An argument that EA should focus more on climate changepost by Ann Garth · 2020-12-08T02:48:06.251Z · EA · GW · 3 comments
Key Points I. The potential co-benefits of climate change mitigation interventions are undervalued in EA (high confidence) and are likely to be greater than the co-benefits associated with other interventions (low confidence). Climate change mitigation co-benefits Comparing co-benefits II. The harms of climate change are likely undervalued in EA (high confidence). Even if full modeling of those harms is impossible, we know enough to prioritize climate change interventions over interventions in other cause areas (low confidence). How to handle uncertainty Directions for further analysis None 3 comments
Acknowledgements: With thanks to Johannes Ackva, John Halstead, Louis Dixon, Aaron Gertler, and Jonah Goldberg for their thoughts on earlier drafts of this post.
- Most climate change mitigation policies have non-climate-related benefits (often referred to as “co-benefits” in the climate literature). Properly accounting for these co-benefits would increase the effectiveness of climate change interventions.
- There are co-benefits associated with interventions in many other issue areas as well, and it is unclear how the co-benefits of climate change mitigation policies compare to the co-benefits of other interventions. I suspect that the co-benefits of climate change mitigation interventions are larger than the co-benefits of interventions in other issue areas, but I am not at all confident in this, and more work in this area is needed.
- The harms of climate change are undervalued in EA, but the extent to which they are undervalued is unclear. Despite this uncertainty, I believe that we know enough to prioritize climate change interventions over interventions in other cause areas; however, I have low confidence in this belief.
I. The potential co-benefits of climate change mitigation interventions are undervalued in EA (high confidence) and are likely to be greater than the co-benefits associated with other interventions (low confidence).
Climate change mitigation co-benefits
Most of the small-scale interventions which mitigate climate change have non-climate-related “co-benefits.” Coal-fired power plants are a significant source of air pollution and thus premature death, so replacing them with renewable power has health benefits. Increasing access to energy in the developing world (including clean energy) will reduce poverty. Other, more controversial, co-benefit claims include claims that distributing clean cookstoves economically benefits families by reducing fuel costs as well as reducing indoor air pollution and associated disease burden, and that rainforest conservation programs that work with local communities provide direct economic, health, and educational benefits and/or access to traditional ways of life, which may improve well-being.
Most of the larger-scale interventions to mitigate climate change also appear to have co-benefits. For example, funding large-scale renewable energy development, improving the grid to make energy transfer more efficient, helping developing nations access clean energy technology, and advising them on how to best implement that technology are all interventions which would reduce energy costs and/or increase access to energy. Founders Pledge identified the four highest-value areas of climate change to work on -- carbon capture and storage, nuclear power, low carbon innovation, and forestry -- and of these, three (nuclear power, low carbon innovation, and forestry) intuitively seem to have associated co-benefits. Similarly, many of the climate change mitigation strategies identified by Drawdown (e.g. reduced food waste, abandoned farmland restoration) fall in this category. And the fact that some developing countries are already investing heavily in carbon-free energy, and are capturing economic co-benefits, suggests that some of these solutions are tractable.
More speculative technologies which are currently being developed might also have significant co-benefits. For example, modular nuclear reactors are a nuclear power technology that is significantly safer than traditional nuclear power plants (and thus, presumably, easier for poorer nations to access), and which can be small enough to fit in the bed of a truck. This portability means that they could reduce global poverty by increasing access to energy in rural areas where conventional fossil fuel plants are difficult or cost-ineffective to build. Their portability would also increase resilience to black swan threats such as a thermonuclear exchange, solar flares, and the like. Another example is Microbial Fuel Cell technology, which uses microorganisms to produce energy from urine and wastewater, producing clean energy while also helping to sanitize the water. There are dozens of other emerging technologies at various stages of development (see here, here, or here for a sampling) which hold the promise of co-benefits, most in the form of significant cost savings (which could then be translated into poverty reduction, investment in other EA priority issues, etc).
