Two things: First, it should be called strongtermism, because it's shorter and also makes the whole thing feel cooler. But second, have you ever read "Fanaticism and Knowledge" by Frank Hong? (https://philarchive.org/rec/HONFAK) I broadly agree with all your points here but I think it's a very plausible way to justify the "discount low risks" approach that solves a lot of the problems you're raising. But it also depends heavily on whether you buy knowledge-first epistemology and some other controversial views. If you haven't read it, you should check it out.
I haven't, though I'll give it a look. Out of curiosity: do you think it helps with avoiding the result that discounters should be Longtermists? Or just that it is a more plausible kind of discounting.
That's a hard question to answer, haha - I'd say it makes certain forms of "tail discounting" much more defensible, but how exactly that maps onto longtermist considerations in particular is harder for me to judge since it's not an area I'm super familiar with. Just in general though it's a good way to fix the "Every individual bad result is rare so discounting means acting like there's no problem" situation you bring up with the torture example.
Yeah I don't think it's decisive in this particular case, it's just an interesting solution to many of the generalized problems attributed to the "ignore small risks" position.
There is an error in your chaos response. You say:
> If Longtermist actions have higher expected impact and higher or roughly equivalent probabilities of actually making things better, then shouldn’t you just pick them?
Yes, but the chaos argument is best understood as challenging the claim that longterm actions have such a higher expected impact. Because our certainly in the effects an action will have decreases with time that reduces the expected effect of actions that impact the far future.
Here is a plausible model. At least while we haven't left earth we can assume that outcomes will always fall between 50 billion people living in pure ecsatasy and 50 billion people living in hellish squalor. Because these bounds prevent arbitrary large positive and negative effects the increased uncertainty of far future effects drags their expected delta impact to 0.
If I have a comment on one of your old posts you linked is it better to post it there or somewhere more recent so that you will see it? Anyway, I wrote a comment on your "Why I'm a fanatic" post,
The post does acknowledge harmful persistent states - extinction, lock-in, extreme power concentration - but it seems to treat longtermist interventions mainly as ways of reducing those risks. What is less clear is how it handles sign uncertainty: the possibility that actions aimed at improving the far future may themselves increase the probability of bad persistent states, or have negative effects of comparable magnitude.
If the future is vast, then both positive and negative long-run effects are vast. So scale alone does not establish positive expected value. One also needs an argument that the intervention shifts the probability distribution in the right direction, rather than merely increasing our influence over a domain where the sign of that influence is highly uncertain.
I think the veil of ignorance is good here. We should pretend we don't know what generation we'd be in. How many resources would we want each generation to spend on the next? It's hard to model this--not least because the chance we exist depends on what we say and whatnot. But it seems like the equilibrium will be quite a high rate of resources to future generations.
Rawls talks about this topic in TOJ, and what he says is a little more complicated. But I think his prior generations test would follow this quite closely once we interpret the OP properly.
By "interepet properly" I mean that we don't know what generation we are. Rawls doesn't go this way, but I think it's an interpretation of the OP that's more accurate to model moral questions.
Genuine question: will you change what you write about / work on as a result? Seems like many of your current areas of interest are not strong-longtermist! Without *some* anchor to the present, it seems hard to justify working on animal welfare, global poverty, political discourse, or basically anything except existential risk. For what it's worth, my interests are also not strong-longtermist compatible...
I think you are equivocating in your (second group of) steps 1-3 in Section 3. For premise 2 to work you need to weight the number of people you affect by how much you knowably affect them. But if you do this, it is not clear why premise 1 should be true.
Integrated over time, this means that moral priority depends on the balance between increasing numbers and decreasing epistemic reliability. Depending on the assumptions, the optimal target of action may lie neither exclusively among present people nor at the furthest possible future, but somewhere along a temporal curve.
However, there may also be plausible interaction effects across generations that further complicate matters: benefiting present people can indirectly benefit future people by enabling actions that positively affect them later on. For example, helping existing people flourish and reproduce may be a precondition for those future people existing at all.
So my intuition would be that relative shorttermism would usually be the outcome of calculations taking all this into account
Whenever I try to pencil it out, you either end up with (a) people now to ~30-50 years from now are the only thing that matters, or (b) the ultra-far-future is the only thing that matters. Your discounting/etc create a real knife's edge that make things tilt one way or the other depending on your specific choices.
I agree that, if we assume the sign of the far-future effect remains positive, the model may tilt either toward near-termism or ultra-longtermism depending on the discounting curve.
But that assumption is doing a lot of work. My worry is not only that positive far-future effects become harder to trace or should be discounted. It is that the sign of the effect itself may become increasingly uncertain. Over very long timescales, an intervention aimed at improving the future may also increase the probability of bad lock-in, power concentration, technological acceleration, or other harmful persistent states.
So the issue is not just “how much should we discount a positive expected far-future effect?” It is whether we are entitled to treat the far-future term as positive in the first place. If both upside and downside scale astronomically, then sheer temporal reach does not by itself favour ultra-longtermism. It only does so if we have good reason to think the intervention shifts the distribution in the right direction.
