63Joined Aug 2019


Sorted by New


Hi John, thanks for the reply, I wasn't expecting one after this long but I am pleased about the thoroughness and thought you put into it.

  1. Yes, I think it's likely to be catastrophic (not extinction risk). If your assertion that it is only likely to be used when warming is 4C+, your whole argument is moot because that is already past the point of no return. From the base that 4C+ is an unacceptable situation, you would immediately be making the case that SG should be funded and accelerated by your own following argument - that currently policy and research are preventing it from being available until well after it is a viable temporary mitigation strategy (if it on balance positive).

  2. In the framing of the desperation-triggered, unilateral application of SG this would not be in the context of "statecraft". If millions are facing water shortages and famines in India they may have bigger concerns than their relations with allies, and equally could criticise them for not contributing to climate change solutions that are disproportionately affecting them. Scapegoating a unilateral SG actor for adverse weather is possible but they could also point to climate change for those effects and say they are trying to combat them, which no one else is doing. This very much gets into the mire of propaganda and spin and I don't think that a direct consequential chain of "public gets angry, state representing that public puts pressure on SG-using state, SG-using state acquiesces" is bulletproof as an argument. Saudi Arabia has such significant political and economic power that they can support terrorism with the West calling them allies. I think these things are far from obvious. I am not so concerned with unilateralism not for the state "peer pressure" reason you give but because it would be a significant financial burden for any one state. I also see it as a potential risk while you appear to discount it heavily, and use this to support 1b) which then kicks the ball down the road for 50 years.

This bet seems severely against your interests so I will treat it as rhetorical - your payout will only be accessible in 30 years and on those time scales inflation, life expectancy, technological change, X-risks and remembering the bet are all not in your favour. Whereas I have the potential to cash out at any time before that deadline. I don't particularly think that it's likely but I would still have taken that bet because of the conditions - 30 years is a long time.

  1. We are, depending on estimates, a decent way off having AGI, so by your argument there is no point doing AGI safety research now because we don't need it now, if there are any risks / costs associated. I think the same arguments in support of that work in this context, time required and complexity of the issue prompt early investment (and that early-stage research is neglected currently). This isn't about the neglect but the timeline - any research is going to incur costs now and pay off later. AGI safety has clear risks in potentially downplaying the risks of AGI and achieving a Dunning-Kruger effect for the entire human race, accelerating development of an inherently uncontrollable technology. It's a case of balancing the costs with the advantages of starting earlier.

  2. Over 90% of warming since the 1970s has been absorbed by the oceans, this conversation will only be of relevance if we are following higher RCP trajectories where it may be warranted to use controversial techniques, but there is a huge thermal sink that will work to bring the atmosphere back into equilibrium even using SG. This undermines the wait and see approach because it would be valuable to prepare in advance of the requirements.

  3. I didn't realise that this was referenced - if other people have investigated then that's fair enough. I didn't think there was any precedent for creating and dispersing particulate matter into the atmosphere and to keep them there, considering weather and localisation and other challenges. We are not able to replicate a volcanic eruption in terms of getting material airborne was my point, so the relevance of global cooling via volcanism isn't ironclad, direct evidence that SAI would work in the same way, as I expect it would depend on the distribution method and location (altitude etc.). If people who have looked into this see this as feasible with current tech then they will likely have more info than me.

  4. The counterfactual of not implementing SAI isn't a flat line for X-risk, in the absence of mitigating effects the risk of existential events increases over time as a result of the indirect effects of climate change unmitigated by SAI (other things being equal). This would somewhat discount the increasing X-risk of SAI.

  5. On rereading I may be misinterpreting, I thought that you were using moral hazard in the standard economic sense but you may be defining it just after as per previous papers as plan B undermining plan A. I agree in that case that my use of the term doesn't make sense, but I'm more familiar with it's use as increasing exposure to risk because the actor doesn't bear the full consequence of that risk, a sort of generalised externality. In that case CO2 emission is a moral hazard because the systemic risk isn't borne just by that actor, which incentivises overproduction.

Again, thanks for the response, I enjoyed the article and your reply has helped me understand the sources of disagreement a bit better. Some of them seem to be purely opinion-based or miscommunication. I also agreed with a lot of your points, although some not for the same reasons that you have given. I would still like to see small scale research on this topic being done and didn't see that much wrong with doing it. I will have to read more about the moral hazard argument sources you mentioned because they should be more convincing than those I have encountered previously.

