I don’t think these examples illustrate that “bewaring of suspicious convergence” is wrong.
For the two examples I can evaluate (the climate ones), there are co-benefits, but there isn’t full convergence with regards to optimality.
On air pollution, the most effective intervention for climate are not the most effective intervention for air pollution even though decarbonization is good for both. See e.g. here (where the best intervention for air pollution would be one that has low climate benefits, reducing sulfur in diesel; and I think if that chart were fully scope-sensitive and not made with the intention to showcase co-benefits, the distinctions would probably be larger, e.g. moving from coal to gas is a 15x improvement on air pollution while only a 2x on emissions):
And the reason is that different target metrics (carbon emissions, reduced air pollution mortality) are correlated, but do not map onto each other perfectly and optimizing for one does not maximize the other.
Same thing with alternative proteins, where a strategy focused on reducing animal suffering would likely (depending on moral weights) prioritize APs for chicken, whereas a climate-focused strategy would clearly prioritize APs for beef. (There’s a separate question here whether alternative proteins are an optimal climate strategy, which I think is not really established).
I think what these examples show is that we often have interventions with correlated benefits and it is worth asking whether one should optimize for both metrics jointly (given that there isn’t really an inherent reason to separate, say, lives lost from climate from lives lost from air pollution it could make sense to prioritize interventions which do not optimize either dimension), but if one decides to optimize for a single dimension not expecting to also accidentally optimize the other dimension (i.e. “bewaring of suspicious convergence”) continues to be good advice (or, more narrowly, is at least not discredited by these examples).
To be clear, I think the original post is uncontroversially right that it’s very unlikely that the best intervention for A is also the best intervention for B. My claim is that, when something is well evidenced to be optimal for A and perhaps well evidenced to be high tier for B, you should have a relatively high prior that it’s going to be high tier or even optimal for some related concern C.
Where you have actual evidence available for how effective various interventions are for C, this prior is largely irrelevant—you look at the evidence in the normal way. But when all interventions targeting C are highly speculative (as they universally are for longtermism), that prior seems to have much more weight.
Just to fully understand—where does that intuition come from? Is it that there is a common structure to high impact? (e.g. if you think APs are good for animals you also think they might be good for climate, because some of the goodness comes from the evidence of modular scalable technologies getting cheap and gaining market share?)
Partly from a scepticism about the highly speculative arguments for ‘direct’ longtermist work—on which I think my prior is substantially lower than most of the longtermist community (though I strongly suspect selection effects, and that this scepticism would be relatively broadly shared further from the core of the movement).
Partly from something harder to pin down, that good outcomes do tend to cluster in a way that e.g. Givewell seem to recognise, but AFAIK have never really tried to account for (in late 2022, they were still citing that post while saying ‘we basically ignore these’). So if we’re trying to imagine the whole picture, we need to have some kind of priors anyway.* Mine are some combination of considerations like
there are a huge number of ways in which people tend to behave more generously when they receive generosity, and it’s possible the ripple effects of this are much bigger than we realise (small ripples over a wide group of people that are invisibly small per-person could still be momentous);
having healthier, more economically developed people will tend to lead to more having more economically developed regions (I didn’t find John’s arguments against randomistas driving growth persuasive—e.g. IIRC it looked at absolute effect size of randomista-driven growth without properly accounting for the relative budgets vs other interventions. Though if he is right, I might make the following arguments about short term growth policies vs longtermism);
having more economically countries seems better for global political stability than having fewer, so reduce the risk of global catastrophes;
having more economically developed countries seems better for global resilience to catastrophe than having fewer, so reduce the magnitude of global catastrophes;
even ‘minor’ (i.e. non-extinction) global catastrophes can have a substantial reduction on our long-term prospects, so reducing their risk and magnitude is a potentially big deal
tighter feedback loops and better data mean we can learn more about incidental-optimisations than we can with longtermism work, including ones we didn’t know at the time we wanted to optimise for—we build up a corpus of real-world data that can be referred to whenever we think of a new consideration
tighter feedback loops also mean I expect the people working on it to be more effective at what they do, and less susceptible to (being selected by or themselves being subject to) systemic biases/groupthink/motivated reasoning etc.
the combination of greater evidence base and tighter feedback loops has countless other ineffable reinforcing-general-good benefits, like greater probability of shutting down when having 0 or negative effect; better signalling; greater reasoning transparency; easier measurement of Shapley values vs rather than counterfactuals; faster and better process refinement etc
I don’t think these examples illustrate that “bewaring of suspicious convergence” is wrong.
For the two examples I can evaluate (the climate ones), there are co-benefits, but there isn’t full convergence with regards to optimality.
On air pollution, the most effective intervention for climate are not the most effective intervention for air pollution even though decarbonization is good for both.
See e.g. here (where the best intervention for air pollution would be one that has low climate benefits, reducing sulfur in diesel; and I think if that chart were fully scope-sensitive and not made with the intention to showcase co-benefits, the distinctions would probably be larger, e.g. moving from coal to gas is a 15x improvement on air pollution while only a 2x on emissions):
And the reason is that different target metrics (carbon emissions, reduced air pollution mortality) are correlated, but do not map onto each other perfectly and optimizing for one does not maximize the other.
Same thing with alternative proteins, where a strategy focused on reducing animal suffering would likely (depending on moral weights) prioritize APs for chicken, whereas a climate-focused strategy would clearly prioritize APs for beef.
(There’s a separate question here whether alternative proteins are an optimal climate strategy, which I think is not really established).
I think what these examples show is that we often have interventions with correlated benefits and it is worth asking whether one should optimize for both metrics jointly (given that there isn’t really an inherent reason to separate, say, lives lost from climate from lives lost from air pollution it could make sense to prioritize interventions which do not optimize either dimension), but if one decides to optimize for a single dimension not expecting to also accidentally optimize the other dimension (i.e. “bewaring of suspicious convergence”) continues to be good advice (or, more narrowly, is at least not discredited by these examples).
Hey Johannes :)
To be clear, I think the original post is uncontroversially right that it’s very unlikely that the best intervention for A is also the best intervention for B. My claim is that, when something is well evidenced to be optimal for A and perhaps well evidenced to be high tier for B, you should have a relatively high prior that it’s going to be high tier or even optimal for some related concern C.
Where you have actual evidence available for how effective various interventions are for C, this prior is largely irrelevant—you look at the evidence in the normal way. But when all interventions targeting C are highly speculative (as they universally are for longtermism), that prior seems to have much more weight.
Interesting, thanks for clarifying!
Just to fully understand—where does that intuition come from? Is it that there is a common structure to high impact? (e.g. if you think APs are good for animals you also think they might be good for climate, because some of the goodness comes from the evidence of modular scalable technologies getting cheap and gaining market share?)
Partly from a scepticism about the highly speculative arguments for ‘direct’ longtermist work—on which I think my prior is substantially lower than most of the longtermist community (though I strongly suspect selection effects, and that this scepticism would be relatively broadly shared further from the core of the movement).
Partly from something harder to pin down, that good outcomes do tend to cluster in a way that e.g. Givewell seem to recognise, but AFAIK have never really tried to account for (in late 2022, they were still citing that post while saying ‘we basically ignore these’). So if we’re trying to imagine the whole picture, we need to have some kind of priors anyway.* Mine are some combination of considerations like
there are a huge number of ways in which people tend to behave more generously when they receive generosity, and it’s possible the ripple effects of this are much bigger than we realise (small ripples over a wide group of people that are invisibly small per-person could still be momentous);
having healthier, more economically developed people will tend to lead to more having more economically developed regions (I didn’t find John’s arguments against randomistas driving growth persuasive—e.g. IIRC it looked at absolute effect size of randomista-driven growth without properly accounting for the relative budgets vs other interventions. Though if he is right, I might make the following arguments about short term growth policies vs longtermism);
having more economically countries seems better for global political stability than having fewer, so reduce the risk of global catastrophes;
having more economically developed countries seems better for global resilience to catastrophe than having fewer, so reduce the magnitude of global catastrophes;
even ‘minor’ (i.e. non-extinction) global catastrophes can have a substantial reduction on our long-term prospects, so reducing their risk and magnitude is a potentially big deal
tighter feedback loops and better data mean we can learn more about incidental-optimisations than we can with longtermism work, including ones we didn’t know at the time we wanted to optimise for—we build up a corpus of real-world data that can be referred to whenever we think of a new consideration
tighter feedback loops also mean I expect the people working on it to be more effective at what they do, and less susceptible to (being selected by or themselves being subject to) systemic biases/groupthink/motivated reasoning etc.
the combination of greater evidence base and tighter feedback loops has countless other ineffable reinforcing-general-good benefits, like greater probability of shutting down when having 0 or negative effect; better signalling; greater reasoning transparency; easier measurement of Shapley values vs rather than counterfactuals; faster and better process refinement etc