Longtermism is the thesis that in a large class of decision situations, the best thing we can do is what is best for the long-term future. The scope question for longtermism asks: how large is the class of decision situations for which this is true? In this paper, I suggest that the scope of longtermism may be narrower than many longtermists suppose. I identify a restricted version of longtermism: swamping axiological strong longtermism (swamping ASL). I identify three scope-limiting factors — probabilistic and decision-theoretic phenomena which, when present, tend to reduce the prospects for swamping ASL. I argue that these scope-limiting factors are often present in human decision problems, then use two case studies from recent discussions of longtermism to show how the scope-limiting factors lead to a restricted, if perhaps nonempty, scope for swamping ASL.
David Thorstad, “The scope of longtermism”
1. Introduction
This post begins a new series on my paper, “The scope of longtermism“. The best way to introduce the paper is to begin with Greaves and MacAskill’s original statement of axiological strong longtermism:
In a wide class of decision situations, the option that is ex ante best is contained in a fairly small subset of options whose ex ante effects on the very long-run future are best.
The scope question for longtermism asks: how wide is the class of situations for which this statement holds? My aim in this paper is to answer the scope question.
2. Background views
When I came to Oxford in 2020, it was widely assumed that if longtermism was true in any contemporary decision context, it was probably true in nearly every decision context. After all, many longtermists think that some actions, such as preventing existential risk, are many orders of magnitude better than anything else we can do today. But almost all of our actions have some impact on our ability to cause or mitigate existential risks, and even a small chance of astronomical benefit is, from an ex ante perspective, extremely weighty. Thus it seems that in almost all decision contexts, ex ante value is dominated by long-run considerations.
In this vein, many leading longtermists suggested that the scope of longtermism is very wide. For example, Hilary Greaves and Will MacAskill originally suggested that longtermism holds in the decision problem where a philanthropist Shivani is deciding between short- and longtermist interventions, then concluded from this that:
Insofar as axiological strong longtermism is true of Shivani’s decision context, it seems likely to be true also of a fairly wide variety of other decision contexts.
Nick Beckstead argued that the short-term effects of paradigmatic global health interventions such as preventing childhood blindness may be swamped by their long-term effects.
Suppose I cure some child’s blindness. We ordinarily think that the main benefit of this is that the child will have a better life … However, curing the child’s blindness creates a ripple effect that carries forward through many people affected by the child, and many people those people affect, and so forth. If these ripple effects continue for a long time, it can be argued that they swamp the proximate benefits that command our attention.
Beckstead went so far as to conclude that because saving is beneficial for the long-term future, and individuals in wealthy countries save more than individuals in poorer countries, “It now seems more plausible to me that saving a life in a rich country is substantially more important than saving a life in a poor country, other things being equal”.
Owen Cotton-Barratt went further, arguing that most everyday decisions, including selecting topics for dinner-table conversation, should be made in pursuit of proxy goals that plausibly correlate with long-term value.
On the flipside, leading opponents of longtermism have often been unwilling to explicitly concede that longtermism correctly describes any decision problem, past, present or future.
My aim in this paper is to chart a path between these two extremes. I defend a view on which the scope of longtermism, while nonempty, is probably much narrower than many longtermists assumed.
3. Unpacking the scope question
To unpack the scope question, we need to get clear on what is being claimed.
3.1 Axiological and deontic longtermism
Longtermism can be construed as an axiological claim about what actions are best, or as a deontic claim about what actions we ought to take. In this series, I am concerned with axiological longtermism.
For some, such as consequentialists, there is a tight link between axiological and deontic longtermism. For others, the link is less tight, and further arguments (such as Greaves and MacAskill’s stakes-sensitivity argument) will need to be made. But that is a project for another day.
3.2 Ex ante and ex post longtermism
Axiological longtermism can be construed as a claim about the ex post effects that actions will actually have, or instead the ex ante effects that actions are expected to have. As in most discussions of longtermism, I focus on ex ante longtermism.
One reason for this is that ex post longtermism is likely to have far wider scope than ex ante longtermism. It isn’t crazy to think that the value which some actions will in fact produce is going to be dominated by the long-term consequences of those actions. But it does not follow that the expected value of those actions will be dominated by their long-term impacts, since there is substantial uncertainty about long-term impacts.
We have already seen a classic statement of ex ante axiological longtermism, namely Greaves and MacAskill’s original statement of axiological strong longtermism (ASL):
(Axiological Strong Longtermism) In a wide class of decision situations, the option that is ex ante best is contained in a fairly small subset of options whose ex ante effects on the very long-run future are best.
I focus on this version of longtermism here.
In recent years, longtermists have proposed a variety of weaker definitions of longtermism. These definitions will not be my focus in this series. In large part, they are not my focus because I think that many longtermists subscribe to stronger versions of longtermism and that many longtermist arguments, if correct, would ground stronger versions of longtermism. I don’t want to argue against weaker claims when most parties to the debate are actually interested in something more like axiological strong longtermism.
3.3 Swamping
Let a longtermist option be an option whose ex ante effects on the very long-run future are near best. The scope of ASL extends to any case in which the ex ante best option is a longtermist option. This can happen in two very different ways.
First, the ex ante best option can be a swamping option: an option whose expected long-term benefits exceed in magnitude the expected short-term effects produced by any option. This is the scenario envisioned by many longtermists, so it will be my target here.
(Swamping Axiological Strong Longtermism (Swamping ASL)) In a wide class of decision situations, the option that is ex ante best is (a) a swamping option and (b) contained in a fairly small subset of options whose ex ante effects on the very long-run future are best.
Second, the best option might not be a swamping option, but may nevertheless be contained in a fairly small subset of options whose ex ante effects on the very long-run future are best. One way for this to be true would be on the convergence thesis that what is best for the short-term is often near-best for the long-term as well. Not everyone should accept the convergence thesis. For example, those impressed by patient philanthropy might think that promoting long-term value requires saving resources at a high rate, whereas promoting short-term value requires spending down resources quickly. But those who accept the convergence thesis are invited to put aside cases of convergence to focus, for the moment, on the scope of swamping ASL.
3.4 Scope-limiting factors
I claim that the scope of swamping ASL is narrower than many longtermists suppose. To support this claim, I identify three scope-limiting factors: probabilistic and decision-theoretic phenomena which, when present in a decision problem, tend to reduce the prospects for swamping ASL to correctly describe that problem.
I argue that the scope-limiting factors are present in many contemporary decision problems and that, when present, they tend to cause headaches for longtermism.
4. The good case: The Space Guard survey
4.1 Introducing the Space Guard Survey
To say that the scope of swamping ASL is narrow is not to say that it is empty. Let’s begin by considering a case that may be within the scope of swamping ASL.
Approximately 66 million years ago, the Cretaceous–Paleogene mass extinction completely transformed the face of the Earth. All land-dwelling mammals with a mass of at least five kilograms, including most of your favorite dinosaurs, became extinct. The event was so catastrophic that it marked the end of the Cretaceous period and the beginning of a new Paleogene era.
In 1980, a team of scientists proposed that the Cretaceous-Paleogene mass extinction was caused by an asteroid impact on Earth. Evidence for the asteroid-impact theory mounted, and by the early 1990s scientists were increasingly confident that they had found the crater of the responsible asteroid.
This news sparked understandable fear. If an asteroid caused the extinction of so many species during the Cretaceous period, might an asteroid impact do the same to us as well? Even a notoriously stingy United States Congress saw reason to act on this fear, spending about $70 million on the Space Guard Survey, a series of projects aimed at tracking potentially dangerous asteroids.
The asteroid that likely killed the dinosaurs was a monster: likely 10-15 kilometers in diameter. For reference, NASA classifies asteroids at least 1 kilometer in diameter as catastrophic, capable of causing global catastrophe, and expresses serious concern about asteroids approaching 10 kilometers in diameter. For this reason, the Space Guard Survey was especially concerned with tracking large near-Earth objects.
The Space Guard Survey was remarkably successful at this goal. By now, the project has mapped at least 95% of near-Earth objects with a diameter exceeding 1 kilometer, and they are fairly confident that they have mapped all near-Earth objects with a diameter exceeding ten kilometers.
In the estimation of the US Congress, the Space Guard Survey was a swamping longtermist option. Given the risk of extinction and the need to prepare, the long-term benefits of the Space Guard survey were thought to dramatically swamp the short-term benefits of other competing programs. Was Congress right?
4.2 Evaluating the Space Guard survey
Following Toby Newberry, Hilary Greaves and Will MacAskill estimate the value of the Space Guard survey as follows.
First, take on board NASA’s estimate that an asteroid of at least 1 kilometer in diameter strikes Earth about once in every 6,000 centuries.
Second, make the conservative estimate that an impact of this size would pose a one-in-a-million chance of causing human extinction, and that early detection of an extinction-causing asteroid would give a one-in-a-thousand chance of survival.
Finally, note that many published estimates of the expected number of future humanlike lives range from 1013 to 1055.
Multiplying through yields an estimate of the expected number of lives saved by the Space Guard survey: (1/6,000)*(1/1,000,000)*(1/1,000)*[1013 to 1055]. Dividing the $70 million cost of the Space Guard survey by the estimated lives saved puts cost-effectiveness in the neighborhood of $7,000 per life at the lower bound, and $7*10-39 at the upper bound. This makes the Space Guard Survey look like a plausible candidate for a swamping option.
Now we saw in the series Mistakes in the moral mathematics of existential risk that there are good reasons to doubt this style of calculation. When the mistakes in this calculation are corrected, the cost-effectiveness of the Space Guard survey is likely to decrease substantially.
At the same time, my aim in this series is not to suggest that the Space Guard survey was not worth funding. I might well have voted to fund the survey myself. Instead, I want to emphasize a different point: the Space Guard survey is a very good candidate to fall within the scope of swamping ASL. Indeed, we will see that the Space Guard survey plausibly avoids all three of the scope-limiting factors developed in this series.
However, this suggests that interventions such as the Space Guard survey may be very special. In many other decision contexts, the scope-limiting factors get a strong take and when they do, the context is much less likely to fall within the scope of swamping ASL.
5. Wrapping up
Today’s post introduced the scope question: how wide is the class of situations for which the ex ante best option is a swamping longtermist option? We then clarified the target of the paper as swamping axiological strong longtermism.
Although I think that the scope of swamping ASL is more limited than many longtermists take it to be, I do not think that swamping ASL has empty scope. For this reason, I introduced the Space Guard Survey and argued that the decision of whether to fund the Space Guard Survey may have been within the scope of swamping ASL.
The main project of this paper and series is to introduce three scope-limiting factors: probabilistic and decision-theoretic phenomena which, when present in a decision problem, tend to reduce the prospects for swamping ASL to correctly describe that problem. The next several posts will introduce the scope-limiting factors and argue that they often obtain in contemporary decision problems.
6. Further resources
If you’d like to engage with this paper beyond the blog series, you can:
- Read the paper: Here
- Watch the talk: Talk here, Handout here
- Read a summary: Here
- Listen to a podcast: Here (Focus: 2:08:00 – 2:15:00)
I hope to have more resources available as time goes on.
Leave a Reply