John Halstead - Research Fellow at the Forethought Foundation. Formerly Head of Applied Research at Founders Pledge and researcher at Centre for Effective Altruism DPhil in political philosophy from Oxford
John G. Halstead
Karma: 8,558
Given that people are sharing evidence on Torres, I thought I would chime in. I agree it would have been better for the OP to share with Zoe before posting, but I also think working with Torres is a mistake.
My relationship with Torres started after I criticised something he wrote about Steven Pinker on Facebook—my critique was about 3 sentences. My critique was supported by others in the community, including Will MacAskill. I think this was the start of Torres becoming disenchanted with EA.
From this point on, he published several now infamous pieces suggesting that I and others in EA support white supremacy. He also sent me numerous messages on Facebook after I had stopped responding. In this Facebook post, Torres inexplicably namechecks me while he is accusing some people of being rapists/paedophiles (their names are redacted)
My whole experience with Torres has been surreal—for one small piece of criticism, he went after me for years. I know he has done the same to others: some people he has gone after have needed counselling, and I think people should take that into account when they interact with Torres.
For people who are confused that Torres, who wrote a book defending the FHI-house view of x-risk in 2017 and endorsed that view until his review of Pinker in 2019, now thinks EA is so bad, it seems to be because he thinks he faced some rejection by the community.
I would say that for all of the ‘non-EA’ reviewers, the review was very extensive, and this was also true of some of the EA reviewers (because they were more pushed for time). The non-EA expert reviewers were also compensated for their review in order to incentivise them to review in depth.
It is true that I ultimately decided whether or not to publish, so this makes it different to peer review. Someone mentioned to me that some people mean by ‘peer review’ that the reviewers have to agree for publication to be ok, but this wasn’t the case for this report. Though it was reviewed experts, ultimately I decided whether or not to publish in its final state.
Thanks for this. Someone else raised some issues with the moist greenhouse bit, and I need to revise. I still think the Ord estimate is too high, but I think the discussion in the report could be crisper. I’ll revert back once I’ve made changes
I’d say the depth of review was similar to peer review yes, though it is true to say that publication was not conditional on the peer reviewers okaying what I had written. As mentioned, the methodology was reviewed, yes. So, this is my view, having taken on significant expert input.
A natural question is whether my report should be given less weight eg than a peer reviewed paper in a prominent journal. I think as a rule, a good approach is to try start by getting a sense of what the weight of the literature says, and then exploring the substantive arguments made. For the usual reasons, we should expect any randomly selected paper to be false. Papers that make claims far outside the consensus position that get published in prominent journals are especially likely to be false. There is also scope for certain groups of scientists to review one another’s papers such that bad literatures can snowball.
This isn’t to say that any random person writing about climate change will be better than a random peer reviewed paper. But I think there are reasons to put more weight on the views of someone who has good epistemics (not saying this is true of me, but one might think it is true of some EA researchers) and also be actually talking about the thing we are interested in—i.e. the longtermist import of climate change. Most papers just aren’t focusing on that, but will use similar terminology. e.g. there is a paper by Xu and Ramanathan which says that climate change is an existential risk but uses that term in a completely different way to EAs.
I will give some examples of the flaws of the traditional peer review process as applied to some papers on the catastrophic side of things.
A paper that is often brought up in climate catastrophe discussions is Steffen et al (2018) - the ‘Hothouse Earth’ paper. That paper has now been cited more than 2,000 times. For reasons I discuss in the report, I think it is both surprising that the paper was published. The IPCC also disagrees with it.
2. The Kemp et al 2022 PNAS paper (also written by many planetary boundaries people) was peer reviewed, but also contains several errors.
For instance, it says “Yet, there remain reasons for caution. For instance, there is significant uncertainty over key variables such as energy demand and economic growth. Plausibly higher economic growth rates could make RCP8.5 35% more likely (27).”