Finally, as a recent forum post [EA · GW] noted, economic growth in developing countries may be more important than randomista interventions, meaning that the potential impact of a climate change mitigation intervention which also had significant economic impacts might be quite large. Of course, it is possible that transitioning away from fossil fuels on a large scale might raise energy costs enough to slow economic growth. However, I suspect this would be unlikely. Given how many different types of renewable energy technologies exist, it seems quite possible to identify low-cost green energy technologies. Furthermore, most of the climate change mitigation technologies listed here are complements to existing energy systems, rather than policy changes to reduce use of fossil fuels, avoiding some of the economic harms of regulation.
It is trivially true that if a given intervention has both direct benefits (e.g. mitigating climate change) and co-benefits, accounting for both direct benefits and co-benefits makes the intervention more cost-effective -- potentially enough so that the intervention would be competitive with other EA priority interventions, even if it was not prior to the consideration of co-benefits. In other words, accounting for co-benefits might be enough to tip the scales in favor of one intervention over another.
If even just a few of the co-benefits of climate change mitigation interventions turn out to be significantly positive, it seems highly likely that that would be sufficient to make many climate change interventions a higher priority than they currently are. However, interventions in other cause areas have co-benefits as well. For example, malaria nets reduce disease, presumably allowing more students to attend school and thus increasing their and their communities’ human capital. Similar cases could be made about other global health and development interventions, and indeed about probably every intervention in any cause area.
It is unclear which interventions have higher co-benefits. However, this seems like it might significantly affect comparisons between interventions and cause areas. It’s possible that a full accounting of climate change mitigation co-benefits would make climate mitigation interventions competitive with GiveWell’s top recommended charities. (Indeed, some EAs believe that from a near-termist perspective, the air pollution reduction co-benefits of climate mitigation interventions are similarly important as mitigating climate change. The benefits of reducing air pollution consistent with a clean energy transition to stay below 2°C have been estimated at $700 billion per year in the US alone, which is one reason why Founders Pledge has prioritized climate change interventions.)
Of course, co-benefits only affect the importance of an issue and don’t affect tractability or neglectedness. Therefore, they may not affect marginal cost-effectiveness. However, in instances where two interventions or two cause areas appear to have quite similar cost-effectiveness, accounting for co-benefits might be a tipping point for prioritizing one intervention over another. Regrettably, however, there has been relatively little work in EA addressing climate change mitigation co-benefits. One notable exception is Founders Pledge, which heavily emphasizes the “triple challenge” of climate change, air pollution, and energy poverty in their climate change mitigation work; Rob Wiblin also discussed this issue recently on Facebook. However, co-benefits are not discussed as often elsewhere in EA, and when discussed they are not usually quantified.
There has been even less work comparing cause areas with co-benefits accounted for; furthermore, some EA organizations do not seem to have seriously considered climate change as a cause area for quite some time (although that might change as climate change becomes more of a focus for EA). This means that, for example, recent direct comparisons between climate change interventions and global health/development interventions have primarily happened on the forum (example [EA · GW]), and have not addressed co-benefits. Therefore, modeling the co-benefits of climate mitigation and of interventions in other cause areas seems to be an important avenue of research.
Modeling the co-benefits of climate change mitigation policies is hard: there is significant uncertainty about the impact of many emerging technologies, and the co-benefits depend on which interventions are used, which is also hard to model. However, it seems intuitively plausible that some of the economic co-benefits of climate change mitigation technologies could be modeled.
One co-benefits scenario is: technology X costs families $B and produces N amount of energy; prior to technology X, producing N amount of energy cost families $A; thus, $A-B is the amount of money families save due to technology X. In this scenario, existing EA modeling about the impact of poverty-reduction efforts like direct cash transfers can give a sense of the impact of $A-B savings for families.
In other cases, renewable energy technology would provide energy in regions where there was not previously access. An extremely cursory search of the existing literature on energy poverty (e.g. here) suggests that work attempting to model the benefits of expanded energy access may have already been done, and it is possible that those models could be used by EA researchers.
