Thanks for this. It’s useful for the community to think about this kind of thing and this is well-argued.
Overall
1. It’s a good point that since the top AI fields seem oversubscribed, it might be worth some people moving into the next best causes. Another possibility is that they should wait until the number of organisations catches up with the number of people. It might even be that the most valuable options is having a reserve of a large number of people who could, with some probability, be a good fit for the highest-impact orgs, even though most of these people never end up working for high-impact orgs. This puts a new slant on the demandingness of EA: rather than making sacrifices by donating, EAs make sacrifices by being prepared to accept the substantial probability of themselves never having impact. This would be hard to take psychologically, but might be the right thing to do in a crowded talent space.
2. On indirect risks, another point I make in the FP report is that while climate change is an indirect stressor of other risks, this suggests to me that working on those terminal risks directly would be a better bet than working on climate change since climate change is such an indirect stressor, is very crowded and seems difficult to make progress on. What do you think of that argument?
ITN
3. I don’t think it is right that problems with high tractability should be de-prioritised. I think what you mean is that we should focus on things that shift the long-term trajectory of humanity. But these could be highly tractable. e.g. the problem of not starting nuclear war was tractable for Vasili Arkhipov, but plausibly had large long-term effects. Having looked at it in some depth, climate change does look an intractable problem overall and this is indeed a reason not to work on it.
4. Another good point on how there could be increasing returns to scale in climate change, as we could affect the huge pool of funds going to the space through engagement.
5. Really, the ITN perhaps shouldn’t be used when we have cost-effectiveness estimates. On the 80k rendering, ITN is literally a cost-effectiveness estimate. But we now have cost-effectiveness estimates of climate charities. If we can make plausible estimates of the impact of bio, AI and nuclear, then we should use those, rather than appealing to the ITN. similarly, for use of time as well as money.
5. It is premature to say that work on climate change could be tractable. I think careful analysis is needed to figure out whether the things you list are indeed a good bet compared to other things that EAs could do.
Details
6. Climate Action Tracker suggests that on current policy, we are in for 3.1 to 3.5C, which is different to the ‘baseline’ trajectory estimate that you give. I think the current policy trajectory is most relevant for that part of your argument. (But note that this is only by 2100)
7. The impact of climate change on food production is in fact predicted to be fairly modest, as I discuss here. Yields might fall by 10-20% but this will be in the context of rising productivity and improvement in the other factors that determine the supply of food.
8. The emphasis on water shortage throughout is a bit overblown. We don’t need to ration water, we just need to price it properly (which is efficient rationing). If we did that, there would be no water problems today or in the future, anywhere (provided people had enough money).
“EAs make sacrifices by being prepared to accept the substantial probability of themselves never having impact. This would be hard to take psychologically, but might be the right thing to do in a crowded talent space.”
My impression has always been that even the most qualified person who goes into the most promising field (say, for example, AI risk reduction) has a low absolute chance of being the person to make a breakthrough in that field, but rather, that part of the point of EA was to get more talented people into those fields (e.g. by increasing the number of jobs) to increase the chance that a breakthrough will be made by someone.
Thanks for your comments—really great to learn more about this topic.
1. Agree, and optimisation in some soft-EA fields could have benefit
2. Agree—all other things equal, it would be better to work directly on an x-risk. But it would benefit EA, for the later reasons given, to acknowledge climate change as a potential stressor on x-risk. The assumption I’m using is that climate change might be more tractable for a bigger pool of people. And if someone is concerned about x-risk but is an expert on renewable energy on the breakthrough of some new technology, then they could understand their work as reducing the overall portfolio of x-risk. And if direct x-risks are heavily oversubscribed, or based on only a small number of agents (e.g. nuclear), then perhaps there’s more leverage for some people on climate change.
3. Agree that we should be focusing on things which affect the overall trajectory of civilisation. Is climate change really an intractable problem? In that case, why do so many smart people at all these universities, and the UN and IPCC have reducing emissions as a goal? Is it maybe intractable to assume that we’ll get to net zero, but is it a worthwhile goal to push to lower the rate of warming to give us more time to adapt?
I don’t profess to have the answer, but I’d be interested in the debate. I worry that this discussion doesn’t have enough input from real experts in this space.
If climate change is intractable, then what’s the next step? Should we be looking at geoengineering, adaptation, resilience? Assuming that climate change is intractable, then here are some other rough ideas of things that could help global welfare:
Early warning systems for floods in areas with anticipated rising sea levels
Research onto how societies should manage heat stress
Research and development of more resilient infrastructure, e.g. energy, food, and water—even if it’s just theoretically a question of pricing water etc. accurately, I’m not confident that in practice globally we’re doing that very well at the moment
4,5 - Interesting to read, I don’t profess to be an expert so would appreciate learning from other perspectives.
6,7,8- I wonder whether it’s possible that some modelling is overconfident on how resilient societies will be to climate change, as we’re densely networked and there might be lots of unanticipated secondary effects, such as mosquitoes affecting another billion people. The latest UK adaptation report acknowledges biodiversity as one of the several areas urgently needing further research.
I’ll focus on point 2 because I think it is the most important. I don’t see the argument for it being true that for the vast majority of people, working on climate change promises more leverage on the problem of nuclear war, than does working directly on nuclear war. Nuclear war is easier to make progress on, more neglected and more important than climate change.