The cite here in note (27) is to Christensen et al (2018), which actually says “Our results indicate that there is a greater than 35% probability that emissions concentrations will exceed those assumed in RCP8.5.” i.e. their finding is about the percentage point chance of RCP8.5, not about an increase in the relative risk of RCP8.5.
Another example: “While an ECS below 1.5 °C was essentially ruled out, there remains an 18% probability that ECS could be greater than 4.5 °C (14).”
The cite here is to the entire WG1 IPCC report (not that useful for checking but that aside...) The latest IPCC report says “a best estimate of equilibrium climate sensitivity of 3°C, with a very likely range of 2°C to 5°C. The likely range [is] 2.5°C to 4°C” The IPCC says “Throughout the WGI report and unless stated otherwise, uncertainty is quantified using 90% uncertainty intervals. The 90% uncertainty interval, reported in square brackets [x to y], is estimated to have a 90% likelihood of covering the value that is being estimated. The range encompasses the median value, and there is an estimated 10% combined likelihood of the value being below the lower end of the range (x) and above its upper end (y). Often, the distribution will be considered symmetric about the corresponding best estimate, but this is not always the case. In this Report, an assessed 90% uncertainty interval is referred to as a ‘very likely range’. Similarly, an assessed 66% uncertainty interval is referred to as a ‘likely range’.
So, the 66% CI is 2.5ºC to 4ºC and the 90% CI is 2ºC-5ºC. If this is symmetric, then this means there is a 17% chance of >4ºC, and a 5% chance of >5ºC. It’s unclear whether the distribution is symmetric or not—the IPCC does not say—but if it is then the ’18% chance of >4.5ºC’ claim in climate endgame is wrong. So, a key claim in that paper - about the main variable of interest in climate science—cannot be inferred from the given reference.
3. Jehn et al have published two papers cited in Kemp et al (2022), one of which says that “More likely higher end warming scenarios of 3 °C and above, despite potential catastrophic impacts, are severely neglected.” This is just not true, but nevertheless made it through peer review. Almost every single climate impact study reports the impact of 4.4ºC. There is barely a single chart in the entire IPCC impacts report that does not report that. We can perhaps quibble over what ‘severely neglected’ means, but it doesn’t mean ‘shown in every single chart in the IPCC climate impacts book’. It is surprising that this got through peer review.
**
As I have said, these are just single studies. I am consistently impressed by how good the IPCC is at reporting the median view in the literature, given how politicised the whole process must be.
**
I also do not think there is any tendency to downplay risks in the climate science literature. If you look at studies on publication bias in climate science, they find that effect sizes in abstracts in climate change papers have a tendency to be significantly inflated relative to the main text. This is especially pronounced in high impact journals. I have also found this from personal experience. Overall, I think in some cases the risks are overstated, in some they are understated, but there is no systematic pattern.
Probably the best way to examine whether my substantive conclusions are wrong would be to raise some substantive criticisms/carry out a redteam—I would welcome this. I emphasise that if my arguments are correct, then the scale of biorisk is numerous orders of magnitude larger than climate change.
One issue I have with these arguments for pluralism and for sometimes obeying something like common sense morality for its own sake and independent of utilitarian justification is that common sense morality is crazy/impossible to follow in almost all normal decision situations if you think it’s implications through properly.
One argument for this is MacAskill’s argument that deontology required paralysis. Every time you leave the house, you foreseeably cause someone to die by changing the flow of traffic. Cars also contribute to air pollution which foreseeably kills people, suggesting that emitting any amount of pollution is impermissible. This violates nonconsequentialist side-constraints. I don’t understand how you can give some weight to this type of view.
This is not the point that we should follow utilitarian morality when the stakes are high from a utilitarian point of view.
This is a very thoughtful critique. What do you make of the argument that The Precipice and WWOTF work well together as a partnership that target different markets and could be introduced at different stages as people get into EA?
“Refugees: ~216 million climate refugees by 2050 (World Bank Groundswell Report) caused by droughts and desertification, sea-level rise, coastal flooding, heat stress, land loss, and disruptions to natural rainfall patterns”
The groundswell report is about voluntary internal migration, so it is not about refugees, who are typically defined as involuntarily displaced people crossing national borders.