Cloud formation was the biggest unknown feedback loop and efforts to model them more accurately has led to the increase in range. The effects only start at unprecedented levels of warming which is why observational data may not fit.

I have tried to lay out in the section on probability, above, that most X-risks are a subset of C-risks as they collapse usually has to happen before an existential event. Most X-risks in their moderate form, such as a nuclear winter lasting a few years, moderate climate change, or a global pandemic seem much more likely to pose a C-risk than an X-risk.

Thanks for the comment! I am aware about the GCRI and GCRs but I don't see the term getting used much and (in both cases) seem to get conflated with X-risks, but I haven't addressed this at all in the piece so I will add an edit.

Thanks for catching the typo. I've been trying to embed the images but it hasn't been working, so I'm contacting support for help.

The analogy I was making was that socially held values are liable to change and (usually) improve over time, and any specific value might not disqualify a future civilisation from being counted as valuable by us today, but at some point in the future there may be sufficient drift to make that claim. This may happen gradually and piecemeal as in the ship of Theseus. The full thought experiment also mentions restoration of rotting parts and asks whether these are also the ship of Theseus, similar to a Renaissance period.

This doesn't seem to be the context in which you were dropping the link, seeing as they all have top-level summary numbers that feed into your boundaries and point 1 doesn't say anything about 2m/year being a lower bound, seeing as you are using it as the upper bound. I would like to see these other estimates of climate mortality as they aren't referenced or seem to feed into point 1.

On a meta-level I think disconnecting sources with the context with which you are referencing them is very unfriendly to the reader as they have to wade through your links to find what you are saying where, but apparently these aren't even sources for adding evidence to your claims. So I am further befuddled by your inclusion of incidental contextualising literature when you haven't included references to substantiate your claims. I also think your edit comes across as quite uncharitable to readers (and self-defeating) if you don't think you can change people's minds.

I think when considering your estimates for 1. it is important to consider the boundaries given by those sources and to contextualise them.

The WHO is only looking at disease burden but even there they are expecting 250k to 2050 (not even looking to 2100) and they estimate that CC will exacerbate malnutrition by 3% of current values - this seems extremely conservative. They don't seem to include the range increases for most other insect-transmitted diseases, just malaria, even within the extremely limited subset of causes they consider.

Impactlab's "big data approach" - they don't give their assumptions, parameters, or considerations - I think this should largely be discounted as a result. It seems to be based on historic and within-trend correlation data, not accounting for risk of any higher-level causes of mortality such as international conflicts, political destabilisation, famine, ecological collapse, climate migration, infrastructure damage etc. that will have an impact and I am guessing aren't accounted for in their correlational databank.

Danny Bressler is only looking at extrapolating inter-personal conflicts. It doesn't include famines, pandemics, increased disease burden, ecosystem collapse, great nation conflicts, etc. etc. etc. that are very likely to be much, much worse than the trends considered in his model. As such his 74 million estimate should be considered an extremely conservative lower bound to the estimated value. He is also showing a significant upwards trend per-year, so the burden should be considered to exacerbate over time.

Overall this seems to cast doubt on 1, 4 and 5. For 2. I have also critiqued John Halstead's work in a previous post, and the Ozy Brennan post is refuting CC as an extinction risk, not as a global catastrophic risk as you use it. He is saying nothing about the chances of >10% likelihood of >10% population decrease. These combined should cause pause for thought when making statement 7.

Hi John, thank you for this piece. I know it's been a long time since you posted this but I wanted to respond to some of your thoughts.

"In my view, solar geoengineering is only likely to be used once warming is quite extreme, roughly exceeding around 4 degrees" - +4C is already endgame and catastrophic in my opinion. Considering that most of the heat is being absorbed by oceans leading to acidification, we'll already be seeing significant sequestration losses as marine animals are unable to build calcium carbonate shells.

"This suggests that for solar geoengineering to be feasible, all major global powers would have to agree on the weather, a highly chaotic system." - individual actors may resort to solar geoengineering without worldwide consensus, especially if countries that aren't suffering from climate change are actively blocking mitigation attempts while still polluting. Understanding the possible ramifications before people begin to experiment through desperation is surely a good thing (e.g. India, China, Saudi Arabia, Brazil).