I have no background in other main EA issue areas, but I wonder if better modeling of co-benefits could be conducted there as well.
Even in instances where full co-benefits modeling is impossible, however, I still believe that more transparency about co-benefits -- such as listing some known co-benefits, even if modeling their exact effect size is impossible -- would be useful. If all/most of the known co-impacts are likely to be positive (co-benefits, as opposed to “co-harms”), we can assume that the intervention is likely to have a more positive impact than modeling of direct impacts shows, even if we don’t know the precise magnitude of this difference.
II. The harms of climate change are likely undervalued in EA (high confidence). Even if full modeling of those harms is impossible, we know enough to prioritize climate change interventions over interventions in other cause areas (low confidence).
Climate change has a number of immediate and direct impacts with immediate and direct harms, but it also has a laundry list of other indirect impacts. Current estimates of the social cost of carbon do not fully capture “Adjustment costs (short-term costs of adaptation), Non-market damages (biodiversity loss, cultural losses, etc.), Tipping points in the climate system (catastrophic climate events, hysteresis etc.), High inertia effects of CO2 (ocean acidification, sea level rise), General equilibrium effects (spillover, trade, etc.), Macro-scale adaptation (long-term restructuring of economy), Political instability and violent conflicts, Large migration flows, and More extreme weather and natural disasters.” Danny Bressler has done extensive work on the issues with the social cost of carbon metric, which is quite flawed. For example, though the social cost of carbon makes some account of the economic cost of premature mortality, it does not seem to include non-economic harms. But loss of lung function due to air pollution has harms beyond lowered productivity, as does being made a refugee by a climate-induced natural disaster, and so on. One of the crucial innovations of EA was the use of DALYs/QALYs to capture such non-economic harms, but conventional climate change models do not seem to have made this leap.
The tail risks of climate change are also quite high -- as this post [EA · GW] notes, there is a significant risk of warming over 4C, the impact of which is “undervalue[d]” by current research. Indeed, Rob Wiblin states that “the probability of getting [even] five or six degree warming just doesn’t seem that low.”
Furthermore, climate change is also a risk factor for other harms (in line with the concept of “existential risk factors” in The Precipice). Many of the harms of climate change, such as natural disasters, resource scarcity, and migration, are seriously destabilizing to the global political order.
Finally, I want to emphasize the unique risk of climate change tipping points. Tipping points are reached when climate change causes changes to the planet’s system which then lock in a vicious cycle of increasing warming. For example, as warmer temperatures melt permafrost, the permafrost releases methane, a greenhouse gas which is currently trapped in permafrost; the now-released methane further increases warming, and so on (examples of other tipping points can be found here). While there is uncertainty about the threshold for these tipping points being triggered, and exactly how strong the effects will be, there is little to no uncertainty about the underlying mechanisms. EA priority cause areas are priorities because they all have horrible harms or potential harms, but climate change is somewhat unique in that its harms are horrible and have time-limited solutions; the growth rate of the harms is larger, and the longer we wait to solve them the less we will be able to do.
The uncertainty, then, is not about whether current models undervalue the impacts of climate change -- they clearly do -- but rather about how much they undervalue those impacts. And I believe better modeling is possible to help answer this question. There has already been research evaluating the economic impacts of climate change/climate change mitigation policies on poverty, migration, disease, air pollution, and severe weather. More in-depth modeling including these as many of these factors as possible would provide a better sense of how much worse climate change is than current models predict.
Finally, the (admittedly limited) work in this area suggests that climate change interventions are quite cost-effective: Danny Bressler created a model [EA · GW] using a more comprehensive estimate of the cost of carbon than standard estimates, and his model found that climate change interventions were more cost-effective than GiveWell’s top global health interventions.
How to handle uncertainty
Even with better modeling, of course, many of the impacts of climate change are impossible to fully account for, and we will never be able to predict the full magnitude of its impacts. Despite this uncertainty, however, I believe we are undervaluing the impact of climate change relative to interventions in other cause areas (though I have low confidence in this belief).
One recent forum model [EA · GW] found that climate change interventions are roughly on par with global health & global poverty interventions (within an order of magnitude). However, there is more uncertainty about the magnitude of the impacts of climate change -- which means more potential downside. If even just a few of the negative impacts of climate change turn out to be serious, it seems highly likely that that would be sufficient to make many climate change interventions a higher priority than some other current priority interventions.
Of course, it is possible that the uncertainty around climate change will go in the opposite direction -- that decarbonization is already baked into our sociopolitical systems, or that climate sensitivity is lower than we expect. In those cases, climate interventions might turn out to be much less impactful than interventions in other cause areas.
However, my read of the current evidence suggests that most of the uncertainty related to climate change impacts is downside risk. Almost all of the potential impacts of climate change which have not been fully modeled (listed above) seem to be negative -- and while we may not be able to determine exactly which of the potential severe harms of climate change will occur, that doesn’t particularly matter if we are reasonably confident that at least some of them will. Furthermore, it seems likely that some of these impacts will occur, or are already doing so. The New York Times and ProPublica recently modeled the migration impacts of climate change, which they suggest will be significant. Other recent work suggests that the economic impacts of severe weather which can be attributed to climate change are higher than previous models have found. Many of the tipping point mechanisms discussed in the linked article are already beginning to happen -- for example, the permafrost is thawing and Arctic sea ice levels are declining -- and there is serious concern that others are closer than expected; a group of prominent scientists recently wrote that, while it’s not too late to prevent tipping points, “[w]e think that several cryosphere tipping points are dangerously close.” And finally, though I can’t point to specific surveys to back this up, my sense from the reading I’ve done on this topic is that climate scientists tend to think we are underweighting (rather than overweighting) the risk.
I worry that we may be downplaying the risks of climate change simply because we can’t fully model those risks. On the basis of the evidence above, I think it is reasonable to prioritize climate change more highly than some of us have.
Directions for further analysis
Because there has been little to no evaluation using a full accounting of either the negative impacts of climate change or the co-benefits of climate mitigation interventions (much less both), it is highly possible that high-priority climate change mitigation interventions have been overlooked under current modeling. When it comes to small-scale interventions, questions have been raised about the efficacy of the two interventions that are best-known in EA, clean cookstoves and forest preservation efforts [EA · GW], but it is unclear how much research has been done to find and evaluate other potential interventions. One possibility is to reevaluate climate mitigation charities which GiveWell considered, but did not contact (e.g. Solar Electric Light Fund); it’s unclear why such groups were not initially contacted by GiveWell, but depending on the co-benefits of their work, they may be more effective than previously realized.
Giving Green is also working on this, though their work is unique because their approach is to “meet climate donors ‘where they are’” [EA · GW] by focusing on popular climate-related giving areas rather than the most effective areas. And of course, Founders Pledge is doing fantastic work in this area -- I donate to their Climate Change Fund monthly!
pp 4 ↩︎
Drawdown’s ranking system has been questioned by EAers [EA(p) · GW(p)]; I include it here primarily as an example of the prevalence of co-benefits, rather than to imply that its issue prioritization is correct. ↩︎
Co-benefits are either not mentioned (such as Open Phil’s shallow investigation into climate change), only briefly mentioned (80,000 Hours briefly mentions air pollution but no other co-benefits), or mentioned but never quantified (such as in this review of Cool Earth). ↩︎
They didn’t release the exact numbers produced by their model (perhaps because its current iteration is focused on the Americas rather than the entire globe), the article about their model states that “rough predictions have emerged about the scale of total global climate migration — they range from 50 million to 300 million people displaced” (for comparison, there are currently 26 million refugees) and also cites work from 2018 that their current model was based on, which finds that “as many as 143 million people would be displaced within their own borders” (relative to 45.7 million people internally displaced worldwide as of 2019). I haven’t analyzed this model myself, so I can’t comment on how confident we should be in its accuracy. However, it does seem to suggest that there is at least a reasonable chance that climate change will significantly increase migration flows. ↩︎
Comments sorted by top scores.