I recently read Vaden Masrani’s post “A case against strong longtermism” for a book/journal club, and noted some reactions to the post as I went. I’m making this post to share slightly-neatened-up versions of those reactions.[1] I’ll split my specific reactions into separate comments, partly so it’s easier for people to reply to specific points.
Masrani’s post centres on critiquing The Case for Strong Longtermism, a paper by Greaves & MacAskill. I recommend reading that paper before reading this post or Masrani’s post. I think the paper is basically very good and very useful, though also flawed in a few ways; I wrote my thoughts on the paper here.
My overall thoughts on Masrani’s post are as follows:
- I think that criticism is very often valuable, and especially so for ideas that are promoted by prominent people and are influencing important decisions. Masrani’s post represents a critique of such an idea, so it’s in a category of things I generally appreciate and think we should generally be happy people are producing.
- However, my independent impression is that the critique was quite weak and that it involved multiple misunderstandings of the Greaves & MacAskill paper in particular, longtermist ideas and efforts more generally, and also some other philosophical ideas.
- Relatedly, my independent impression is that Masrani’s post is probably more likely to cause confusions or misconceptions than it is to usefully advance people’s thinking and discussions.
- All that said, I do think that there are various plausible arguments against longtermism that warrant further discussion and research.
- Some are discussed in Greaves and MacAskill’s paper.
- One of the best such arguments (in my view) is discussed in Tarsney’s great paper “The epistemic challenge to longtermism”.
- See also Criticism of effective altruist causes and What are the leading critiques of "longtermism" and related concepts.
(Given these views, I was also pretty tempted to call this A Case Against “A Case Against Longtermism”, but I didn’t want to set off an infinitely recursive loop of increasingly long and snarky titles!)
(Masrani also engaged in the comments section of their original post, wrote some followup posts, and has discussed similar topics on a podcast they host with Ben Chugg. I read most of the comments section on the original post and listened to a 3 hour interview they had with Fin and Luca of the podcast Hear This Idea, and continued to be unimpressed by the critiques provided. But I haven’t read/listened to the other things.)
[1] This seemed better than just making all these comments on Masrani’s post, since I had a lot of comments and that post is from several months ago.
This post does not necessarily represent the views of any of my employers.
Masrani writes:
This is true, but seems to be responding to tone rather than the substance of the argument. And given that (I think) we're interested in the substantive question rather than the social legitimacy of the criticism, I think that it is more useful to engage with the strongest version of the argument.
The actual issue that is relevant here, which isn't well identified, is that naive expected value fails in a number of ways. Some of these are legitimate criticisms, albeit not well formulated in the paper. Specifically I think that there are valuable points about secondary uncertainty, value of information, and similar issues that are ignored by Greaves and MacAskill in their sketch of the ideal decision-theoretic reasoning.
That's roughly true for me saying "In any case, “utterly oblivious” seems to me to be both a rude phrasing and a strong claim."
But I don't think it's true for my comment as a whole. Masrani makes specific claims here, and the claims are inaccurate.
I think steelmanning is often really useful. But I think there's also valuing in noticing when a person/post/whatever is just actually incorrect about something, and in trying to understand what arguments they're actually making. Some reasons:
So here I'm actually not solely interested in the substantive questions raised by Masrani's post, but also in countering misconceptions that I think the post may have generated, and giving indications of why I think people might find it more useful to engage with other criticisms of longtermism instead (e.g., the ones linked to in the body of my post itself).
One final thing worth noting is that this was a quickly produced post adapting notes I'd made anyway. I do think that if I'd spent quite a while on this, it'd be fair to say "Why didn't you just talk about the best arguments against longtermism, and the points missing from Greaves & MacAskill, instead?"
Yeah, I imagine there are many things in this vicinity that Greaves & MacAskill didn't cover yet that are relevant to the case for strong longtermism or how to implement it in practice, and I'd be happy to see (a) recommendations of sources where those things are discussed well, and/or (b) other people generate new useful discussions of those things. Ideally applied to longtermism specifically, but general discussions - or general discussions plus a quick explanation of the relevance - seems useful too.
I definitely don't mean to imply with this post that I see strong longtermism as clearly true; I'm just quickly countering a specific set of misconceptions and objections.