"We have had about 1 degree of warming thus far and, according to an IMF report, a further 1 degree of warming would be economiclly positive for many regions, especially Canada, Russia and Eastern Europe, and even potentially China (IMF report page 15)." - I think this is sketchy at best. The caveat footer 9 on page 14 should indicate how limited their conclusion is, not counting weather effects, migration, ecological effects, etc.

"Russia is a crucial factor here: global warming seems likely to bring numerous economic benefits for Russia, freeing up the Russian Arctic for exploration and thawing potential farmland." - the US and Canada have been far more disruptive for climate change global agreements, the Paris agreement was largely stymied by Republican Congress. Permafrost thawing doesn't free up usable farmland in significant amounts, these are still primarily extremely low viability / low human density forestland in Siberia. In fact, Russia is set to lose out significantly from permafrost thaw.

"Solar geoengineering research has clear risks and, given that we cannot deploy it at least for the next 50 years, there is no need to incur these costs now." - this argument doesn't hold weight for AGI research, and I don't think it should for solar geoengineering. SG is highly neglected and as a fraction of CC research is minimal. The research will take decades to filter through to policy and international agreements, so it is worth starting research (not implementing) well before we are forced to use it.

"This would give us at least 20 years to cover the technical details and a governance framework." - A lot of the warming is already locked into the ocean. Giving another 30 years before starting to research will likely be too late. I'm not in favour of implementing solar geoengineering now but researching the viability of these measures now seems to be promising, if not for application then for global security to dissuade rogue actors from implementing the measures with false / incomplete information / encourage preventative policy decisions. This requires fundamental technical research to assess the risks.

"This seems to me like enough time, given that: Solar geoengineering is probably technically feasible with adaptations to various different current technologies." - I'm not sure the current technologies that you are referring to that can be adapted, but the more promising interventions are all much larger scale such as sulfur aerosol injections and have almost no precedent (volcanic eruptions can only tell us so much).

Finally "Another risk of solar geoengineering research is that it will uncover new technologies that could destabilise global civilisation. I discuss weaponisation risks in section 3.2 of my paper. " - As your paper says, current information on SAI indicates that it will take a highly technically adept state actor decades of spending tens of billions of dollars and will still not be a permanent doomsday device (and will be obvious to other states and easily counteracted). All in all I find it difficult to imagine that SAI research will discover something that is easier and cheaper to generate a doomsday device than already exists in a conventional nuclear weapons stockpile. Additionally this doomsday device would also exterminate the user, whereas nuclear weapons can be directed at other states with no immediate, direct blowback (of course the political and social cost and likely retaliation from affiliated states are the reasons why we haven't seen this happen yet). So the implication is that the malicious actor would also have to be suicidal. This doomsday device would also take time to work, which would give time to find a counteraction, and if research inadvertently discovers this application the time to find a solution will be the time from that research until the time of implementation.

Adding to this, currently climate change is projected to be a major stressor on international politics which can exacerbate nuclear X-risk, as well as expanding vectors for natural pandemic risks, among others - so this should also be in consideration when considering if SAI may uncover new X-risks as the baseline p(X-risk) for the coming decades is likely to be a curve rather than flat.

In conclusion the injection of CO2 and methane into the atmosphere may already constitute a moral hazard and dangerous weather manipulation method, and I think that we should be researching (not implementing) potential technical geoengineering solutions in order to prevent the expected outcomes of climate change as well as many other potential (part) solutions (As mentioned by others SAI doesn't reduce CO2 levels and so does nothing for ocean acidification and other related issues). We should evaluate the risks and if (as I expect we will find) them to be too high due to uncertainty, we can use that information to construct international policy around this issue.

I do not disagree that the Vice piece and the think tank research are likely alarmist and unrepresentative, but unfortunately in my opinion John Halstead's analysis and the underlying IPCC reports are entirely too optimistic on the flipside. I think this leaves a lot of room for further serious evaluation of the potential existential risks on climate change.

Firstly, John Halstead's review of existing literature. I was privileged enough to go to his talk EA Global London 2018 which was a summary of this work. It is a very large and understudied field and partially as a result of this his work has focused on direct and existential risks.

Indirect risks seem to include a huge host of pressures on society, infrastructure, and other trappings of civilisation where the impulse of these pressures could be very significant in causing widespread collapse. I don't think he specifically states here but he goes with the more clear, academic definition of existential risk of the extinction of the human species.

