www.jimbuhler.site
Also on LessWrong (with different essays).
Jim Buhler
Karma: 1,138
...if you think welfare is net positive either way, yes. This seems like a tough case to make. I see how one can opt for agnosticism over believing net negative but I doubt there exists anything remotely close to a good case that WAW currently is net positive (and not just highly uncertain).
We take the meat-eater problem[3] seriously, but we don’t at all think that the conclusion is to avoid donating in the Global Health and Development (GHD) space: the effects might actually even out if e.g. further development reduces the total amount of natural space, potentially counterbalancing increased meat consumption by reducing the number of suffering wild animals.
Is the positive effect on wild animal welfare really your crux for finding GHD net positive? If yes, that means you think WAW is more pressing than improving human health. And it feels weird to advocate for improving human health despite the meat-eating pb because of wild animal suffering. If you really think that, it feels like you should just advocate for reducing wild animal suffering instead (unless you think GDH happens to be the best way to do that).
Damn, a friend just made me realize that I had mispelled your first name too. So sorry aha
I think if we only do spatiotemporal bracketing, it tells us to ignore the far future and causally inaccessible spacetime locations, because each such location is made neither determinately better off in expectation nor determinately worse off in expectation.
Oh helpful thanks, this reasoning also works in my sniper case, actually. I am clueful about the “where Emily is right after she potentially shoots” ST location so I can’t bracket out the payoff attached to her shoulder pain. This payoff is contained within this small ST region. However, the payoffs associated with where the bullet ends aren’t neatly contained in small ST regions the same way! I want the terrorist dead because he’s gonna keep terrorizing some parts of the world otherwise. I want the kid alive to prevent the negative consequences (in various ST regions) associated with an innocent kid’s death. Because of this, I arguably can’t pin down any specific ST location other than “where Emily is right after she potentially shoots” that is made determinately better or worse off by Emily taking the shot. Hence, ST bracketing would allow C but not A or B.
To the extent that I’m still skeptical of C being warranted, it is because:1) I find it weird that finding action-guidance depends on my inability to pin down any specific ST location other than “where Emily is right after she potentially shoots” that is made determinately better or worse off. Say I had a crystal ball randomly showing me a prison cell in Argentina that, for some reason, is empty if Emily shoots and filled with starving people if she doesn’t. ST bracketing would now tell me shooting is better… It feels wrong to decide based on isolated ST regions in which I happen to know what happens depending on whether Emily shoots. There are plenty of other ST regions that would be made better or worse off. I just can’t say where/when they are. And whether or not I can say this feels like it shouldn’t matter.[1]
2) I’m confused as to why we should bracket based on ST regions rather than on some other defensible value-bearers that may give a conflicting result.
And I guess all this also applies to A’ vs B’ vs C’ and whether to bracket out near-term effects. Thanks for helping me identify these cruxes!
I’ll take some more time to think about your point about bracketing out possibilities and AGI by date X.- ^
And that’s one way to interpret Anthony’s first objection to bracketing? I can’t actually pin down a specific ST location (or whatever value-bearer) where donating to AMF is determinately bad, but I still know for sure such locations exist! As I think you alluded to elsewhere while discussing ST bracketing and changes to agriculture/land use, what stops us from acting as if we could pin down such locations?
Interesting, thanks for the update!
If you weren’t doing [B] with moral weights, though, you would presumably have to worry about things other than effects on soil animals. So, ultimately, [B] remains an important crux for you.
(You could still say you’d prioritize decreasing uncertainty on moral weights if you thought there was too much uncertainty to justify doing [B], but the results from such research might never be precise enough to be action-guiding. You might have to endorse B despite the ambiguity, or one of the three others.)
Extinction forecloses all option value — including the option for future agents to course-correct if we’ve made mistakes. Survival preserves the ability to solve new problems. This isn’t a claim about net welfare across cosmic history; it’s a claim about preserving agency and problem-solving capacity.
I think it still implicitly is a claim about net welfare across the cosmos. You have to believe that preserving option value will actually eventually lead to higher net welfare across the cosmos[1]---belief which I argue relies on judgment calls. (And the option-value argument for x-risk reduction was kind of already infamously known as a bad one in the GPR literature, including among x-risk reducers.)
You might say individuals can act on non-longtermist grounds while remaining longtermist-clueless. But this concedes that something must break the paralysis, and I’d argue that “preserve option value / problem-solving capacity” is a principled way to do so that doesn’t require the full judgment-call apparatus you describe.