As I mentioned in my other reply, I don't see as much value in responding to weak-man claims here on the forum, but agree that they can be useful more generally.
Regarding "secondary uncertainty, value of information, and similar issues," I'd be happy to point to sources that are relevant on these topics generally, especially Morgan and Henrion's "Uncertainty," which is a general introduction to some of these ideas, and my RAND dissertation chairs work on policy making under uncertainty, focused on US DOD decisions, but applicable more widely. Unfortunately, I haven't put together my ideas on this, and don't know that anyone at GPI has done so either - but I do know that they have engaged with several people at RAND who do this type of work, so it's on their agenda.
So you've shown that Masrani has made a bunch of faulty arguments. But do you think his argument fails overall? i.e. can you refute its central point?
tl;dr: Yes, I think so, for both questions. I think my comments already did this, but that I didn't make it obvious whether and where this happened, so your question is a useful one.
I like that essay, and also this related Slate Star Codex essay. I also think this might be a generically useful question to ask in response to a post like the one I've made. (Though I also think there's value in epistemic spot checks, and that if you know there are a large number of faulty arguments in X but not whether those were the central arguments in X, that's still some evidence that the central arguments are faulty too.)
Your comment makes me realise that probably a better structure for this post would've been to first summarise my understanding of the central point Masrani was making and Masrani's key arguments for that, and then say why I disagree with parts of those key arguments, and then maybe also add other disagreements but flag that they're less central.
The main reason my post is structured as it is is basically just that I tried to relatively quickly adapt notes that I made while reading the post. But here's a quick attempt at something like that (from re-skimming Masrani's post now, having originally read it over a month ago)...
---
Masrani writes:
As noted in some of my comments:
---
(I should note again that I read the post over a month ago and just dipped in quickly to skim for a central point to refute, so it's possible there were other central points I missed.)
I also expect that the post mentioned various other things that are related to better arguments against longtermism, e.g. the epistemic challenge to longtermism that Tarsney's paper discusses. But I'm pretty sure I remember the post not adding to what had already been discussed on those points. (A post that just summarised those other arguments could be useful, but the post didn't set out to be that.)
Hey! Can't respond most of your points now unfortunately, but just a few quick things :)
(I'm working on a followup piece at the moment and will try to respond to some of your criticisms there)
My central point is the 'inconsequential in the grand scheme of things' one you highlight here. This is why I end the essay with this quote:
> If among our aims and ends there is anything conceived in terms of human happiness and misery, then we are bound to judge our actions in terms not only of possible contributions to the happiness of man in a distant future, but also of their more immediate effects. We must not argue that the misery of one generation may be considered as a mere means to the end of securing the lasting happiness of some later generation or generations; and this argument is improved neither by a high degree of promised happiness nor by a large number of generations profiting by it. All generations are transient. All have an equal right to be considered, but our immediate duties are undoubtedly to the present generation and to the next. Besides, we should never attempt to balance anybody’s misery against somebody else’s happiness.
Just wanted to flag that I responded to the 'proving too much' concern here: Proving Too Much
Hey Vaden!
Yeah, I didn't read your other posts (including Proving Too Much), so it's possible they counter some of my points, clarify your argument more, or the like.
(The reason I didn't read them is that I read your first post, read most comments on it, listened to the 3 hour podcast, and have read a bunch of other stuff on related topics (e.g., Greaves & MacAskill's paper), so it seems relatively unlikely that reading your other posts would change my mind.)
---
Hmm, something that strikes me about that quote is that it seems to really be about deontology vs consequentialism - and/or maybe placing less moral weight on future generations. It doesn't seem to be about reasons why strong longtermism would have bad consequences or reasons why longtermist arguments have been unsound (given consequentialism). Specifically, that quote's arguments for its conclusion seem to just be that we have a stronger "duty" to the present, and that "we should never attempt to balance anybody’s misery against somebody else’s happiness."
(Of course, I'm not reading the quote in the full context of its source. Maybe those statements were meant more like heuristics about what types of reasoning tend to have better consequences?)