I thought that was what was meant by AGI? I agree that the operationalisation doesn’t state that explicitly, but I thought it was implied. Do you think I should change in the report?
Which impacts do you think I have missed? Can you explain why the perspective you take would render any of my substantive conclusions false?
I’m not sure what you’re talking about with self-citation. When do I cite myself?
Another way to look at it is to think about the impacts including in climate-economy models. Takakura et al (2019), which is one of the more comprehensive, includes:
Fluvial flooding
Coastal inundation
Agriculture
Undernourishment
Heat-related excess mortality
Cooling/heating demand
Occupational-health costs
Hydroelectric generation capacity
Thermal power generation capacity
I discuss all of those except cooling/heating demand and hydro/thermal generation capacity, as they seem like small factors relative to climate risk. In addition to that, I discuss tipping points, runaway greenhouse effects, crime, civil and interstate conflict, ecosystem collapse.
Oh sorry, I thought you meant ‘did they leave negative comments about these things’. Lots of people looked at the overall report and were free to point out things I missed.
I still don’t really understand why you have such an issue with the methodology. I took my methodology to be—pick out all of the things in the climate literature that are relevant to the longtermist import of climate change, review the scientific literature on those things, and then arrive at my own view, send it to reviewers, make some revisions, iterate.
I think it might help to make this discussion more concrete if you gave an example of what you mean by a cascading risk. It’s hard to defend the arguments in the report when I’m not sure what you are saying I have missed in my analysis. I talk about risks to the food system, and the spillover effects that might come from that (eg conflict), I talk about purported effects on crime, I talk about drought, I talk about tipping points etc. What is the casual story you have in mind?
The substantive discussion is the outline all of the various impacts that I have discussed and summarising the literature on economic costs, which tends to find costs of 4ºC are on the order of 5% of GDP. Unless something is radically missing from these analyses, I’m not sure how climate change could make a large difference to the chance of recovery from collapse.
How about a scenario where a multitude of factors eg climate related damages, civil conflict, interstate conflict, bioweaponary, natural disasters and economic collapse all work in concert with each other?
I discuss the potential impacts of climate damages, civil conflict, interstate conflict and the economic impact of climate change at considerable length. Even if these all work in concert with each other, my substantive conclusion is unaffected. I also explicitly discuss the possibility that climate change will cause the use of bioweapons in the report.
What I am trying to suggest is that by shutting down the possibilities to bio and nuclear war, you reduce the role that climate change could play in bringing about collapse.
I don’t shut down the possibility, I argue against it at considerable length.
However, I really think a lot of your key ideas could so with better citation/definition. I also think that ignoring a lot of these concepts which are common in the literature, and then putting the burden on me to hash out the arguments in a comment on my weekend, rather than actually addressing these concepts, even if you were to reject them, in your 400 page report, is a little odd.
You have made a series of conceptual criticisms of the report. I have said that my conceptual approach is exhaustive, which is true, but you seem to think this is unsatisfactory. I don’t think it is unreasonable for you to explain to me what you think I have missed.
A direct climate impact is an impact of climate change for which the proximal cause of the damage is not human-on-human interaction. An example would be something like heat stress deaths or crop failures from drought. An indirect climate impact is an impact for which the proximate cause is human-on-human damage, but for which the ultimate cause is climate change. An example would be crime, conflict or undermines institutions.
Saying something is the most obvious isn’t evidence or a justification. Your report is 400 pages long, I am pretty sure you have space to justify this core part of your methodological approach. Also, just because one thing is the “most” obvious doesn’t mean others aren’t worthy of consideration.
I think appeals to common sense and what is obvious are often permissible in arguments. Which indirect effect do you think is more important?
I was expecting the discussion to be more like ‘hear is why you are wrong about emissions/climate sensitivity/runaway greenhouse/impacts on the food system/impacts on conflict/...’