Nice, that’s the crux! Yeah so I tentatively find something like bracketing out long-term effects more principled (as a paralysis breaker) than option-value preservation. I have no clue whether reducing the agony of the many animals we can robustly help in the near term is overall good when considering the indirect long-term effects, but I find doing it anyway far more justifiable than “reducing x-risks and let future people decide what they should do”. I would prefer the latter if I bought the premises of the option-value argument for x-risk reduction, but I wouldn’t be clueless and wouldn’t have a paralysis problem to begin with, then.
I don’t see any good reason to believe enabling our descendants is impartially better than doing the exact opposite (both positions rely on judgment calls that seem arbitrary to me). However, I see good (non-longtermist) reasons to reduce near-term animal suffering rather than increase it.- ^
Unless you intrinsically value the existence of Earth-originated agents or something, and in a way where you’re happy to ignore the welfarist considerations that may leave you clueless on their own. In this case, you obviously think reducing P(extinction) is net positive. But then,
i) you don’t think that reducing P(extinction) is good for the welfare of all sentient beings, like most EAs. You care about x-risk reduction for other reasons and you have to be clear about that.
ii) you have to argue that it is not arbitrary to intrinsically value the existence of Earth-originated agents (or whatever non-welfarist and/or non-impartial thing) independently of what these agents do to moral patients. The reason why EAs like impartial welfarism is because anything else seems unfairly arbitrary, so if I have to give up on impartial welfarism in order to value reducing P(extinction), I solve a problem by creating a new one (and this solution to the problem would feel suspiciously convenient to me).
- ^
Nice, thanks! (I gave examples of charities/work where you’re kinda agnostic because of a crux other than AI timelines, but this was just to illustrate.)
Assuming that saving human lives increases welfare, I agree doing it earlier increases welfare more if TAI happens earlier.
I had no doubts you thought this! :) I’m just curious as to whether you see reasons for someone to optimize assuming long AI timelines, despite low resilience in their high credence in long AI timelines.
(Hey Vasco!) How resilient is your relatively high credence that AI timelines are long?
And would you agree that the less resilient it is, the more you should favor interventions that are also good under short AI timelines? (E.g., the work of GiveWell’s top charities over making people consume fewer unhealthy products, since the latter pays off far later, as you and Michael discuss in this thread.)
it seems pretty likely to me that aquatic noise reduces populations (and unlikely that it increases them), both fish and invertebrates, by increasing mortality and reducing fertility.
What about trophic cascades? Maybe the populations most directly affected and reduced by aquatic noise were essential for keeping overall wild animal populations down?
Do you think aquatic noise is like some specific forms of fishing that determinately reduce overall populations? Is it because you think it directly affects/reduces all populations (unlike some other specific forms of fishing) such that trophic cascades can hardly compensate?
if we’re clueless whether Emily will feel pain or not then the difference disappears. In this case I don’t have the pro-not-shooting bracketing intuition.
Should this difference matter if we’re not difference-making risk-averse or something? In both cases, C is better for Emily in expectation (the same way reducing potential termite suffering is better for termites, in expectation, even if it might make no difference because they might not be sentient).
Now, new thought experiment. Consider whatever intervention you find robustly overall good in the near-term (without bracketing out any near-term effect) and replace A, B, and C with the following:A’) Bracket in the bad long-term effects (-> don’t intervene)
B’) Bracket in the good long-term effects (-> intervene)
C’) Bracket in the near-term effects (-> intervene)
Do you have the pro-C’ intuition, then? If yes, what’s different from the sniper case?
Interesting! This resembles Michael St. Jules’ hedging proposal.
But the animal representatives, even if they aren’t thrilled by retaining another omnivore, have more pressing priorities than trying to help animals by eliminating meat-eaters one by one.
Is it that obvious? I find it hard to come up with interventions that reduce farmed animal suffering (and farmed animal suffering only, ignoring all the rest) more robustly than “minimize the number of entities that incidentally perpetuate this farmed animal suffering”. I feel like this should actually be pretty high up the list for someone who strictly spotlights farmed animals (like the farmed animal representatives are supposed to in your spotlighting parliament, if I understand correctly).
Why don’t the managers of the Animal Welfare Fund support preventing the existence of more potential meat-eaters, then? (via, e.g., increasing x-risks.) I presume becausemost of their biggest donors don’t spotlight farmed animals, are concerned about effects on other things, and would stop donating to the fund if the managers were to do that.
the fund managers themselves, like most farmed animal people, also don’t spotlight farmed animals and are concerned about effects on other things.
Curious if you agree.
