Taking into account the expected benefits of denuclearization (i.e. fewer deaths and injuries from nuclear war), the expected costs (i.e. more deaths and injuries from conventional war due to weakened deterrence), and the tractability of lobbying for denuclearization, I find that the marginal expected value of lobbying for denuclearization to be 248 DALYs per USD 100,000, which is around 40% as cost-effective as giving to a GiveWell top charity.
- The results potentially overestimate the risks insofar as I focus on the three major nuclear war risks (i.e. NATO-Russia, US-China and India-Pakistan), and leave off the more minor flash points (e.g. Israel-Iran, NK-US) when calculating the average deaths from nuclear war.
- On the other hand, the results may also potentially underestimate the risks insofar as existing evaluations of the probability of nuclear war, which this analysis relies on, likely over-anchor on major conflicts and neglect the additional risks from more minor ones. This effect balances out the first issue, but to what extent, it is hard to say.
- Results also fail to take into account the changed risk in war between NATO, the US, Russia, China, India or Pakistan and other countries, whether nuclear or non-nuclear, though this is unlikely to be significant insofar as no plausible estimate of the decreased chance of conventional war can outweigh the risks of nuclear war.
- Results are very sensitive to hard-to-calibrate tractability estimates, and reasonable people can disagree (e.g. on the right outside views to use; on inside view estimates; and on how you weigh the two). A lot of the calls here rely heavily on my pre-existing academic and professional knowledge of geopolitics, but my knowledge is far from perfect, and in any case reasonable people will disagree.
- It would be valuable to consult experts on tractability – especially country-specific national security experts – if deeper research is done.
- There are potentially more tractable ways of reducing risk that aren't lobbying for denuclearization (e.g. mitigation work through strengthening global food systems, like what ALLFED does), which I do not evaluate here.
Expected Benefit: Averting Nuclear War Fatalities
The main expected benefit of denuclearization is, of course, preventing people from dying from nuclear war. I model this in the following way.
Moral Weights: I take the value of averting one death to be 29.3 DALYs. This is calculated as a function of (a) a human's full healthy life expectancy of 63.69, (b) a minor age-based philosophical discount, and (c) assuming we save someone of the median age in the relevant population. For more details, refer to CEARCH's evaluative framework.
Scale: I calculate the average number of deaths across the three major nuclear war scenarios: NATO-Russia, US-China, and India Pakistan. In each case I take into account both direct violent death as well as death by starvation from nuclear winter. Multiple estimates are used and aggregated in each case to even out random error.
For the direct deaths from nuclear exchange in the NATO-Russia case, I take Luisa's estimate (51 million deaths), Wellerstein et al's estimate (34.1 million deaths) and Toon, Robock & Turco's estimate that considers a 1,100 warheads US attack on Russia and 1,000 warheads Russia attack on the US (100,000,000 deaths), and then calculate a weighted average (37.43 million deaths). A far greater weighting given to the Wellerstein et al estimate, due to Luisa's estimate relying on a probability distribution of countervalue targeting that isn't empirically well-grounded, and Toon, Robock & Turco looking only at countervalue targeting, making it a likely overestimate.
For the deaths from starvation in nuclear winter in the NATO-Russia case, I take Luisa's estimate (5.5 billion deaths), Xia et al's estimate (5 billion deaths), and Harwell's older estimate as adjusted for population growth (1.65 billion deaths), and then calculate a simple average (4.05 billion deaths).
In total, this yields around 4.09 billion deaths after a NATO-Russia nuclear exchange.
For the direct deaths from nuclear exchange in the US-China case, I take Kristensen, Norris & McKinzie's estimate (14.14 million deaths), by averaging the fatality figures from multiple scenarios to obtain US and Chinese fatalities respectively and then summing them. I also take Toon, Robock & Turco's estimate (217.57 million deaths), by considering a 1,100 warheads US attack on China and 200 warhead Chinese attack on the US, with the casualties under the latter calculated by (a) assuming that China launches its full 200 warhead arsenal, (b) taking that casualties scale with diminishing marginal returns from warhead use due to larger cities being destroyed first, and (c) using the 1,000 warheads Russian attack on the US as a baseline for casualties. Finally, I take an extrapolation of Miyazaki's estimate (179.64 million deaths). Here, I calculate Chinese fatalities by averaging the US's Alert Force and Full Force attacks to obtain a casualty ratio that is then applied to China's total population, the result of which is then subject to the fatality-to-casualty ratio from Toon, Robock & Turco. Meanwhile, I calculate US fatalities by taking the US death figures from Soviet attack as a baseline and adjusting for the smaller Chinese arsenal by (similar to above) assuming that China launches all 200 of its warheads, and that casualties scale with diminishing marginal returns from warhead use due to larger cities being destroyed first. And all that done, I take a weighted average of the three estimates (44.89 million deaths), where a far greater weighting is given to the Kristensen, Norris & McKinzie estimate due to Toon, Robock & Turco looking only at countervalue targeting, making it a likely overestimate, and due to the extrapolated Miyazaki estimate relying on additional uncertain extrapolations.
For the deaths from starvation in nuclear winter in the US-China case, I use the same figure calculated for NATO-Russia (4.05 billion deaths).
In total, this yields around 4.09 billion deaths after a NATO-Russia nuclear exchange.
For the direct deaths from nuclear exchange in the India-Pakistan case, I take Toon et al's estimate (87.5 million deaths), a US intelligence estimate (10.5 million deaths), and McKinzie et al's estimate (2.9 million deaths), and then calculate a weighted average (74.03 million deaths) – a much higher weightage is put on the more recent Toon et al estimate vs the other two considerably older ones, since both India and Pakistan have been increasing their number of nuclear weapons over the past two decades
For the deaths from starvation in nuclear winter in the India-Pakistan case - this is more limited than in the superpower scenarios, and I calculate this by taking Xia et al's estimate (2 billion deaths), Helfand's estimate (1 billion deaths) and Robock and Toon's estimate (1 billion deaths), and taking a single average (1.33 billion deaths).
In total, this yields around 1.51 billion deaths after an India-Pakistan nuclear exchange.
And finally, I weigh each individual outcome (i.e. NATO/Russia nuclear war, US/China nuclear war, and India/Pakistan nuclear war) by the relative probability of occurrence, as Luisa has most helpfully collated here, to calculate the total deaths from an average nuclear war (2.51 billion deaths).
Persistence: The risk of nuclear war persists over time, and solving it brings long-term benefits. When summing the per annum benefits, however, they have to be discounted by various factors.
Firstly, I discount for the probability of the solution not persisting (i.e. renuclearization even after initial denuclearization success). As a baseline, we have to note that no state that has denuclearized (i.e. South Africa, Ukraine, Belarus & Kazakhstan) have yet renuclearized, which should put our priors close to zero. At the same time, it is clear that there is a temptation for states to renuclearize if their security conditions worsen (c.f. Zelensky's statements on the Budapest Memorandum). Hence, I assign around a 0.1% per annum chance to renuclearization.
Secondly, I discount for the probability of the problem being counterfactually solved anyway (i.e. countries deciding to denuclearize sans intervention). One potential way of calculating the reversal rate is to look at the number of years in which countries have given up their nuclear weapons, divided by the total number of years in which states have had nuclear weapons. This would theoretically give a base rate of denuclearization of 0.008, given 4 country-years where denuclearization (by South Africa, Ukraine, Belarus and Kazakhstan) occurred, out of a potential 504 country-years (where countries are nuclear armed). However, this approach is fatally flawed insofar as there is self-selection into denuclearization by countries who do not have significant security concerns such that the true base rate for the remaining nuclear powers will be far lower.
Instead, I approach this issue in the following way. First, let us consider what actors are looking to improve the situation. On the governmental side, there are some attempts at moving the P5 and hence the world towards denuclearization. Similarly, there are anti-nuclear charities (e.g. ICAN and the like) working on the matter. In contrast, businesses do not see this as something that concerns them. On the whole, therefore, I take the base rate of denuclearization to be the sum of the chances of pro-denuclearization voices within government succeeding in pushing for self-reform, and of non-profit lobbying from the outside being successful; each of them I take to be equal to the general denuclearization lobbying success rate of 0.00000016 (the calculations of which are discussed in the tractability section of this report). Further, I very marginally adjust the reversal rate up by 0.02 relative to the base due to two favourable trends - economic growth increasing the chance of democratization and hence democratic peace, thus reducing security concerns and increasing the odds of denuclearization; and also cultural shifts bringing increased liberalization and hence the chance of regime change (e.g. in Russia and China) or reduced nationalism (e.g. with respect to India and Pakistan), which will in turn reduce security tensions and increase the odds of denuclearization. Overall, this gives a reversal rate of 0.00003%.
Thirdly, I discount for the probability of the world being destroyed anyway (i.e. a general existential risk discount). Here, I take into account the probability of total nuclear annihilation, since the benefits of saving people from nuclear war in one year is nullified if they had already died in a previous year. For the exact risk of total nuclear annihilation, I take it to be one magnitude lower than the risk of nuclear war itself (discussed below), since nuclear war may not kill everyone. Of course, this probability will shift as a result of any efforts to denuclearize, but the chances of success are sufficiently small (as will be discussed) that it does not materially change our results. I do not take into account other existential risks like supervolcano eruption and asteroid impact, since the chances of those occurring at all is very marginal, let alone the chances of such events killing everyone and not just most people. Overall, therefore, I treat the general existential risk discount to be just the risk of nuclear war but adjusted a magnitude down, or 0.07%.
Fourthly, I apply a broad uncertainty discount of 0.1% to take into account the fact that there is a non-zero chance that in the future, the benefits or costs do not persist for factors we do not and cannot identify in the present (e.g. actors directing resources to solve the problem when none are currently doing so).
Value of Outcome: Overall, the raw perpetual value of averting nuclear war fatalities is 2.75 * 1013 DALYs.
Probability of Occurrence: Of course, the chances of nuclear war occurring (and hence the chance that it is a problem whose solution brings benefits) is very low indeed. To calculate the probability of nuclear war, I take a weighted average of the probabilities collated by Luisa: the probability of nuclear weapons being used during war as based on historical frequency (1.4%), of accidental nuclear war between the US and Russia as based on fault tree analysis and historical frequency (0.9%), of nuclear attack as based on expert survey (2.21%), of nuclear war killing at least 1 million as based on expert survey (0.39%), and nuclear detonation by a state actor causing at least 1 fatality as based on superforecaster prediction (0.4%). The weighting significantly favours the superforecasters' more conservative prediction, because of the following reasons. Firstly, the estimate of the probability of intentional nuclear war based on historical frequency is likely biased upwards due to historical use being in a MAD-free context. Secondly, the probability of accidental nuclear war based on historical close calls is highly uncertain due to the difficulty of translating close calls to actual probabilities of eventual launch. And thirdly, experts are notoriously bad at long-range forecasts, relative to superforecasters. Overall, this weighted average produced a per annum probability of nuclear war of 0.7%.
Expected Value: Overall, the expected value of averting nuclear war fatalities is 1.88 * 1011 DALYs.
Expected Benefit: Averting Nuclear War Injuries
Beyond preventing death, another expected benefit of denuclearization is preventing injuries from nuclear war. My modelling is as follows.
Moral Weights: I take the value of averting a typical injury in nuclear war to be 5.88 DALYs. This is calculated as a function of (a) the average disability weight for all injuries, (b) a minor age-based philosophical discount and (c) assuming we save someone of the median age in the relevant population. For more details, refer to CEARCH's evaluative framework.
Scale: The approach I take to injuries is the same as the one I took to fatalities, in that I calculate the average number of injuries across the three major nuclear war scenarios of NATO-Russia, US-China, and India Pakistan, with the aggregation of multiple estimates used for robustness.
For injuries from nuclear exchange in the NATO-Russia case, I take Luisa's fatality estimate and extrapolate using the Wellerstein et al ratio of fatalities to injuries (85.85 million injuries), the Wellerstein et al estimate itself (57.4 million injuries) and the Toon, Robock & Turco estimate (75 million injuries), to get a weighted average (58.56 million injuries) that gives far greater weightage to the Wellerstein et al estimates for reasons discussed in the previous section on nuclear war fatalities.
For injuries from nuclear exchange in the US-China case, I take the Kristensen, Norris & McKinzie estimate (24.35 million injuries). by averaging the fatality figures from multiple scenarios to obtain US and Chinese fatalities respectively and then summing them. I also take the Toon, Robock & Turco estimate (142.57 million injuries), deploying the same methodology and assumptions as when calculating US-China nuclear war fatalities. Finally, I take an extrapolation of Miyazaki's estimate (89.96 million injuries), using the same methodology as when extrapolating for US-China nuclear war fatalities, except with the extra step of applying the injury-to-casualty ratio from Toon, Robock and Turco when calculating US injuries. Ths yields a weighted average (39.63 million injuries), where I give a far greater weight to the Kristensen, Norris & McKinzie estimate for reasons mentioned in the previous section.
For injuries from nuclear exchange in the India-Pakistan case, I use the Toon et al estimate of fatalities, appling a 1:1 fatality-to-injury ratio as estimated by a previous study on India-Pakistan nuclear war (87.5 million injuries). I also use a US intelligence estimate (4.5 million). Further, I utilize the McKinzie et al estimate (1.5 million injuries). A weighted average is then calculated (73.42 million injuries), favouring the Toon et al estimate for reasons previously outlined.
All in all, weighing each war scenario by relative probability of occurrence, that gets us to 63.72 million injuries in an average nuclear war.
Persistence: The same discounts discussed in the section on nuclear war fatalities are applied here.
Value of Outcome: Overall, the raw perpetual value of averting nuclear war injuries is 1.4 * 1011 DALYs.
Probability of Occurrence: The same probability of nuclear war calculated previously is applied.
Expected Value: All in all, the expected value of averting nuclear war injuries is 9.56 * 108 DALYs.
Expected Cost: More Conventional War Fatalities
We've been discussing the benefits of denuclearization, but it is important not to forget that it does come with an expected cost – that of more conventional war fatalities - insofar as nuclear weapons deter conventional conflict. To model this, I incorporate the following variables.
Moral Weights: As described in the nuclear war fatalities section.
Scale: I look at the deaths from conventional war between NATO-Russia, US-China and India-Pakistan, with the aggregation of multiple estimates consistently used.
For deaths from conventional war between NATO and Russia, I use various outside views to estimate this. My first reference class is the war that most resembles a NATO-Russia conflict – the 2022 Russian invasion of Ukraine (72,000 deaths) – assuming the mid-point of the conflict is the present day of 1st October 2020 (i.e. the war continues for an additional period of time equal to how long it has lasted so far). My second reference class is a recent war involving the US – the Iraq War (409,000 deaths). And my third reference class is another recent war involving Russia – the War in Donbass, preceding the current invasion (12,000 deaths). This yields a weighted average (95,000 deaths), using a far higher weight for the 2022 Russian invasion of Ukraine, as it is most like a NATO-Russia war, in terms of weapons used and doctrines employed.
For deaths from conventional war between the US and China, I again use various outside views. My first reference class is the last US-China war – the Korean War (3.33 million deaths). My second reference class is a recent US war - so Iraq again. And my third reference class is a recent war involving China (for a given definition of recent, one supposes) – the Sino-Vietnamese War (56,000). This yields a weighted average (2.81 million deaths), wherein the Korean War is weighted most heavily given that it actually involved full-scale industrial warfare between US and China.
For deaths from conventional war between India and Pakistan, our job is easier insofar as the outside views we consult are simply past India-Pakistan Wars – the Indo-Pakistani War of 1947–1948 (9000 deaths); the Indo-Pakistani War of 1965 (3000 deaths), with fatalities calculated by taking casualties and applying the fatality-to-injury ratio from the previous war; and finally the Indo-Pakistani War of 1971, inclusive of the Bangaldesh genocide (1.66 million deaths). A simple average is then calculated (558,000 deaths) – the 1971 war is not penalized for the genocide, given the real potential for ethnic cleansing in the context of increased Hindutva nationalism and hostility to Muslims (on the Indian side), and obviously the Pakistani Army's history of genocide (on the other).
Persistence: The same discounts as before are applied.
Value of Outcome: Overall, the raw perpetual disvalue of more fatalities from conventional war between NATO-Russia, US-China and India-Pakistan are -1.04 * 109 DALYs, -3.08 * 1010 DALYs, and -6.01 * 109 DALYs respectively.
Probability of Occurrence: Relying on the academic literature, I estimate the reduction in per annum probability of conventional war between NATO-Russia, US-China and India-Pakistan as 0.000133 (sum of the reduced risk that NATO attacks Russia and vice versa), 0.000133 (sum of the reduced risk that the US attacks China and vice versa) and 0.000067 (reduced risk that India attacks Pakistan) respectively. Note that the base reduction in the risk that a country is attacked when it has nuclear weapons is 0.0000667 - but only if it has a potential first use nuclear posture (whether officially as for the US, Russia, and Pakistan, or arguably de facto with respect to Chinese policy on Taiwan); if a country has a official and de facto no first use policy, as for India, there is no measurable deterrent value.
Expected Value: All in all, the expected disvalue of more conventional fatalities is -4.65 * 106 DALYs.
Expected Cost: More Conventional War Injuries
Another expected cost is that of more conventional war injuries, which I model as such.
Moral Weights: As described in the nuclear war injuries section.
Scale: I look at the injuries from conventional war between NATO-Russia,US-China and India-Pakistan, with the aggregation of multiple estimates consistently used.
For injuries from conventional war between NATO-Russia, I utilize the same reference classes as when calculating fatalities – the 2022 Russian invasion of Ukraine (188,000 injuries), the Iraq War (254,000 injuries), and the War on Donbass (30,000 injuries) – and calculate the weighted average (180,000 injuries) using a far higher weightage for the Ukraine War, again because it most resembles a potential NATO-Russia conflict.
For injuries from conventional war between US-China, I employ the same reference classes as for fatalities – the Korean War (1.25 million injuries), the Iraq War (254,000 injuries), and the Sino-Vietnamese War (69,000 injuries) – and calculate the weighted average (1.07 million injuries) using a far higher weightage for the Korean War, again because it actually involves the US and China fighting a full-scale industrial war.
For injuries from conventional war between India-Pakistan, I once more resort to the reference classes of past conflicts between the two countries – the Indo-Pakistani War of 1947–1948 (17000 injuries); the Indo-Pakistani War of 1965 (4000 injuries), with fatalities calculated by taking casualties and applying the fatality-to-injury ratio from the previous war; and finally the Indo-Pakistani War of 1971 (32,000 injuries). A simple average is then computed (18,000 injuries).
Persistence: The same discounts discussed in the section on nuclear war fatalities are applied here.
Value of Outcome: Overall, the raw perpetual disvalue of more injuries from conventional war between NATO-Russia, US-China and India-Pakistan are -3.95 * 108 DALYs, -2.35 * 109 DALYs, and -3.92 * 107 DALYs respectively.
Probability of Occurrence: Same probabilities as discussed in the previous section on conventional war fatalities.
Expected Value: In sum, the expected disvalue of more conventional war injuries is -3.69 * 105 DALYs.
To assess the tractability of working on denuclearization – specifically, through lobbying governments – I use both an inside view (developed through a Fermi estimate that breaks down successful denuclearization into the necessary steps) as well as an outside view.
Let us discuss the inside view first. Since no government would willingly give up their nuclear weapons when their rivals – who potentially threaten them – continue to possess nuclear weapons, the question of whether a country X will give up nuclear weapons basically reduces to whether they will agree to denuclearize, conditional on their rivals agreeing to denuclearize. In certain cases, it will also be conditional on a country's superpower ally agreeing, since the latter will put effective pressure on the former to follow along with nuclear disarmament.
First, let us consider the probability of persuading the US to denuclearize conditional on Russia and China agreeing to denuclearize. While NATO has conventional superiority against Russia, especially post-Ukraine, there will be little trust that Russia will adhere to the terms of denuclearization over an extended period, given its tearing up the Budapest Memorandum and its flouting the Intermediate-Range Nuclear Forces (INF) Treaty, and also given the general air of distrust after Ukraine. On the China side of things, while the United States still retains conventional superiority for now, the balance of power is likely to change in such a way that disfavours the United States, keeping nuclear weapons to deter conventional aggression against itself and its allies more attractive. Overall, I rate the probability of success here to be around 10%.
Second, we have the probability of persuading the UK to denuclearize conditional on Russia and the US agreeing to denuclearize. At baseline, the same considerations vis a vis Russia apply (i.e. distrust over disregarded treaties, plus the general ill-will over the Ukraine invasion). Moreover, while the US would probably put pressure on the UK to agree if it itself agreed to denuclearize, this is counterbalanced by the uncertainty as to whether the US will commit to European security in the long term, especially if it pivots to Asia, all of which would create some pressure for the UK to retain its nuclear weapons as a hedge against a future Russian threat. Overall, I rate the probability of success here to be around 10%.
Third, we have the probability of persuading France to denuclearize conditional on Russia and the US agreeing to denuclearize. The same considerations discussed with respect to the UK apply here. Overall, I rate the probability of success here to be around 10%.
Fourth, we have the probability of persuading Russia to denuclearize conditional on NATO and China agreeing to denuclearize. This is extremely unlikely, given that Russia is in a position of conventional military inferiority relative to NATO, and would (in its view, fairly or not) be risking potential NATO military intervention within its sphere of influence (e.g. Ukraine, Belarus, Georgia) or invasion of Russia itself, if it did not have its nuclear deterrence. Further, the Russians do have to worry to some extent about the Chinese, who are aggressive in expanding their self-interest and who they do not trust not to take advantage of them. Overall, I rate the probability of success here to be around 1%.
Fifth, we have the probability of persuading China to denuclearize conditional on the US and India agreeing to denuclearize. China is currently still in a position of conventional inferiority relative to the United States, and nuclear deterrence is key to potentially warding the Americans off from intervening and supporting Taiwan when China attempts reunification – which is of course a key objective for them given the degree of nationalistic fervour over Taiwan being part of China, and given the CCP's legitimacy being tied up in national restoration. On the India side of things, China would be fairly comfortable with denuclearization given its conventional superiority; indeed, denuclearization would give it the chance to annex Indian territory without risking nuclear retaliation. Overall, I rate the probability of success here to be around 1%.
Sixth, we have the probability of persuading India to denuclearize conditional on Pakistan and China agreeing to denuclearize. On the Pakistan side, India holds comfortable conventional superiority, and any denuclearization would indeed favour it given that it would hence be able to take the whole of Kashmir or indeed more Pakistani territory without fear of nuclear retaliation. On the China side, however, India is in the opposite position of military inferiority, and would not give up its ability to deter significant conventional aggression on its northern border. Overall, I rate the probability of success here to be around 1%.
Seventh, we have the probability of persuading Pakistan to denuclearize conditional on India agreeing to denuclearize. This is extremely unlikely, and then some. Pakistan is in a position of conventional inferiority, and without its nuclear weapons to deter conventional aggression, it would fear not just losing a war, but potentially the annexation of the whole country in the (perceived) manner of the Bangladeshi War of Liberation, or indeed ethnic cleansing in the manner of the Partition of India or other anti-Muslim pogroms within India (c.f. Gujarat). Overall, I rate the probability of success here to be around 0.1%.
Finally, we have to consider the probability of coordinating various persuasion efforts such that they occur around the same time frame. This would be necessary, because in-principle agreement for denuclearization cannot be obtained without at least some material expression of interest by potential foes in denuclearization. That said, while such coordination would be difficult, the problem is not insurmountable. Overall , I rate the probability of success here to be around 10%.
Multiplying this through, this puts the inside view probability of persuading states to denuclearize at 0.00000000001% – vanishingly low.
Moving on to the outside view, I use three reference classes.
The first is the failure of the International Campaign to Abolish Nuclear Weapons (ICAN) to persuade any of the existing nuclear powers to denuclearize, even as it played no role in the denuclearizing of the four previous nuclear powers that did actually denuclearize (i.e. a literal 0% success rate). ICAN is a useful benchmark here, because it is probably the most successful nuclear disarmament NGO to date – it contributed to the ratification of the Treaty on the Prohibition of Nuclear Weapons (TPNW), and won a Nobel Peace Prize in the process.
The second reference class is the success of negotiations (starting from 1980) in eventually bringing the Chemical Weapons Convention into force (in 1997). I calculate the success rate (i.e. 5.55% chance recurring) by taking the year in which ratification occured, including by the key powers of the US and USSR, and divide through by the years in which they could have ratified whether they actually did or not.
The third reference class is the success of the 1968 British proposal that eventually led to the 1975 Biological Weapons Convention, to which both the US and USSR were parties. I calculate the success rate (i.e. 12.5%) by taking the years in which ratification occured, including by the key powers of the US and USSR, and divide through by the years in which they could have ratified whether they actually did or not.
Then, I take a weighted average of the three reference class probabilities (0.000018%), with a far higher weight placed on the actual nuclear case study compared to the chemical and biological ones. This is for three reasons. Firstly, security considerations militate against denuclearization in a way they do not for chemical and biological weapons, making the weapons control efforts for the latter two areas unrepresentative of success on the nuclear front. Secondly, there is almost certainly endogeneity insofar as the biological and chemical arms control proposals are made in conditions where they are perceived to stand some chance of success, such that the measured rate of success would be overstated relative to their true probability of success all things equal. Third, the more pessimistic view seems generally right, per Luisa's assessment that it is unlikely that countries that are non-compliant with the TPNW will ratify it, or that TPNW supporters will be influenced not to pursue, host or manufacture nuclear weapons or to join a nuclear weapons alliance – such that existing anti-nuclear efforts are ineffective, and the upshot of which is that progress is fundamentally intractable.
We can now proceed to adjust the outside view with the inside view – I give more weight to the outside view than the inside view, given that the inside view is subject to the usual worries about inferential uncertainty, thus yielding an overall estimate of the probability of success of 0.000016%
Meanwhile, in terms of how much money a nonprofit lobbying for denuclearization would require for its operations, we can again resort to both an outside and inside view. The outside view is that ICAN has expenses of around USD 2,000,000 per annum, and assuming 10 years to success by an EA charity, that translates to USD 20,000,000 in total. The inside view is thus – my sense is that you would minimally require a team of maybe 10 people working across multiple countries full time, and that they would need perhaps 10 years or so to succeed. Assuming funding of around USD 50,000 per team member per annum, in line with past CE incubatee expenditures (covering not just salaries but other costs like office rental, lobbying expenditures etc), that translates to USD 5,000,000 in total. Combining these two estimates – equally, since the usual worries over inferential uncertainties for the inside view is counterbalanced by the fact that an EA or EA-identified organization will almost certainly be more cost-effective than ICAN – we get an estimate that such an EA nuclear disarmament organization would probably need no less than USD 12,500,000.
And all in all, this means that the proportion of problem solved per additional USD 100,000 spent is 0.000000001.
Marginal Expected Value of Lobbying for Denuclearization
Combining everything, the marginal expected value of lobbying for denuclearization is 248 DALYs per USD 100,000 spent, making this only around 40% as cost-effective as a GiveWell top charity.
I'm curious about the ethical decisions you've made in this report. What's your justification for evaluating current lives lost? I'd be far more interested in cause-X research that considers a variety of worldviews, e.g. a number of different ways of evaluating the medium or long-term consequences of interventions.
I think the issue of worldview diversification is a good one, and coincidentally something I was discussing with Sam the other day - though I think he was more interested in seeing how various short-termist stuff compare to each other on non-utilitarian views, as opposed to, say, how different longtermist causes compare when you accept the person affecting view vs not.
So with respect to the issue of focusing on current lives lost (I take this to mean the issue of focusing on actual rather than potential lives, while also making the simplifying assumption that population doesn't change too much over time) - at a practical level, I'm more concerned with trying to get a sense of the comparative cost-effectiveness of various causes (assuming certain normative and epistemic assumptions), so worldview diversification is taking a backseat for now.
Nonetheless, would be interested in hearing your thoughts about this issue, and on cause prioritization more generally (e.g. the right research methodology to use, what causes you think are being neglected etc). If you don't mind, I'll drop you an email, and we can chat more at length?
Sure, happy to chat about this!
Roughly I think that you are currently not really calculating cost-effectiveness. That is, whether you're giving out malaria nets or preventing nuclear war, almost all of the effects of your actions will be affecting people in the future.
To clarify, by "future" I don't necessarily mean "long run future". Where you put that bar is a fascinating question. But focusing on current lives lost seems to approximately ignore most of the (positive or negative) value, so I expect your estimates to not be capturing much about what matters.
(You've probably seen this talk by Greaves, but flagging it in case you haven't! Sam isn't a huge fan, I think in part because Greaves reinvents a bunch of stuff that non-philosophers have already thought a bunch about, but I think it's a good intro to the problem overall anyway.)
On the cluelessness issue - to be honest, I don't find myself that bothered, insofar as it's just the standard epistemic objection to utilitarianism, and if (a) we make a good faith effort to estimate the effects that can reasonably be estimated, and (b) have symmetric expectations as to long term value (I think Greaves has written on the indifference solution before, but it's been some time), existing CEAs would still yield be a reasonably accurate signpost to maximization.
Happy to chat more on this, and also to get your views on research methodology in general - will drop you an email, then!
I agree with (a). I disagree that (b) is true! And as a result I disagree that existing CEAs give you an accurate signpost.
Why is (b) untrue? Well, we do have some information about the future, so it seems extremely unlikely that you won't be able to have any indication as to the sign of your actions, if you do (a) reasonably well.
Again, I don't purely mean this from an extreme longtermist perspective (although I would certainly be interested in longtermist analyses given my personal ethics). For example, simply thinking about population changes in the above report would be one way to move in this direction. Other possibilities include thinking about the effects of GHW interventions on long-term trajectories, like growth in developing countries (and that these effects may dominate short-term effects like DALYs averted for the very best interventions). I haven't thought much about what other things you'd want to measure to make these estimates, but I would love to see someone try, and it seems pretty crucial if you're going to be doing accurate CEAs.
Are you gonna include all these values in a big table somewhere?
All values are listed within the CEA itself, as linked to in the summary - it's probably easier to follow there, rather than in the writeup!
Sorry but is there gonna be an easy way to see the comparisons of different CEAs you do?
Apologies if I'm misunderstanding, but if you're referring to comparing the headline results of various CEAS (e.g. nuclear war, fungal disease, asteroids, future topics etc), they'll all be listed here (https://exploratory-altruism.org/research/). Once the list gets longer, I'll probably work to put everything into a single excel/google sheet for easier comparison.
That sounds great, thanks
Worth flagging that I believe this is based on a (loosely speaking) person-affecting view (mentioned in Joel and Ben's back-and-forth below). That seems to me to bias the cost-effectiveness of anything that poses a sizable extinction risk dramatically downward.
At the same time, I find both the empirical work and the inside-view thinking here very impressive for a week's work, and it seems like even those without a person-affecting view can learn a lot from this.
Wow, many thanks, quite an eye-opener though I'm quite new to the literature on this question and to the forum itself!
Just a worry about the modelling which may have been taken care of anyway through the mention of inferential uncertainty (I haven't checked the definition of this so far).
So, here goes: I wonder whether the inclusion of wars of more than say 20 years ago in the calculations (for example for the conventional war injuries) are pertinent, since the conditions then were significantly different. More specifically, for example, there was no internet and no possibility of cyber-warfare and the world was a far less interconnected place.
More generally, I wonder whether even considering the last ten or twenty years or even two years the political conditions are such that render each conflict a sui generis event, and whether this should be a worry for any modeller.
I think these are fair points, and in particular I'm worried about the reliance on Korean War data to model US-China conflict - if I had more time, I would go look at the expected deaths in a Taiwan conflict, but there aren't any really available as far as I can tell.
From a bigger picture perspective, all this probably doesn't matter too much, insofar as the costs of more fatalities/casualties from more conventional war get swamped by the benefits of reduced nuclear risk anyway.
yeah, I agree that what's most important is the bigger picture, and regarding that, I totally agree with your conclusions!
(though, coming just after listening to Ian Morris's podcast on his history book (here: https://80000hours.org/podcast/episodes/ian-morris-big-picture-history/) I just got the thought that politics may dictate a quantification also for the scenaria for nuclear conflict as politically speaking, use of a nuclear weapon by say Russia would be qualitatively different to use by say Pakistan because of (for lack of better words) there's a 'hierarchy' of states judged on their power on the world stage. Something which I think it would be hard for a model to capture. But I may be wrong and you have covered this, and as I said before, I do agree with your conclusions and estimates