No I didn’t get any of that. I don’t want to put words in their mouths, but Peter overall seemed very positive. I’m less sure what Goodwin and James thought, but they didn’t say anything massively negative, though perhaps they thought it
I think this is a good place to have discussions about claims in specific sections (rather than the whole report) if people would like
Hi Noah,
(Just noting that I’m not ignoring your comments about methane clathrates, but I don’t think you were asking for a response there, but were instead just highlighting some issues for you to look into? Correct me if I’m wrong)
Yes I note that there is deep uncertainty about sea level rise once warming passes 3ºC and that sea level rise might be much higher than estimated. I discuss the impacts this might have in the sea level rise section and the economic costs section
I agree that many specific tipping points haven’t made their way into IPCC models
The model should be shared now.
Yes that is correct re my assessment of the other existential risks. I’m taking a view similar to Toby Ord and I suppose the rest of the EA community about where the main risks are. Of course, my main goal in the report is not to make this substantive case; I largely take it as given.
I don’t really see how viewing climate change as a cascading risk would change the overall risk assessment. If you argue that climate change is a large cascading risk then you would have to think that climate would play an important role in starting the cascade from collapse to extinction. I don’t see how it could do that and explain why at length in the report. Can you lay out a concrete scenario that sketches this cascading risk worry that isn’t already discussed in the report?
The report does suggest that climate change would make civilisational recovery harder but for plausible levels of warming, it would not be a large barrier to recovery and this should be clear from the substantive discussion in the report
Moreover, you suggest that “there is some chance of civilisational collapse due to nuclear war or engineered pandemics,” essentially suggesting other causes of civilisational collapse that are less direct, and therefore could be made more likely due to climate change, are negligable. This assumption should be stated and evidenced, and yet you seem to include no sources on this.
The whole report is about whether climate change could lead to civilisational collapse or something close to it. What other mechanisms do you have in mind that are not already discussed in the report?
The influence of climate change on great power war or war more generally seems like the most obvious indirect risk of climate change that could make a substantial difference to the scale of climate change. It is often argued that climate change is a threat multiplier for conflict risk. I discuss the literature on this at length. What other indirect risks do you think might be comparably important?
It does seem that you think that viewing climate change as a cascading risk would make a large difference to my conclusion. I don’t understand what you think this cascading risk actually is that is not already discussed in the report.
I’m not sure which comment you are referring to? I argued that the direct/indirect approach is conceptually exhaustive, which is trivially true.
The reviewers for each section were as follows. Josh Horton reviewed a section on solar geoengineering which I am still in the process of revising for a later version.
Peter Watson, Goodwin and James Ozden provided comments on various sections in the report.
Without wanting to pre-empt the reviewer comments if I am allowed to provide them, there was agreement with what I had written and I accepted the vast majority of proposed revisions. I think the main disagreement was that Keith Wiebe disagreed with some of my claims about extreme warming and agriculture. I think maybe Danny Bressler moderately disagreed with my assessment of Burke et al (2015), but not completely sure.
I will ask the experts if I can share their feedback. I did ask a couple of them to do this but after a long review process they didn’t respond so I decided not to ask the other experts if I could share theirs as I thought it would be weird to have comments on some parts but not others. Maintaining interest in the process from experts can be difficult because they sank a lot of time into reviewing the report and have other things to do so there is a risk of over-asking and them not wanting to engage any more.
the forum did offer the chance of having agree/disagree on the post, I just forgot to respond. I think it is a beta feature but happy for it to be used on this post
These are good points, I will amend
hi david, this was from before he was banned from the forum but after his beef with me started—this was while he was doing all the white supremacy articles about me, beckstead and others. he had a long-standing dispute with the people mentioned, and independently at the time he was especially annoyed at me for criticising him. I think that is what led him to namecheck me in his allegation.
I hadn’t heard of one of the people he was accusing at the time that he wrote the facebook post. I have no idea whether or not the allegations are true, I just don’t understand why he involved me in them.