But if I recall correctly, your post mostly focused on arguments that stronglongtermist would have bad consequences or that longtermist arguments have been unsound. And "we should never attempt to balance anybody's misery against somebody else's happiness" is either:
So I think that that quote arrives at a similar conclusion to you, but it might show very different reasoning for that conclusion than your reasoning?
Do you have a sense of what the double crux(es) is/are between you and most longtermists?
Masrani seems to take (some of) Greaves and MacAskill’s examples and tentative views about what strong longtermism might indicate one should prioritise as a logically necessary consequence of the moral view itself. In particular, Masrani seems to assume that longtermism necessarily focuses solely on existential risk reduction. But this is actually incorrect.
I felt uncomfortable with and confused by the section of the post that was about jargon and euphemisms.
Also worth noting that there are a bunch of other more accessible descriptions of longtermism out there and this is specifically a formal definition aimed at an academic audience (by virtue of being a GPI paper)
Masrani seemed to jump to a strange, uncharitable, and incorrect conclusion about the history of longtermist thought.
Masrani wrote:
Masrani says that "longtermism encourages us to treat our fellow brothers and sisters with careless disregard for the next one thousand years, forever". But strong longtermism being true now doesn't mean it always was true and always will be true, as Greaves and Macaskill themselves note.
Masrani writes:
At least in some places, Masrani seems to think or imply that longtermism doesn’t aim to influence any events that occur in the next (say) 1000 years. But in reality, longtermists mostly focus on influencing the further future via influencing things that happen within the next 1000 years (e.g., whether an existential catastrophe occurs).
This seems to agree with his criticism - that we care about the near-term only as it affects the long term, and can therefore justify ignoring even negative short term consequences of our actions if it leads to future benefits. It argues even more strongly for abandoning otherwise short term beneficial interventions with small longer term impacts.
Obvious examples of how this goes wrong include many economic planning projects of the 20th century, where the short term damage to communities, cities, and livelihoods was justified by incorrect claims about long term growth.
tl;dr: I basically agree with everything except "This seems to agree with his criticism", because I think (from memory) that Masrani was making a stronger and less valid claim. (Though I'm not totally sure; it may have just been slightly sloppy writing + the other misconception that longtermism is necessarily solely focused on existential risk reduction.)
---
I think there's a valid claim similar to what Masrani said, and that that could reasonably be seen as a criticism of longtermism given some reasonable moral and/or empirical assumptions. Specifically, I think it's true that:
I would mostly "bite the bullet" of this critique - i.e., say that we can't prioritise everything at once, and if the case for strong longtermism holds up then it's appropriate that we prioritise the long-term at the expense of the short-term. And then I do think we should remain vigilant of ways our thinking, priorities, actions, etc. could mirror bad instances of "ends justify the means" etc.
But I could understand someone else being more worried about this objection.
Also, FWIW, I think the Greaves and MacAskill paper maybe fails to acknowledge that strong longtermism actions might be very strange or net-negative from a near-term perspective, rather than just not top priorities. (Though maybe I just forgot where they said this.) I made a related comment here.
---
We could steelman Masrani into making the above sorts of claims and then have a productive discussion. But I think it's also useful to sometimes just talk about what someone actually said and correct things that are actually misleading or common misconceptions. And I think Masrani was making a stronger claim (though I'm now unsure, as mentioned at the top), which I also think some other people actually believe and which seems like a misconception worth correcting (see also). (To be fair, I think Greaves & MacAskill could maybe have been more careful with some phrasngs to avoid people forming this misconception.)
E.g. Masrani writes:
And:
And:
(But again, I now realise that this might have just been slightly sloppy writing + the x-risk misconception, and also that Greaves & MacAskill may have been slightly sloppy with some phrases as well in a way that contributed to this. So I think this point isn't especially important as a critique of the post.
Though I guess my original statement still seems appropriate hedged: "At least in some places, Masrani seems to think or imply that longtermism doesn’t aim to influence any events that occur in the next (say) 1000 years." [emphasis added])
I think we basically agree.
And while I agree that it's sometimes useful to respond to what was actually said, rather than the best possible claims, that type of post is useful as a public response, rather than useful for discussion of the ideas. Given that the forum is for discussion about EA and EA ideas, I'd prefer to use steelman arguments where possible to better understand the questions at hand.