These two put the frame for the whole analysis as entirely too narrow for what most people are asking, and why the Vox piece is misrepresenting his work by saying it likely isn't an existential threat. In my opinion the responses of multiple competing nations to a world with increasingly shrinking and variable resources is a critical factor and while it is nearly impossible to forecast likely interactions, I don't think they're going to be pretty. Nuclear war is a separate X-risk and its probability is likely to be vastly increased with severe climate effects to start with, but there are several other indirect and interdependent mechanisms that could exacerbate the rate of collapse.

The definition of existential risk as more widely accepted in the EA community is a lot more general than that seemingly assumed by John Halstead or stated in the Vox piece - the permanent or sufficiently long-term curbing of the progress rate of the human species is included, generally to sustained existence at pre-industrial levels. While he does state when talking for the Vox article that he doesn't see even +10C being a threat to this broader definition, I at least don't see any substantiation of that in his research document, and throughout the Vox piece (and in his analysis) this seems to be a throw-away comment rather than the overriding definition that is being used is either strict extinction or is confused and assumed.

Interestingly as an aside, in Halstead's reasoning he seems to have updated from his previous dismissal of ecological collapse where he now sees it as a more complex and uncertain issue. There were some pre-Eocene extinctions related to carbon release events (the counterfactual being a large part of his justification for the theory's dismissal), as well as the paleontological record not accounting for unique Anthropocene pressures on populations including "speed, magnitude, and spatial scale" (from the IPCC). I see these as critical factors for the determination of adequate adaptation response in ecosystems / specific species at the high-impulse end where we seem to be operating. At longer timescales of thousands of years (still very high impulse climate change for paleontology) this does allow for some Darwinian selection and migration, whereas we are operating on a scale of approx. 200 years (and we are already living through a mass extinction - whether this leads to ecological collapse is another matter).

Aside from the framing of the research with narrow definitions (and leaving out a significant chunk of relevant mechanisms that could contribute to climate change being an X-risk by confining it to direct effects only), for me there is a real problem with Halstead's analysis in that he bases it largely on the IPCC reports, which are (in my opinion) egregiously optimistic. The simplest reason for this is the case of incentive. IPCC stands for "Intergovernmental Panel on Climate Change" and while it does take into account the consensus of climate scientists it is also an intergovernmental report that has specifically political objectives and the language of which is curbed and adjusted in various ways to further various agendas. This includes a Summary for Policymakers (SPM) which includes a "final round of written comments on the SPM, before governments meet in plenary session to approve the SPM line by line and accept the underlying report." This is the final stage but also governments have individual and serious involvement in the assessments made and to assess how palatable the conclusions are in order to try to foster (or impede) positive progress on these issues in their home countries. [It has been speculated that one of the reasons why the IPCC reports are so illegible for laypeople - excessive stating of confidence intervals and complex justifications - is to allow policymakers to more easily ignore it as they are, largely, laypeople themselves, and this is by design and for plausible deniability]. There is also widespread consensus that (for reasons above and others) the IPCC reports also aim to be conservative and non-alarmist in order to prompt change by making it appear more feasible.

Additional glaring problems with a lot of climate models is that many don't account for feedback loops, and there are plenty of reports about gross underestimations of releases and warming effects, and we don't have very good data on runaway effects and how much they could accelerate warming, in addition to how fragile the Earth's global ecosystem is. It is only relatively recently that it was calculated by NASA that 27.7 million tons of phosphorous-rich sand is being carried by cross-Atlantic winds from the Sahara to the Amazon (where soils are notoriously poor, and as such this transfer is important for the flourishing of the Amazon). Since the Amazon is about the size of the Indian sub-continent, a change in wind patterns that stopped this nutrient transfer could prove highly damaging to this rainforest. As far as I know transfer mechanisms such as this are also not considered in many climate change models.

As a brief summary: I don't think that alarmism such as "high confidence of extinction" is helpful or accurate, but there is a significant chance that climate change could be an X-risk which isn't being studied in anywhere enough detail, especially since climate change is an immediate threat where we could be "locking in" permanent changes (and possibly runaway destabilisation effects, such as clathrate melting / Arctic sea ice melting / permafrost thaw) for hundreds of years inside the next 30 or 40 years, even by conservative and heavily adulterated IPCC estimates. It would be good to know if those changes we're locking in will soon lead to the extinction of the species or (in my opinion far more likely) the regression of humanity to pre-industrial levels, potentially for millennia.