Robin studies climate change.
Robin
Karma: 260
I think we’re getting closer to an agreement. I would be more tempted if your thesis were “energy will become much more expensive at some times of day/year, as will certain minerals, and this will depress GDP compared to naive expectations.” It’s not obvious to me that low energy storage does more than require heavy industry to relocate to more consistent climes and/or stop for a few days each year, which would depress GDP but hardly to the level of existential threat.
I think most of these are economic points about how expensive it is to make the transition, rather than showing it’s impossibile. It certainly won’t be cheap in any individual sector, but as a fraction of the global economy we aren’t necessarily talking very large amounts of investment for these changes, and many governments already have plans and incentives to make this happen. A lot of this analysis feels like you trying to make a new Integrated Assessment Model (IAM) from scratch without writing down equations, and I think the disagreements you have with existing IAMs are not as substantial as you think. Things like land use constraints for biofuels are typically included in good models, as is the inefficiency of hydrogen, e.g. based on IEA values in [1]. You might dispute the numbers but they’re fundamentally reasonable. Land use is a problem if you want to power a large fraction of the world this way, but not if you just want to power a few small sectors like aviation, and provides some defence against renewable variability. The truth is there is no silver bullet for these transitions, but a range of viable portfolios that are hard to calculate without numbers.
I am more techno-optimist than you, and therefore think that we can sustain a mild continuous increase in energy use from only the improvements in the efficiency and affordability of renewable energy (as in the Ren scenario) and this enables a large increase in GDP, if that’s something society wants. I don’t think this is indefinitely required anyway; I don’t think it’s a particular problem for society if GDP grows subexponentially in 50 years time, or even remains constant at a high level for everyone.
I think you would like this paper [2], which makes a similar point regarding energy investment requirements. Although a lot of IAMs don’t include this, for small modular technology like solar cells and wind, it’s not that big an issue (whereas it is for nuclear). This is also why bioenergy is so popular in spite of the low efficiency—no adaptations required to use it. As stated above, I believe the inefficiency of green hydrogen is accounted for in good models.
[1] https://www.sciencedirect.com/science/article/pii/S0306261922002501#b0170 for hydrogen, https://www.pik-potsdam.de/en/institute/departments/activities/land-use-modelling/magpie for land use models
[2] https://www.science.org/doi/10.1126/science.aaz8060
These are not new technologies—thin film and primarily-organic PV have been commercially available for decades. They don’t out-compete silicon based on price point/efficiency, not unviability [1-2]. The organic films are again very thin, so very little land is required to grow the material to make them (the question would be how many times over a piece of land could produce the feedstock to cover itself in a year, I’m sure it would be tens of times). Similarly, the volume of copper and zinc mined in a year is enough to put a few nanometers around the world, and a few years of that would generate a fair amount of power already (not that I recommend doing this). Also, silicon itself isn’t scarce, just the dopants, which are required in extremely small quantities.
You can already buy electric trucks [3] and smelt iron by hydrogen [4]. Planes (much harder to decarbonise) can already be powered by biofuel [5].
Their properties are less good but if they were much cheaper we would spend more money researching them to make them better. The comparison between manufacture energy requirement and storage energy requirement is irrelevant because the storage happens cyclically more than 100 times once you’ve got 100 batteries you use the to make the 101st. You don’t address iron oxide batteries in your work, nor do you investigate things like compressed CO2. Several of your arguments substitute technological challenge and economic considerations with material ones, most notably your section on hydrogen, which for grid-level storage does not suffer from any of the fundamental-material problems your work otherwise attempts to demonstrate.
While the total energy requirements of the world increased, they increased much more slowly than GDP; this is sufficient to demonstrate decoupling. The decoupling I showed is for global values, so commenting that someone has to produce things somewhere doesn’t pose a problem. For point c., you must be using a very weird measurement of efficiency (do you mean the fraction of GDP spent on oil?). Graphs like this [6] show that the energy required per unit GDP has been declining. The direction of the link between energy consumption and GDP is disputed [7, 8]. The main Giraurd document you cite arguing energy → growth does not appear to be peer-reviewed, and both it and the peer-reviewed Ayres document end their analysis before renewable energy becomes a notable fraction of the total.