(I'll put a bundle of smaller, disconnected reactions in this one thread.)
Masrani writes:
Masrani seems to confuse (1) pure time discounting / a pure rate of time preference with (2) time discounting for other reasons (e.g., due to the possibility that the future won’t come to pass due to a catastrophe; see Greaves).
---
Another, separate point about discounting:
Masrani writes:
But as far as I can tell, this is false, at least taken if literally; instead, how concerned one should be about a given moment in time depends in part on what’s happening at the time (e.g. how many moral patients there are, and what they’re experiencing).
Masrani writes:
Yes, this seems to be a problem, but it's also a problem with naive expected value thinking that prioritizes predictions without looking at adaptive planning or value of information. And I think Greaves and MacAskill don't really address these issues sufficiently in their paper - though I agree that they have considered them and are open to further refinement of their ideas.
But I don't beleive that it's clear we predict things about the long term "with above chance accuracy." If we do, it's not obvious how to construct the baseline probability we would expect to outperform.
Critically, the requirement for this criticism to be correct is that our predictions are not good enough to point to interventions that have higher expected benefit than more-certain ones, and this seems very plausible. Constructing the case for whether or not it is true seems valuable, but mostly unexplored.
Yeah, I agree with your first two paragraphs. (I don't think I understand the third one; feel free to restate that, if you've got time.)
In particular, it's worth noting that I agree that it's not currently clear that we can predict (decision-relevant) things about the long-term with above chance accuracy (see also the long-range forecasting tag). Above, I merely claimed that "we very often predict things that depend on things we don’t fully understand, and with above chance accuracy" - i.e., I didn't specify long-term.
It does seem very likely to me that it's possible to predict decision-relevant things about the long-term future at least slightly better than complete guesswork. But it seems plausible to me that our predictive power becomes weak enough that that outweighs the increased scale of the future, such that we should focus on near-term effects instead. (I have in mind basically Tarsney's way of framing the topic from his "Epistemic Challenge" paper. There are also of course factors other than those two things that could change the balance, like population ethical views or various forms of risk-aversion.)
This seems like a super interesting and important topic, both for getting more clarity on whether we should adopt strong longtermism and on how to act given longtermism.
---
I specified "decision-relevant" above because of basically the following points Tarsney makes in his Epistemic Challenge paper:
Agree that this is important, and it's something I've been thinking about for a while. But the last paragraph was just trying to explain what the paper said (more clearly) were evaluative practical predictions. I just think about that in more decision-theoretic terms, and if I was writing about this more, would want to formulate it that way.
Masrani focuses quite a bit on the idea that longtermism relies on comparisons to an infinite amount of potential future good. But Greaves and MacAskill's paper doesn't actually mention infinity at any point, and neither their argument nor the othe standard arguments I've seen rely at all on infinities.
Masrani writes:
Masrani seems to sort-of implicitly assume that (a) people will have strong ulterior motives to bend the ideas of strong longtermism towards things that they want to believe or support anyway (for non-altruistic reasons), and thus (b) we must guard against a view or a style of reasoning which is vulnerable to being bent in that way. But I think it would be more productive and accurate to basically “assume good faith”.
This ties into another point: Many of the framings and phrasings in Masrani’s post seem quite “loaded”, in the sense of making something sound bad partly just through strong connotations or rhetoric rather than explicit arguments in neutral terms.
This can happen unconsciously, though, e.g. confirmation bias, or whenever there's arbitrariness or "whim", e.g. priors or how you weight different considerations with little evidence. The weaker the evidence, the more prone to bias, and there's self-selection so that the people most interested in longtermism are the ones whose arbitrary priors and weights support it most, rightly or wrongly. (EDIT: see the optimizer's curse.) This is basically something Greaves and MacAskill acknowledge in their paper, although also argue applies to short-term-focused interventions:
That being said, I suspect it's possible in practice to hedge against these indirect effects from short-term-focused interventions.
I haven't read your post, so can't comment.