For the moral trade between the representatives of human victims of malaria and farmed animal representatives to be fair, in your setup, the preferences of the latter would have to actually stoplight farmed animals the same way the former spotlights human victims of malaria. I.e., the preferences of farmed animal representatives in your spotlighting parliament should not be those of real farmed animal advocates who are not spotlighting farmed animals (otherwise, they would obviously be pro-x-risks and stuff despite the downsides on other beings, the same way the representatives of human malaria victims are anti-poverty despite the meat-eater pb).
I would still say there are actions which are robustly beneficial in expectation, such as donating to SWP. It is possible SWP is harmful, but I still think donating to it is robustly better than killing my family, friends, and myself, even in terms of increasing impartial welfare.
It’s kinda funny to reread this 6 months later. Since then, the sign of your precise best guess flipped twice, right? You argued somewhere (can’t find the post) that shrimp welfare actually was slightly net bad after estimating that it increases soil animal populations. Later, you started weakly believing animal farming actually decreases the number of soil nematodes (which morally dominate in your view), which makes shrimp welfare (weakly) good again.
(Just saying this to check if that’s accurate because that’s interesting. I’m not trying to lead you into a trap where you’d be forced to buy imprecise credences or retract the main opinion you defend in this comment thread. As I suggest in this comment, let’s maybe discuss stuff like this on a better occasion.)
I suspect Vasco is reasoning about the implications of epistemic principles (applied to our evidence) in a way I’d find uncompelling even if I endorsed precise Bayesianism.
Oh so for the sake of argument, assume the implications he sees are compelling. You are unsure about whether your good epistemic principles E imply (a) or (b).[1]
So then, the difference between (a) and (b) is purely empirical, and MNB does not allow me to compare (a) and (b), right? This is what I’d find a bit arbitrary, at first glance. The isolated fact that the difference between (a) and (b) is technically empirical and not normative doesn’t feel like a good reason to say that your “bracket in consequentialist bracketing” move is ok but not the “bracket in ex post neartermism” move (with my generous assumptions in favor of ex post neartermism).- ^
I don’t mean to argue that this is a reasonable assumption. It’s just a useful one for me to understand what moves MNB does and does not allow. If you find this assumption hard to make, imagine that you learn that we likely are in simulation that is gonna shut down in 100 years and that the simulators aren’t watching us (so we don’t impact them).
- ^
I find impartial consequentialism and indeterminate beliefs very well-motivated, and these combined with consequentialist bracketing seem to imply neartermism (as Kollin et al. (2025) argue), I think it’s plausible that metanormative bracketing implies neartermism.
Say I find ex post neartermism (Vasco’s view that our impact washes out, ex post, after say 100 years) more plausible than consequentialist bracketing being both correct and action-guiding.
My favorite normative view (impartial consequentialism + plausible epistemic principles + maximality) gives me two options. Either:(a) long-term effects dominate, and I’m clueless.
(b) near-term effects dominate, and I know what to do (without having to use consequentialist bracketing, let’s assume).
Would you say that what dictates my view on (a)vs(b) is my uncertainty between different epistemic principles, such that I can dichotomize my favorite normative view based on the epistemic drivers of (a)vs(b)? (Such that, then, MNB allows me to bracket out the new normative view that implies (a) and bracket in the new normative view that implies (b), assuming no sensitivity to individuation.)
If not, I find it a bit arbitrary that MNB allows your “bracket in consequentialist bracketing” move and not this “bracket in ex post neartermism” move.
I was implicitly assuming the probability of hitting the kid or the terrorist is high enough that where the bullet ends strictly matters more than Emily’s pain. If I misunderstood you and this doesn’t address your point, we could also assume that Emily only might have shoulder pain if she takes the shot. Then the difference you point to disappears, right? (And this changes nothing to the thought experiment, assuming risk neutrality and stuff.)
This also makes this second difference disappear, right? On B and C, we’re actually clueful on the out-bracket (the terrorist dwarfs Emily, so it’s better to shoot in expectation). So it’s symmetric to cluefulness on the out-bracket on A.
Spent some more time thinking about this, and I think I mostly lost my intuition in favor of bracketing in Emily’s shoulder pain. I thought I’d share here.
The problem
In my contrived sniper setup, I’ve gotta do something, and my preferred normative view (impartial consequentialism + good epistemic principles + maximality) is silent. Options I feel like I have:
A) Bracket out the kid, but not the terrorist (-> shooting is better)
B) Bracket out the terrorist, but not the kid (-> no shooting is better)
C) Bracket out both the kid and the terrorist, but not Emily[1] (-> no shooting is better)
D) Flip a coin, whatever. This illustrates radical cluelessness.
All these options feel arbitrary, but I have to pick something.
Comparing poisons
Picking D demands accepting the arbitrariness of letting perfect randomness guide our actions. We can’t do worse than this.[2] It is the total-arbitrariness baseline we’re trying to beat.
Picking A or B demands accepting the arbitrariness of favoring one over the other, while my setup does not give me any good reason to do so (and A and B give opposite recommendations). I could pick A by sorta wagering on, e.g., an unlikely world where the kid dies of Reye’s syndrome (a disease that affects almost only children) before the potential bullet hits anything. But I could then also pick B by sorta wagering on the unlikely world where a comrade of the terrorist standing near him turns on him and kills him. And I don’t see either of these two wager moves as more warranted than the other.[3]Picking C, similarly, demands accepting the arbitrariness of favoring it over A (which gives the opposite recommendation), while my setup does not give me any good reason to do so. I could pick C by wagering on, e.g., an unlikely world where time ends between the potential shot hurting Emily’s shoulder and the moment the potential bullet hits something. But I could then also pick A by wagering on the unlikely world where the kid dies of Reye’s syndrome anyway. And same pb as above.[4] And this is what Anthony’s first objection to bracketing gestures at, I guess.
While I have a strong anti-D intuition with this sniper setup, it doesn’t favor C over A or B for me, at the very moment of writing.[5]
Should we think that our reasons for C are “more grounded” than our reasons for A, or something like that? I don’t see why. Is there a variant of this sniper story where it seems easier to argue that it is the case (while conserving the complex cluelessness assumption)? And is such a variant a relevant analogy to our real-world predicament?
- ^
Without necessarily assuming persons-based bracketing (for A, B, or C), but rather whatever form of bracketing results in ignoring the payoffs associated with one or two of the three relevant actors.
- ^
Our judgment calls can very well be worse than random due to systematic biases (and I remember reading somewhere in the forecasting literature that this happens). But if we believe that’s our case, we can just do the exact opposite of what our judgment calls say and this beats a coin flip.
- ^
It feels like I’m just adding non-decisive mildly sweet considerations on top of the complex cluelessness pile I already had (after thinking about the different wind layers, the Earth’s rotation, etc). This will not allow me to single out one of these considerations as a tie-breaker.
- ^
This is despite some apparent kind of symmetry existing only between A and B (not between C and A) that @Nicolas Mace recently pointed to in some doc comment—symmetry which may feel normatively relevant although it feels superficial to me at the very moment of writing.
- ^
In fact, given the apparent stakes difference between Emily’s shoulder pain and where the bullet ends, I may be more tempted to act in accordance to A or B, deciding between the two based on what seems to be the least arbitrary tie-breaker. However, not sure whether this temptation is, more precisely, one in favor of endorsing A or B, or in favor of rejecting cluelessness and the need for bracketing to begin with, or something else.
If most of the value we can influence is in the far future
To be clear, you don’t necessarily assume this in the paper, and you don’t need to, right? You need bracketing to escape cluelessness paralysis, even if you merely think it’s indeterminate whether most of the value we can influence is in the far future, afaiu.
One could try to argue that the second-order effects of near-term interventions are negligible in expectation (see “the washing out hypothesis”). But I don’t think this is plausible.
So even if this were plausible (as Vasco thinks, for instance), this wouldn’t be enough to think we don’t need bracketing. One would need to have determinate-ish beliefs that rule out the possibility of far future effects dominating.
Yup, something a variety of views can get behind. E.g., not “buying beef”.
For “consensual EAA interventions” above, I think I was thinking more “not something EAs see as ineffective like welfare reforms for circus animals”. If this turned out to be the safest animal intervention, I suspect this wouldn’t convince many EAs to consider it. But if, say, developing alternatives to rodents as snake food turned out to be very safe, this could weigh a lot in its favor for them.
Nice, thanks for engaging! :)
It sounds like this is actually the core crux of your view, then. If so, it might be worth making that explicit in the post. As it stands, the discussion of WAW could give the impression that it plays a more decisive role in your evaluation than it ultimately does, whereas your judgment seems to rest mainly on the effects on human welfare, given what you say here.
I also think this position of yours (that is now revealed) invites further scrutiny. Given how many more animals are plausibly affected by GDH compared to humans, concluding that AW is not the most important factor appears to rely on specific assumptions about moral weights that privilege humans to an extent that would be very controversial if it were made explicit. It could be helpful to spell those assumptions out, or at least acknowledge that they’re doing significant work here.