Paris-compatible targets are all well low fossil fuel supply except possibly the NEG scenario. They don’t model specific mineral use because they understand that technology on that granular a level changes more quickly than it can be integrated into the models, e.g. the handwringing over the need for cobalt in batteries is getting pretty dated [9].
[1] https://www.fortunebusinessinsights.com/industry-reports/organic-solar-cell-market-101555
[2] https://www.alliedmarketresearch.com/thin-film-solar-cell-market
[3] https://www.volvotrucks.com/en-en/trucks/alternative-fuels/electric-trucks.html
[4] https://cen.acs.org/environment/green-chemistry/steel-hydrogen-low-co2-startups/99/i22
[5] https://newatlas.com/aircraft/airbus-a380-biofuel-first-flight/
[6] https://yearbook.enerdata.net/total-energy/world-energy-intensity-gdp-data.html
[7] https://www.nowpublishers.com/article/Details/IRERE-0121
[8] https://link.springer.com/article/10.1007/s41247-021-00090-x
I think a deeper look at several of these points shows that it’s not as bad as it seems.
1) It is already quite possible to make solar cells and batteries without any particularly rare metals [1], and some solar cells can be constructed either from films with active areas only nanometers thick (meaning only a few million tons are required to coat the world in them) or entirely out of organic components [2]. Similarly, while the most commercially viable batteries at present may involve somewhat scarce metals like lithium, it’s possible to make them out of most substances, including iron, which is the 4th most abundant element on earth, as well as storing energy in compressed air or capturing hydrogen from water. When materials get scarce, technology is directed to solve these problems; there is not a physics-based limit on human energy consumption at anything near our current level.
2) Energy use per person has been falling in many developed nations for some time as GDP per capita rises, and energy use per person globally has not been rising that fast (about 12% over the last 4 decades) [3], whereas GDP per capita at PPP has > doubled. So, assuming that population stagnates as currently predicted and computational advances continue to deliver about as many efficiency savings as they cost in energy, I see no reason to assume an ever-increasing energy requirement. Obviously an AI explosion could unsettle this, but is not inevitable. The flipside of this, as you comment yourself, is that stagnant energy supply places some sort of limit on the development rate of AI in its current architecture, although historically energy requirements for compute have halved every few years, so not necessarily a very strong limit. As compute takes over more of the economy, it’s even possible to argue that we expect the energy requirements of many sectors of it to decrease at this rate.
3) How exactly one transitions to largely renewable (carbon-neutral) future is an extensive area of research, but it is safe to say that there are a huge variety of plausible ways to do this, many of them allowing for moderate growth in total energy use. For instance, here is total energy use under the IPCC IMP emissions scenarios, all constructed by different socioeconomic modellers and but the first two leading to under 2C of warming [4 and figure below].[2] https://onlinelibrary.wiley.com/doi/full/10.1002/ente.201402153
[3]
[4] https://www.ipcc.ch/report/ar6/wg3/downloads/report/IPCC_AR6_WGIII_Chapter_03.pdf
Strong agree, and part of this is just that EAs should be more modest about how much their assessments of sector impact out-perform other people’s. In the long term, weird second-order social impacts of interventions matter a lot more than the direct impact. For example, the (disputed) effects of abortion on crime rates https://www.sciencedirect.com/science/article/pii/S0047272721001043?via%3Dihub and female employment/social engagement https://link.springer.com/article/10.1007/s12122-004-1028-3 may create social spirals that continue long after the medical harm of the pregnancy and therefore considerably bump abortion rights up the virtual list of longtermist goals, but these effects are very hard to assess in simple models.
I respect you immensely for writing this, but some degree of altruism is required for being an effective altruist—not an infinite duty to self-sacrifice, but the understanding that you can be trusted to do so on big things, and costly signs you will do so are helpful. 10% giving is one such costly sign and it’s not required that you do all of them (I also think you overestimate the fraction of EAs who are vegan). However I think the disjunction between wanting the best for the world and wanting to have a high profile by improving the world occurs everywhere; in the fairly plausible world where AI alignment is impossible, your most effective action is probably either not working on AI or being subtly so bad at it that the field suffers, neither of which will win you much status (assuming you can’t prove that alignment is impossible). This is a general instance of the problem outlined here: “I am most motivated by the prospect of creating a radically better world as opposed to securing our current one from catastrophe”. A biased motivation combined with the unilateralist’s curse can easily give your actions negative expected utility but positive expected status-payoff: you don’t lose face if everyone goes extinct. There are lots of plausible real examples of this, like geo-engineering or gain-of-function research. Which way you’d fall on these questions in practice is a much better test of whether you’re “actually EA” than whether you buy cheap things.
On a more institutionally level, it is unhelpful for EA to become associated with narcissism (which in some circles it already is). Since the cost is borne by the movement not the individual we expect misalignment until being EA is harmful to your reputation, so some degree of excluding narcissists with marginally positive expected personal impact is warranted.
For those abroad unfamiliar with quite how unpopular Dominic Cummings is, here’s an article arguing he was the most unpopular man in the country at the time
and here’s a poll from May 2021 showing only 14% of people trusted him on government handling of COVID, by comparison with 34% for the prime minister. https://docs.cdn.yougov.com/iszcru07g6/TheTimes_Coronahandling_Cummings_Results_210520.pdf
This is an incredibly important question. It is also an incredibly dangerous one. There are many real EA whose views on this topic constitute either an X-risk or an S-risk to EAs with only subtly different assessments: people who, given a truly aligned omnipotent AGI would either wipe out the majority of humans or create many lives others view as unhappy. Historically, well-intentioned eugenicists have killed many people who self-identify as having worthwhile lives.
I also think there is a miscalibration in the creation test; many humans instinctively view people similar to them as competition, and either like or dislike the idea of clones of themselves for other reasons. The advantage of the suicide test is that you are centering your judgement on a real person who can express their actual preferences in the moment, rather than a hypothetical case. That seems worth a lot of offset error to me.
I agree with the comments that you are massively overconfident in the applicability of your logic and your title, but raise a good point for a marginal-but-political voter. However in practice both USA parties are actively ridding themselves of moderates, so an important first step is to push for voting reforms that make it possible for young, sane people to get into positions of power in the Republican party without either actually or pretending to hold a swathe of clearly harmful views about e.g. climate change, abortion access and pandemic preparedness. Republicans know their demographics have bad optics, so unless you actually believe they are the better party, you want to be sure to get a good deal on policy in exchange for fixing them.
I can’t imagine much worse for the prospects of EA than to be associated with Dominic Cummings. He of course now claims to have had very little influence on COVID policy, since it was widely regarded as a shambles during his time. Whatever the truth of this, it’s precisely through association with widely hated figures like this that EA risks evaporating as quickly as it has grown. This is also why we need to make a strong distinction between “the abstract idea of maximising positive impact,” which would survive this, and “the social movement that this forum is part of”, a member of a rather short-lived reference class (albeit with some good signs at the moment, as you say).
I feel like it defeats the whole point of April 1st if articles clearly declare themselves to be parodies at the top of the page. Although it’s still a fair bet that some people won’t notice.
Why 80 000 hours should recommend more people become drug lords
This info was not present at the time I wrote this reply. As you say, most of this doesn’t apply in the case of Russians, but it seems unwise to even discuss the optimal actions for Russians in a non-anonymised public chat.
I’m deeply skeptical that this is a good action for anyone who isn’t personally tied to Ukraine. Foreigners fighting is a substantial propaganda boost for Putin (“Look, Ukraine is a Western stooge after all”) and risks muddying the water around whether or not the nation they are from is at war with Russia, potentially spiralling the conflict. NATO is holding off doing any act tantamount to declaring war for a reason. Ukrainians have the benefit of local geographic knowledge, being trusted by the community and speaking the co-ordinating languages; what they lack is high-tech military equipment. As you can probably tell, my skepticism also includes people who DO have military training, as this exacerbates the risk of it being perceived as state-sponsored assistance, even if these people are more useful militarily. I’m more sympathetic towards volunteers that have medical training and are going in a strictly non-combat facility.
Indeed, although CATF is still very North America-centric. I’d be more excited to learn about a similar charity acting in China (assuming it wanted Western money).
This is a really great discussion piece and a very mature response from Giving Green (GG) to it. I would mostly second jackva’s comments, and will just raise a few additions points.
There seems to be a misapprehension on the core criticism. The fundamental criticism of this article is not just that GG don’t do things quantitatively, but that they have completely neglected the “cost” side of cost-benefit analysis. The qualitative metrics don’t attempt to account for either the actual scale of money needed to do anything, nor the potential negatives of the actions. It appears designed to convince me that the charities accomplish anything at all, rather than that they do good things efficiently.
Minor disagreements with the article
(Flagged to avoid the impression of consensus here rather than to divert the discussion onto these topics)I would challenge the article’s assertion that CATF has no significant downsides. Quite apart from CCS debates, the new distributed nuclear faculties it proposes have elevated risk of enabling nuclear terrorism, which (besides being bad itself) can trigger further backlash against established nuclear generation. There are fundamental safety reasons why nuclear technology develops so slowly now, and why it isn’t widely distributed. There are fundamental geology reasons why CCS at high flow rates is hard, and the trend of attractive-looking test cases that are ultimately distracting failures is historically real.
I’m also confused by the article’s criticism of BURN, which seems to be a valid company that sells carbon offsets. It clearly has social co-benefits from pollution reduction, and it’s not claiming anything other than high-confidence emissions reductions. The write-up about BURN is worded misleadingly but not conceptually flawed.
Disagreement with the response and additional problems
In terms of analytic models, I agree that GHG emission changes are hard to establish with exactitude, but not obviously harder than the social benefits of any other intervention. In many cases I would expect the error bars in GG’s analytic evaluations to be smaller than So Give’s, and have a fairly uncontroversial choice of meaningful unit (e.g. tCO2-equivalent/$) unlike in social problems. Given that several options are pre-packaged as carbon offsets for a particular price, GG really just need an estimate of the probability that the emissions will really be trapped/avoided for some lifetime.
Ironically, the large range of things that are fundamentally difficult to include quantitatively are lumped into a single category: “co-benefits”. I don’t know who this is for. If I was donating from a pot of money that cared about development per se, I would go to Givewell or So Give to work out how. If I care about community buy-in (a really important aspect in the success of many projects), I need to know how those co-benefits are distributed, not just that someone somewhere benefits, and this should really be a part of “Causality”. If I care about biodiversity or literally any other green metric than global warming, I would be consistently disappointed by how little was written here.
Finally, I struggle with way this entire discussion revolves around US policy change. I suspect the amounts of money spent on lobbying in the USA is hugely more than in most countries, but I don’t see any consideration of neglectedness at the country level. Your research priorities document claims: “We focused on US policy because the US is the world’s second-largest emitter, it has outsized global influence, and because Giving Green’s staff are most familiar with the US policy systems.” The first two are not sufficient to demonstrate it’s optimal (it’s not even the first choice!), and the third is a problem with GG rather than the basis for a conclusion. Have you looked at state-level interventions? The whole world is even bigger than the US (and China), so why not look at global movements? Can you not either diversify your staff or expand their sphere of knowledge?
Conclusions
The amount of opinions expressed here should be interpreted in three ways: firstly that a lot of people are really unhappy with the current GG methodology, secondly that they are willing to offer free advice and assistance, and thirdly that what GG are trying to do is really valuable, hence why it’s so important to get it right.
TL;DR: Assuming everything can be fit with a linear trend completely overwhelms the importance of working out what that trend is in these extreme cases, so while instructive for median behaviour, I don’t believe this approach is sufficient to assert anything about tail probabilities.
It’s good to see so much work summarised in one page, but the cost of this is rigour. I agree with the problems with using ECS as mentioned above, and add that, since these trajectories do not result in net 0 CO$_2$ emissions at 2100, it’s not even a good approach to estimate the temperature in 4100 (in the imaginary world where really slow and hard-to-model things stay the same, but easy-to-model things don’t, except CO$_2$ [1]). It’s also worth noting that TCRE normally assumes a linear CO$_2$-T relationship rather than logarithmic, although this is disputed [2], and not really designed for changes of many degrees. A similar problem exists for carbon intensity. You assume an exponential decay, but so far we’ve seen a pretty linear one. (This would imply negative emissions will eventually happen on their own!) While it’s good that you put so much effort into probability distributions of these values, it doesn’t help if you’re wrong about the equation they go in.
Regarding constant carbon intensity improvements (geometric, linear or otherwise) and extra effort, I’m not really clear what you’re proposing needs conserved – a conservation of the current level of effort into decarbonisation, or a conserved rate of change of effort into decarbonisation (since we’ve clearly been putting more effort in recently). It feels like you’re implying a constant effort derivative, i.e. slowly increasing carbon price and legislation.
You (and many others) complain the IPCC does not report extremes of the ECS PDF, then complain about what they are. The IPCC specifically makes a point of not quoting values for these extremes because there isn’t any consensus on it. We do not have > 95% confidence that the full simulations aren’t missing some big factor, in the same way we missed the breakdown of the ozone layer until after it was observed. The presence of the ozone hole, and various other weird new atmospheric chemistries, places similar limits on our confidence in paleoclimate data, as does the unprecedented rate of CO$_2$ release [3]. This does indeed increase the importance of the priors, which is why the fact we can’t agree on them is so problematic. By this point I don’t think it’s possible to disentangle true priors from decades of simulations, back-of-the-envelope calculations and climate history, and since we want to use all of these factors later, none of them can be considered truly prior. The degree of agreement between old and new estimates of ECS is interesting but irrelevant, since it doesn’t include those tails.
Your final point, that the ‘median view’ is that Earth system feedbacks are less important, is inconsistent with the degree of rigour shown elsewhere in the article. You aren’t interested in the median view of these, you’re interested in the 95th percentile views. And that should feature some of these ZOMG WE’RE GOING TO DIE!!!1 papers.
[1] Beyond equilibrium climate sensitivity, Knutti et al 2017 http://iacweb.ethz.ch/staff/mariaru/BeyondEquilibriumClimateSensitivity/KnuttiRugensteinHegerl17.pdf
[2] Implications of non-linearities between cumulative CO2 emissions and CO2 -induced warming for assessing the remaining carbon budget, Nicholls et al.
https://iopscience.iop.org/article/10.1088/1748-9326/ab83af/pdf
[3] Anthropogenic carbon release rate unprecedented during the past 66 million years, Zeebe et al.
I did a crude calculation in DICE2016R, which doesn’t take into account a wide range of effects nor most of the points in my comment below about elasticity. In terms of damage to the economy, the social cost of carbon for 10 years, 20 years and 30 years is about $5, $10 and $14, verses a current total social cost of carbon of $37. This is just taking the social cost of carbon now minus the discounted social cost of carbon in the future for the optimised development pathway. It’s about an order of magnitude lower in the non-optimised (baseline) pathway for DICE. General disagreement over the social cost of carbon between models may make this vary over orders of magnitude and the DICE model is low compared to models like PAGE.
Something that the authors of this book perhaps should have highlighted is that DICE’s main virtue is its simplicity: it is far from being either the only or the best IAM for most analyses. However, to appreciate how badly calibrated the damage function is, here’s a note from the documentation:
“However, current studies generally omit several important factors (the economic value of losses from biodiversity, ocean acidification, and political reactions), extreme events (sea-level rise, changes in ocean circulation, and accelerated climate change), impacts that are inherently difficult to model (catastrophic events and very long term warming), and uncertainty (of virtually all components from economic growth to damages). I [Nordhaus] have added an adjustment of 25 percent of the monetized damages to reflect these non-monetized impacts. ”
(Quote is from the manual of DICE 2013R, http://www.econ.yale.edu/~nordhaus/homepage/homepage/documents/DICE_Manual_100413r1.pdf , still valid for 2016R version as per https://www.pnas.org/content/114/7/1518/tab-figures-data .)
For most reasonable emissions pathways, temperature and linked physical effects depend only on cumulative emissions*. Delaying a given emission by some time therefore does not impact the amount of climate change it causes, so from a climate-focused perspective we don’t see any change in the harm of emissions with time (this may not be true at very low net emissions rates but is at rates similar to present-day). This would mean that the only time delaying emissions would have any climatic benefit would be if they are delayed until a time when net emissions are negative (in which case the world experiences a lower peak cumulative emissions than it would do when emitting without the delay, which we assume is less bad). It’s not clear if and when this will happen, and climate-based discounting would be 0 before that point.
This suggests that for all climatic ‘badness functions’ (effects on humans/ecology) no discounting is needed, however this may depend on the rate of change as well as the state of the system, and human impacts may also depend human development, equality and preparation for climate change. As we hope that the rate of emission will begin to decrease soon, this would mean that delayed emissions might be less impactful in the future. It’s going to be very assumption/IAM-dependent as to how much though. It’s also not clear that this generates a positive discount rate—it’s possible that people seeing more climate change sooner incentivises more research/investment in averting it, which takes time to pay off.
It’s important to distinguish two different factors that could be relevant when discussing this – one is the social cost of carbon (potentially measured in money lost, or in more egalitarian DALY losses), the other is the carbon market value of carbon. If one assumes the existence of a well-functioning global carbon market, then emissions at times after this may be largely absorbed by elasticity. However at times prior to this/if the market is not comprehensive, the ‘offsetting’ may be just displacing consumption.
A lower limit on the discount rate could come from the probability of catastrophic events (which may be a function of pure time, carbon concentration and derivative of concentration). In the event of a nuclear war, meteorite impact etc. our climate may no longer be determined primarily by emissions concentrations, hence carbon released after this period is of lower importance.
Getting closer, anyway. Maybe we will have to
We already know how to solve the blackouts problem via dedicated generation (or storage) for high-impact sectors. In a renewable economy, very large amounts of energy are available very cheaply at certain times, so for instance a factory with a 1-day battery that can produce at night before sunny days is able to work nearly as efficiently as they do now. You aren’t actually relocating very much of the economy (only very heavy industry) and this constantly relocates towards incidentally-sunnier countries anyway for labor-cost reasons.
I’m not an economist so I don’t know that it’s pointful to get into a long debate about economics, but it’s pretty clear from how governments can reshape the economy during wartime implies that they have tremendous capacity to restructure the economy when they want to. Your analysis doesn’t make sense to me because green tech is something that makes energy; it’s an energy loan, not an energy expenditure.
Investment in a green transition solves much of the long-term underinvestment problem at the same time, so the historic underinvestment is not really an additive factor.
Good IAMs don’t take GDP as an external input, and the fact that one you cite does is a bad sign. I had not heard of this IAM before (I work around IAMs professionally but don’t code them myself), but it doesn’t seem to understand the basics of the laws of supply and demand, assumes fixed demand and then complains when this can’t be fulfilled. This means it supports your point that GDP is suppressed, but doesn’t qualify by how much. It also assumes, for instance, that it is impossible to increase recycling rates. The possibility of recycling largely solves your point c), since recycling requires no additional land and less energy than we currently spend making plastic—it’s simply that the processing is not financially incentivised by the low cost of oil.
The GTK report isn’t an academic document and doesn’t have an obvious IAM attached.
I’m unclear why you jump from cars to plastic. At any rate, only the waste fraction plus the growth in the number of vehicles (or plastic) requires new material, the rest can be recycled. There appears to just be a disagreement between two models over the number of vehicles in existence at the moment, not a modelling discrepancy.
I don’t think its wise to generalise from 2008 to the future of the world, but I also don’t know enough to argue about this.
Factories constantly update and move anyway, following cheap labor, as discussed above. China seems to hope to seriously bulk out its energy grid for renewables in about half a year https://www.reuters.com/business/energy/chinas-state-grid-invest-22-bln-ultra-high-voltage-power-lines-report-2022-08-03/