Ah, I see. Tyler Cowen actually has a paper that's sort of about this problem, called 'The Epistemic Problem Does Not Refute Consequentialism' - in part of it, he talks about the "high variance" of future outcomes (could be very positive, could be very negative)
It's funny that in this article you defend that more beings is more better, and we should care about long-term survival, but you have argued elsewhere that it would be better to just pave half the world so that there would be less insects and therefore less suffering, even though we need functional ecosystems, including those insects you want to remove from the planets surface, to survive.
Secondly, the fact that you consider 'space' as what's needed for long-term survival severely weakens how serious I can take your arguments.
I do agree that we need to care about the future, and future generations, but I think putting the emphasis on a far away future is wrong. Mainly because of uncertainty; for one thing the actions you might now take are more likely to be undone the further away in time we come. Secondly you don't know what is needed for future survival. The fact that some of the longtermist crowd for some reason promote going onto different planets, a certain death, as what's needed for survival shows that longtermism has too high a risk of promoting delusional ideas to be capable of realizing their own goals of long term survival. The best thing you can do is to ensure that the next few generations survive, and will have at least the same level of resources needed for survival as we do (but preferably even more, especially of knowledge).
There’s no formal inconsistency between saying that we want 10^54 happy humans and that we do not want quintillions of suffering insects.
And as a factual matter, the latter is not necessary to the former.
Putting “space” in “scare quotes” is not a “serious” “argument.” Space both reduces existential risks and increases the total amount of energy humans can use to create more happy humans.
You're forgetting the part we're I'm saying we need those insects.
And I'm not using "scare quotes" I was using it as a shorthand, considering that 'space' itself is not an answer, but it's just a convenient way of referring to all the things we're supposedly going to do there.
And space doesn't reduce any risk, it merely ensures that the people going out there will die sooner rather than later.
I think there is some possibility that no actions are better in expectation than any other due to weird stuff with infinities. That said, this doesn't seem like a guarantee, and we can maybe reject it on Pascalian grounds. Also maybe something something hyperreals.
Schwitzgebel argues that we should discount the far future. This seems a bit crazy--it implies that a trillion people being tortured in a bunch of years is less bad than one person stubbing their toe today.
I like this post and will use it in debate class.
Two things: First, it should be called strongtermism, because it's shorter and also makes the whole thing feel cooler. But second, have you ever read "Fanaticism and Knowledge" by Frank Hong? (https://philarchive.org/rec/HONFAK) I broadly agree with all your points here but I think it's a very plausible way to justify the "discount low risks" approach that solves a lot of the problems you're raising. But it also depends heavily on whether you buy knowledge-first epistemology and some other controversial views. If you haven't read it, you should check it out.
I haven't, though I'll give it a look. Out of curiosity: do you think it helps with avoiding the result that discounters should be Longtermists? Or just that it is a more plausible kind of discounting.
That's a hard question to answer, haha - I'd say it makes certain forms of "tail discounting" much more defensible, but how exactly that maps onto longtermist considerations in particular is harder for me to judge since it's not an area I'm super familiar with. Just in general though it's a good way to fix the "Every individual bad result is rare so discounting means acting like there's no problem" situation you bring up with the torture example.
Having now read it I don't think it helps avoid the view that risk discounting still implies longtermism.
Yeah I don't think it's decisive in this particular case, it's just an interesting solution to many of the generalized problems attributed to the "ignore small risks" position.
I am very much enjoying this series of articles on the morality of a strange, nonhuman species.
There is an error in your chaos response. You say:
> If Longtermist actions have higher expected impact and higher or roughly equivalent probabilities of actually making things better, then shouldn’t you just pick them?
Yes, but the chaos argument is best understood as challenging the claim that longterm actions have such a higher expected impact. Because our certainly in the effects an action will have decreases with time that reduces the expected effect of actions that impact the far future.
Here is a plausible model. At least while we haven't left earth we can assume that outcomes will always fall between 50 billion people living in pure ecsatasy and 50 billion people living in hellish squalor. Because these bounds prevent arbitrary large positive and negative effects the increased uncertainty of far future effects drags their expected delta impact to 0.
If I have a comment on one of your old posts you linked is it better to post it there or somewhere more recent so that you will see it? Anyway, I wrote a comment on your "Why I'm a fanatic" post,
The post does acknowledge harmful persistent states - extinction, lock-in, extreme power concentration - but it seems to treat longtermist interventions mainly as ways of reducing those risks. What is less clear is how it handles sign uncertainty: the possibility that actions aimed at improving the far future may themselves increase the probability of bad persistent states, or have negative effects of comparable magnitude.
If the future is vast, then both positive and negative long-run effects are vast. So scale alone does not establish positive expected value. One also needs an argument that the intervention shifts the probability distribution in the right direction, rather than merely increasing our influence over a domain where the sign of that influence is highly uncertain.
I think the veil of ignorance is good here. We should pretend we don't know what generation we'd be in. How many resources would we want each generation to spend on the next? It's hard to model this--not least because the chance we exist depends on what we say and whatnot. But it seems like the equilibrium will be quite a high rate of resources to future generations.
Rawls talks about this topic in TOJ, and what he says is a little more complicated. But I think his prior generations test would follow this quite closely once we interpret the OP properly.
By "interepet properly" I mean that we don't know what generation we are. Rawls doesn't go this way, but I think it's an interpretation of the OP that's more accurate to model moral questions.
Genuine question: will you change what you write about / work on as a result? Seems like many of your current areas of interest are not strong-longtermist! Without *some* anchor to the present, it seems hard to justify working on animal welfare, global poverty, political discourse, or basically anything except existential risk. For what it's worth, my interests are also not strong-longtermist compatible...
I think you are equivocating in your (second group of) steps 1-3 in Section 3. For premise 2 to work you need to weight the number of people you affect by how much you knowably affect them. But if you do this, it is not clear why premise 1 should be true.
Integrated over time, this means that moral priority depends on the balance between increasing numbers and decreasing epistemic reliability. Depending on the assumptions, the optimal target of action may lie neither exclusively among present people nor at the furthest possible future, but somewhere along a temporal curve.
However, there may also be plausible interaction effects across generations that further complicate matters: benefiting present people can indirectly benefit future people by enabling actions that positively affect them later on. For example, helping existing people flourish and reproduce may be a precondition for those future people existing at all.
So my intuition would be that relative shorttermism would usually be the outcome of calculations taking all this into account
Whenever I try to pencil it out, you either end up with (a) people now to ~30-50 years from now are the only thing that matters, or (b) the ultra-far-future is the only thing that matters. Your discounting/etc create a real knife's edge that make things tilt one way or the other depending on your specific choices.
I agree that, if we assume the sign of the far-future effect remains positive, the model may tilt either toward near-termism or ultra-longtermism depending on the discounting curve.
But that assumption is doing a lot of work. My worry is not only that positive far-future effects become harder to trace or should be discounted. It is that the sign of the effect itself may become increasingly uncertain. Over very long timescales, an intervention aimed at improving the future may also increase the probability of bad lock-in, power concentration, technological acceleration, or other harmful persistent states.
So the issue is not just “how much should we discount a positive expected far-future effect?” It is whether we are entitled to treat the far-future term as positive in the first place. If both upside and downside scale astronomically, then sheer temporal reach does not by itself favour ultra-longtermism. It only does so if we have good reason to think the intervention shifts the distribution in the right direction.
Ah, I see. Tyler Cowen actually has a paper that's sort of about this problem, called 'The Epistemic Problem Does Not Refute Consequentialism' - in part of it, he talks about the "high variance" of future outcomes (could be very positive, could be very negative)
It's funny that in this article you defend that more beings is more better, and we should care about long-term survival, but you have argued elsewhere that it would be better to just pave half the world so that there would be less insects and therefore less suffering, even though we need functional ecosystems, including those insects you want to remove from the planets surface, to survive.
Secondly, the fact that you consider 'space' as what's needed for long-term survival severely weakens how serious I can take your arguments.
I do agree that we need to care about the future, and future generations, but I think putting the emphasis on a far away future is wrong. Mainly because of uncertainty; for one thing the actions you might now take are more likely to be undone the further away in time we come. Secondly you don't know what is needed for future survival. The fact that some of the longtermist crowd for some reason promote going onto different planets, a certain death, as what's needed for survival shows that longtermism has too high a risk of promoting delusional ideas to be capable of realizing their own goals of long term survival. The best thing you can do is to ensure that the next few generations survive, and will have at least the same level of resources needed for survival as we do (but preferably even more, especially of knowledge).
There’s no formal inconsistency between saying that we want 10^54 happy humans and that we do not want quintillions of suffering insects.
And as a factual matter, the latter is not necessary to the former.
Putting “space” in “scare quotes” is not a “serious” “argument.” Space both reduces existential risks and increases the total amount of energy humans can use to create more happy humans.
You're forgetting the part we're I'm saying we need those insects.
And I'm not using "scare quotes" I was using it as a shorthand, considering that 'space' itself is not an answer, but it's just a convenient way of referring to all the things we're supposedly going to do there.
And space doesn't reduce any risk, it merely ensures that the people going out there will die sooner rather than later.
BB's said elsewhere he favors reducing insect populations at the margins, but not to the extent that humanity would go extinct.
Do you have a response to Eric Schwitzgebel's infinite washout objection (presented in his article The Washout Argument Against Longtermism)?
I think there is some possibility that no actions are better in expectation than any other due to weird stuff with infinities. That said, this doesn't seem like a guarantee, and we can maybe reject it on Pascalian grounds. Also maybe something something hyperreals.
Schwitzgebel argues that we should discount the far future. This seems a bit crazy--it implies that a trillion people being tortured in a bunch of years is less bad than one person stubbing their toe today.