That said, FWIW, my independent impression is that "cluelessness" isn't a useful concept and that the common ways the concept has been used either to counter neartermism or counter longtermism are misguided. (I write about this here and here.) So I guess that that's probably consistent with your conclusion, though maybe by a different road. (I prefer to use the sort of analysis in Tarsney's epistemic challenge paper, and I think that that pushes in favour of either longtermism or further research on longtermism vs neartermism, though I definitely acknowledge room for debate on that.)
I think Tarsney's paper does not address/avoid cluelessness, or at least its spirit, i.e., the arbitrary weighting of different considerations, since
Does some s-risk (e.g. AI safety, authoritarianism) work reduce some extinction risks and so increase other s-risks, and how do we weigh those possibilities?
I worry that research on longtermism vs neartermism (like Tarsney's paper) just ignores these problems, since you really need to deal with somewhat specific interventions, because of the different considerations involved. In my view, (strong) longtermism is only true if you actually identify an intervention that you can only reasonably believe does (much) more net good in the far future in expectation than short-term-focused alternatives do in the short term in expectation, or, roughly, that you can only reasonably believe does (much) more good than harm (in the far future) in expectation. This requires careful analysis of a specific intervention, and we may not have the right information now or ever to confirm that a particular intervention satisfies these conditions. To every longtermist intervention I've tried to come up with specific objections to, I've come up with objections that I think could reasonably push it into doing more harm than good in expectation.
Of course, what should "reasonable belief" mean? How do we decide which beliefs are reasonable and which ones aren't (and the degree of reasonableness, if it's a fuzzy concept)?
Basically, I agree that longtermist interventions could have these downside risks, but:
I think this gets at part of what comes to mind when I hear objections like this.
Another part is: I think we could say all of that with regards to literally any decision - we'd often be less uncertain, and it might be less reasonable to think the decision would be net negative or astronomically so, but I think it just comes in degrees, rather than applying strongly to some scenarios and not at all applying to others. One way to put this is that I think basically every decision meets the criteria for complex cluelessness (as I argued in the above-mentioned links: here and here).
But really I think that (partly for that reason) we should just ditch the term "complex cluelessness" entirely, and think in terms of things like credal resilience, downside risk, skeptical priors, model uncertainty, model combination and adjustment, the optimizer's curse, best practice for forecasting, and expected values given all that.
Here I acknowledge that I'm making some epistemological, empirical, decision-theoretic, and/or moral claims/assumptions that I'm aware various people who've thought about related topics would contest (including yourself and maybe Greaves, both of whom have clearly "done your homework"). I'm also aware that I haven't fully justified these stances here, but it seemed useful to gesture roughly at my conclusions and reasoning anyway.
I do think that these considerations mostly push against longtermism and in favour of neartermism. (Caveats include things like being very morally uncertain, such that e.g. reducing poverty or reducing factory farming could easily be bad, such that maybe the best thing is to maintain option value and maximise the chance of a long reflection. But this also reduces option value in some ways. And then one can counter that point, and so on.) But I think we should see this all as a bunch of competing quantitative factors, rather than as absolutes and binaries.
(Also, as noted elsewhere, I currently think longtermism - or further research on whether to be longtermist - comes out ahead of neartermism, all-things-considered, but I'm unsure on that.)
I don't think it's usually reasonable to choose only one expected value estimate, though, and this to me is the main consequence of cluelessness. Doing your best will still leave a great deal of ambiguity if you're being honest about what beliefs you think would be reasonable to have, despite not being your own fairly arbitrary best guess (often I don't even have a best guess, precisely because of how arbitrary that seems). Sensitivity analysis seems important.
I would say complex cluelessness basically is just sensitivity of recommendations to model uncertainty. The problem is that it's often too arbitrary to come to a single estimate by combining models. Two people with access to all of the same information and even the same ethical views (same fundamental moral uncertainty and methods for dealing with them) could still disagree about whether an intervention is good or bad, or which of two interventions is best, depending basically on whims (priors, arbitrary weightings).
At least substantial parts of our credences are not very sensitive to arbitrariness with shorttermist interventions with good evidence, even if on the whole the expected value is, but the latter is what I hope hedging could be used to control. Maybe you can do this just with longtermist interventions, though. A portfolio of interventions can be less ambiguous than each intervention in it. (This is what my hedging post is about.)
tl;dr: I basically agree with your first paragraph, but think that:
Agreed. But: