Thanks for this work and the nuanced write up.
I guess one way to make these findings potentially stronger / more informative would be to see whether effect size varies as one would expect with familiarity of Just Stop Oil (i.e. more change with more exposure)?
I think it's important to see the nuance of the disagreement here.1. My critique is of what strikes me as overconfident and overconfidently stated reasoning on what seems a critical point in the overall prioritization of climate -- as Haydn writes, few sophisticated people buy the "climate is a direct extinction risk", so while this is a good hook it is not where the steelmanned case for climate concern is and, whatever one assumes the exact amount of risk to be, indirect existential risk plausibly is the majority of badness from climate from a longtermist lens.2. My critique does not imply and I have never said that we should work on climate change to address biorisk. The reasoning of the article can be poor and this can be critiqued while the conclusion might still be roughly right. 3. That said, work on existential risk factor is quite under-developed methodologically so I would not update much from what has been said on that so far, I think this is what footnote 25 also shows, the mental model on indirect risks is not very useful / imposes a particular simplified problem structure which might be importantly wrong.4. As you know, I broadly agree with you that a lot of the climate impacts literature is overly alarmist, but I still think you seem too confident on indirect risks, there are many ways in which climate could be quite bad as a risk factor, e.g. perceived climate injustice could matter for bio-terrorism, or there could be geopolitical destabilization and knock-on effects in relevant regions such as South Asia.
Strongly agree with Haydn here on the critique. Indeed, focusing primarily on direct risks and ignoring the indirect risks or, worse, making a claim about the size of the indirect risks that has no basis in anything but stating it confidently really seems unfortunate, as it feels like a strawman.Justification for low indirect risk from the article:"That said, we still think this risk is relatively low. If climate change poses something like a 1 in 10,000 risk of extinction by itself, our guess is that its contribution to other existential risks is at most an order of magnitude higher — so something like 1 in 1,000.25"And then footnote 25"How should you think about indirect risk factors? One heuristic is that it’s more important to work on indirect risk factors when they seem to be worsening many more direct problems at once and in different ways. By analogy, imagine you’re in a company and many of your revenue streams are failing for seemingly different reasons. Could it be your company culture making things less likely to work smoothly? It might be most efficient to address that rather than the many different revenue problems, even though it’s upstream and therefore less direct.
But this doesn’t seem to be the case with climate change and direct extinction risks — there aren’t many different ways for humanity to go extinct, at least as far as we can tell. So it’s less important to reduce upstream issues that could be making them worse vs trying to fix them directly. This means that if you’re chiefly worried about how climate change might increase the chance of a catastrophic global pandemic, it seems sensible to focus directly on how we prevent catastrophic global pandemics, or perhaps the intersection of the two issues, vs focusing primarily on climate change.↩"Problems with this:1. There is no basis on which to infer to a magnitude relationship between direct existential risk of climate and indirect existential risk of climate. It is totally possible for climate to have a very low probability of being a direct existential risk while still being a significant indirect existential risk factor, as far as I can see there is no substantive argument for the two to not vary by more than 1 order of magnitude.2. There is equally no basis, as far as I can tell, for working directly on the problem. What the heuristic impliclty assumes to be true, without justification as far as I can tell, is that problems are solvable by direct work. This is not necessarily true, for example it is perfectly conceivable that AI safety outcomes depend entirely on the state of geopolitics in 2035 and that direct domain-specific work has no effect at all. This is an extreme example, but there seems to be no argument why we should assume a specific and definitely higher share of solvability from direct work.
I strongly share John's intuition that this is primarily an artefact of talking about desired temperature targets in the IPPC report rather than a change in the foci of the research that the IPCC reports on.Would it be possible to test this by denoting which share of the 0-2 degree mentions are surrounded by words like "Paris Agreement", "policy targets", "ideally", "well below" etc. (i.e. words that typically co-occur with the statement of the ambition of the Paris Agreement) . Or, alternatively, by focusing on the climate science & impact sections of the IPCC reports?
FWIW, your calculation seems still optimistic to me, still, e.g. assuming quite a high elasticity (cost of coal is not such an important part of the cost of producing electricity with coal) and, if I understand your reasoning correctly, a fairly high chance of additionality (by default, coal is in structural decline globally).
Thanks for this!FWIW, Founders Pledge's climate work is explicitly focused on supporting solutions that work in the worst worlds (minimizing expected climate damage) and we're thinking a lot about many of those issues both from a solution angle and a cause prioritization angle (I think the existential risk factors you allude to are by far the most important reasons to care about climate from a longtermist lens).That being said, you are making a lot of very strong claims based on fairly limited evidence and it would require significantly more work to get credible estimates on the various risk pathways and to justify confident statements about large effect sizes. It also seems that the sources do not fully support the claims, e.g. I clicked on the link to justify the estimate for IPCC underestimating climate impacts which led me to a study which is 10 years old, i.e. essentially carries 0 information for the current state on that question. So I'd be careful to jump to quite extreme (in the sense of confident) conclusions. Given the vast uncertainties and the heterogeneity of climate-relevant science one can justify almost any conclusion from cherry-picking a subset from the peer-reviewed literature, so it's really crucial to consider review papers and a wide range of estimates before making bold claims.
It's a btit hard to comment in detail given the research process is not very clear. One question, though: There's a fair amount of public and private money focused on funding research to decarbonize aviation fuels (including Breakthrough Energy), what is the rationale for thinking that small philanthropic donations to private companies can make a difference?
(Working at Founders Pledge)1. To your question on accounting for deadweight losses etc., it is true that this is not included, rather this is an estimate of marginal changes from donations. But the factors not included in the calculation are not only deadweight losses (and other costs), but also lots of benefits, e.g. economic benefits from technological leadership. This is parallel to GiveWell analyses which only focus on mortality/direct income gains and ignore a lot of other follow-on benefits and costs.2. The air pollution benefits of clean energy advocacy are plausibly in the same ballpark as climate benefits (depends on how severe climate change turns out) and benefits from overcoming energy poverty are also very significant (though hard to causally pin-down given the relationship between energy demand growth and human welfare is bidirectional, I explore this a bit more here).3. One thing that is very different between GiveWell recommendations on global health and FP recommendations on climate is the attitude towards uncertainty -- GiveWell recs have a high uncertainty avoidance whereas CATF and other estimates are meant to be risk-neutral estimates leveraging a fairly indirect theory of change (policy advocacy > policy change > technological change > changed emissions trajectory). So, in that sense the absence of risk-neutral global health recommendations biases the argument in favor of climate.
Much has been already discussed, but I wanted to add some clarifications and additional arguments on some points.
And, more generally, on uncertainty and the relative uncertainty of different interventions:
1. None of our grants was driven by the cost-effectiveness analysis that Matthew cites as justifying or being the basis of our grants. This is also clear from the report, as we (i) only include it in the background section, (ii) have plenty of qualifiers about this explaining a logic / theory of change (indeed, it is first referenced in a section on theory of change, irrespective of a specific grant, p. 86), (iii) the cost-effectiveness example is a historical case from 2018 and we have only started making grants from the Climate Fund in 2020, and (iv) the rationale for our grants, including for CATF, is also explained in the report. From the two grants we made to CATF from the Climate Fund, the first was to capitalize on the Biden moment across the entire spectrum of technologies CATF focuses on (CCS, but also hydrogen, advanced nuclear, geothermal, methane etc.) and the second was for globalization of CATF activity and organizational investment.
As I emphasize whenever I talk or write about it, these cost-effectiveness analyses are clearly wrong and they (i) primarily serve as explaining a logic and important stylized facts (such as the stark difference between local and global effects) and (ii) they are not the only and certainly not the primary piece of analysis when making a grant – a lot of other kinds of evidence and considerations go into grantmaking. Indeed, because these cost-effectiveness analyses are ultimately so arbitrary at FP we now focus on understanding the source of impact differentials – analyzing the climate philanthropy and action landscape – rather than getting more precise with specific analyses, as impact differentials are more important to make the right prioritization decisions (again, more in the report and also the reason we do not publish lots of BOTEC cost-effectiveness analyses these days; lots more on this in the report).
2. All money that FP and Giving Green direct to CATF are philanthropic dollars, c(3), not c(4), so they are not part of the lobbying budgets and expenses discussed above.
ON CCS3. Why philanthropic CCS support is valuable despite a powerful fossil lobby: 2 gives part of the answer why CATF’s CCS work is valuable despite a powerful well-funded fossil fuel industry. Insofar as one accepts that CCS support is a bet we should take (Kim & Dan have elaborated on this so I won’t add detail here) from a climate angle, it does matter that there is one credible environmental org aligned with Democrats (there are also Republican climate orgs, like ClearPath) that pushes for it, it can make the difference between this being entirely dismissed as fossil fuel or Manchin demand to being an option that has support from clearly climate-motivated actors. CATF also has different incentives than fossil fuel lobbyists in shaping the legislation, so this is another positive aspect of CCS-philanthropic-funding that is not just lobbying expenditure.4. Learning curves of CCS: The learning curve that we assume for CCS is 10% (based on an 8% conservative rate in a report by Global CCS institute, see report for details), compared to a learning curve of 25% or so one would use for solar. So we do indeed adjust for the fact that for a technology like CCS, large-scale not easily mass manufacturable infrastructure, the learning rate is much lower than for more modular technologies. 45Q supports a whole lot of different CCS applications – coal, gas, carbon removal etc – and it doesn’t seem particularly plausible that none of them will have significant technological progress because of a more ambitious 45Q (as discussed above, we do not specifically focus on coal CCS, indeed I am personally more optimistic on other forms of CCS). Within the CCS bucket, there are technologies with high modularity where indeed we should expect higher learning curves than for coal CCS.
ON UNCERTAINTY5. Should we care about the degree of uncertainty? There’s a bit of a semantic stop-sign dynamic in some of the discussions of uncertainty here, suggesting that higher uncertainty would be a reason to be less enthusiastic, more skeptical etc, to use it as a disqualifier. But from an impact-maximizing perspective there isn’t really a case per se for treating uncertainty as an impact-reducing feature. There are, of course, interactions between uncertainties and how they are resolved which we discuss in our report, but these considerations are around the interactions of uncertainties, not the width of uncertainty.I think there is something like intuitive risk aversion going on, where uncertainty leads to skepticism (“maybe it will have zero effect”), whereas the positive case is mentally discounted. But for uncertain technological trajectories as well as ambitious policy changes, the upside can be extremely large.
6. If we care about uncertainty, we need to look at the entire theory of change: As discussed, I don’t think uncertainty should be a qualifier from an impact perspective. But regardless, statements like “innovation advocacy is more uncertain than interventions to deploy mature tech” are partial and, for this reason, possibly quite misleading. We all agree that the innovation process is more uncertain than the deployment of existing technology, but here are a couple of uncertainties that run the other way (which are more uncertain for deploying mature technology) that would need to be taken into account in an overall account of uncertainty of different philanthropic bets:
So whether or not uncertainty of Rewiring America or Evergreen or CATF or Carbon180 is higher is very much an open question. It depends on how the entire sequence of uncertainties from (i) funding additionality (will this be funded anyway?) to (ii) activity additionality (will the funded activity happen anyway, if not by charity X then by someone else?), (iii) probability of policy change, (iv) naive emissions reductions from policy change and (v) emissions reductions adjusted for policy additionality plays out. We cannot just single out (iv) and assign it to the entire intervention and conclude that innovation advocacy is more uncertain than mature tech deployment just because one step in the theory of change of the former is more uncertain.
On 1, agree that this is complicated because of electrification, but even in 2030 80% of electricity are unlikely to translate into more than, say, 40% of energy, given a lot of energy-intensive processes are not easily electrifiable (heavy-duty transport, industry etc.). In any case, these are good goals. But more technology-inclusive peer countries (France, UK) are much more successful.On 2, blue hydrogen (natural gas w CCS) is cheaper than green hydrogen for the next decade or so, so it is good if that is included.I think the current coalition is better than some counterfactuals on climate (continued Grand Coalition, government led by the Greens), but overall still fairly disappointing.