Thanks for this. It’s useful for the community to think about this kind of thing and this is well-argued.
Overall
1. It’s a good point that since the top AI fields seem oversubscribed, it might be worth some people moving into the next best causes. Another possibility is that they should wait until the number of organisations catches up with the number of people. It might even be that the most valuable options is having a reserve of a large number of people who could, with some probability, be a good fit for the highest-impact orgs, even though most of these people never end up working for high-impact orgs. This puts a new slant on the demandingness of EA: rather than making sacrifices by donating, EAs make sacrifices by being prepared to accept the substantial probability of themselves never having impact. This would be hard to take psychologically, but might be the right thing to do in a crowded talent space.
2. On indirect risks, another point I make in the FP report is that while climate change is an indirect stressor of other risks, this suggests to me that working on those terminal risks directly would be a better bet than working on climate change since climate change is such an indirect stressor, is very crowded and seems difficult to make progress on. What do you think of that argument?
ITN
3. I don’t think it is right that problems with high tractability should be de-prioritised. I think what you mean is that we should focus on things that shift the long-term trajectory of humanity. But these could be highly tractable. e.g. the problem of not starting nuclear war was tractable for Vasili Arkhipov, but plausibly had large long-term effects. Having looked at it in some depth, climate change does look an intractable problem overall and this is indeed a reason not to work on it.
4. Another good point on how there could be increasing returns to scale in climate change, as we could affect the huge pool of funds going to the space through engagement.
5. Really, the ITN perhaps shouldn’t be used when we have cost-effectiveness estimates. On the 80k rendering, ITN is literally a cost-effectiveness estimate. But we now have cost-effectiveness estimates of climate charities. If we can make plausible estimates of the impact of bio, AI and nuclear, then we should use those, rather than appealing to the ITN. similarly, for use of time as well as money.
5. It is premature to say that work on climate change could be tractable. I think careful analysis is needed to figure out whether the things you list are indeed a good bet compared to other things that EAs could do.
Details
6. Climate Action Tracker suggests that on current policy, we are in for 3.1 to 3.5C, which is different to the ‘baseline’ trajectory estimate that you give. I think the current policy trajectory is most relevant for that part of your argument. (But note that this is only by 2100)
7. The impact of climate change on food production is in fact predicted to be fairly modest, as I discuss here. Yields might fall by 10-20% but this will be in the context of rising productivity and improvement in the other factors that determine the supply of food.
8. The emphasis on water shortage throughout is a bit overblown. We don’t need to ration water, we just need to price it properly (which is efficient rationing). If we did that, there would be no water problems today or in the future, anywhere (provided people had enough money).
“EAs make sacrifices by being prepared to accept the substantial probability of themselves never having impact. This would be hard to take psychologically, but might be the right thing to do in a crowded talent space.”
My impression has always been that even the most qualified person who goes into the most promising field (say, for example, AI risk reduction) has a low absolute chance of being the person to make a breakthrough in that field, but rather, that part of the point of EA was to get more talented people into those fields (e.g. by increasing the number of jobs) to increase the chance that a breakthrough will be made by someone.
Thanks for your comments—really great to learn more about this topic.
1. Agree, and optimisation in some soft-EA fields could have benefit
2. Agree—all other things equal, it would be better to work directly on an x-risk. But it would benefit EA, for the later reasons given, to acknowledge climate change as a potential stressor on x-risk. The assumption I’m using is that climate change might be more tractable for a bigger pool of people. And if someone is concerned about x-risk but is an expert on renewable energy on the breakthrough of some new technology, then they could understand their work as reducing the overall portfolio of x-risk. And if direct x-risks are heavily oversubscribed, or based on only a small number of agents (e.g. nuclear), then perhaps there’s more leverage for some people on climate change.
3. Agree that we should be focusing on things which affect the overall trajectory of civilisation. Is climate change really an intractable problem? In that case, why do so many smart people at all these universities, and the UN and IPCC have reducing emissions as a goal? Is it maybe intractable to assume that we’ll get to net zero, but is it a worthwhile goal to push to lower the rate of warming to give us more time to adapt?
I don’t profess to have the answer, but I’d be interested in the debate. I worry that this discussion doesn’t have enough input from real experts in this space.
If climate change is intractable, then what’s the next step? Should we be looking at geoengineering, adaptation, resilience? Assuming that climate change is intractable, then here are some other rough ideas of things that could help global welfare:
Early warning systems for floods in areas with anticipated rising sea levels
Research onto how societies should manage heat stress
Research and development of more resilient infrastructure, e.g. energy, food, and water—even if it’s just theoretically a question of pricing water etc. accurately, I’m not confident that in practice globally we’re doing that very well at the moment
4,5 - Interesting to read, I don’t profess to be an expert so would appreciate learning from other perspectives.
6,7,8- I wonder whether it’s possible that some modelling is overconfident on how resilient societies will be to climate change, as we’re densely networked and there might be lots of unanticipated secondary effects, such as mosquitoes affecting another billion people. The latest UK adaptation report acknowledges biodiversity as one of the several areas urgently needing further research.
I’ll focus on point 2 because I think it is the most important. I don’t see the argument for it being true that for the vast majority of people, working on climate change promises more leverage on the problem of nuclear war, than does working directly on nuclear war. Nuclear war is easier to make progress on, more neglected and more important than climate change.