that seriously harm us and never get better over time.

Although I think there are reasons to keep B&C's general conception of an existential risk, Figure A is still tautological, inconsistent, and empirically inadequate. I have argued that Figure C solves these problems, and as such should be integrated into future conceptions of existential and global catastrophic risks.

* * *

(this is a work in progress; comments are welcome)