Book review: The Doomsday Machine
The Doomsday Machine by Daniel Ellsberg:
…is a book about designing the end of the world as we know it, chronologically through Daniel Ellsberg’s career as a nuclear war planner. It’s well written, and Ellsberg makes a compelling hero. He’s most famous for leaking the Pentagon Papers, government documents on the Vietnam War that contributed to Nixon’s resignation.
This book came out of a second set of documents he photocopied and intended to release after his trial for the Pentagon Papers, but lost in an act of nature. Early on, he describes this second planned leak as the one that he fully expected to put him in jail for the rest of his life, and how he felt the loss of those documents as both a tragedy for the nation, and a blessing that allowed him to spend the following decades beside his wife. It’s the kind of thing that makes you glad you’re driving alone when your audiobook is making you tear up in the desert along the Washington-Idaho border.
But all this just helps – the real meat of the book is in the systems he describes.
Let’s talk about nuclear winter real quick. (My favorite line on dates.) Ellsberg puts this at the end, which makes sense chronologically, but it’d be burying the lede for an x-risk focused blog, so let’s get it out there now:
All of our plans for cold war were decided before anyone knew about nuclear winter. I feel like I should capitalize that – Nuclear Winter. It’s the hypothesized event where nuclear explosions cause fired in cities that launch so much ash into the stratosphere that it blots out the sun for months and makes it impossible for plants to grow, killing most human and large animal life. There’s uncertainty around the specifics, but its existence is generally agreed upon in the scientific community.
All US strategy during the early Cold War hinged on this idea of “general war”, an all-out nuclear exchange with Russia and China. General War included dropping enormous nuclear weapons on literally every single city in both Russia and China. Obviously, this is atrocious enough – this level of calamity was expected to kill something like 20% of the world’s population at the time, mostly from fallout.
But every time general war was mentioned, a little voice in my head yelled “nuclear winter!” – that the death toll is actually >90% of humanity, Americans, Russians, Chinese, and everyone else alike, unbeknownst to everybody at the time. My loose impression is that there’s not substantial reason to believe that nuclear war planning policy ever shifted to account for this fact.
Another quick takeaway: the US planned on making the first nuclear strike on Russia and China throughout the Cold War. Today we have a perception that the US only plans for using a second strike, but almost the entirety of planning material is based on the supposition of the US using nuclear weapons first. Again, there’s little reason to suspect this has changed now.
Through this book, I was repeatedly reminded of the Litany of Jai: Almost nobody is evil, almost everything is broken. The problems described in the book aren’t the result of insanity or complete carelessness, but instead a horrifying spider web of incentives, laid unwittingly by people with limited goals and limited knowledge. It’s a sinister net of multipolar traps. If you follow this web down, as Ellsberg does, you find yourself looking into the yawning chasm of a nuclear apocalypse – not built on purpose, but built nonetheless.
Let’s look at how some of these tangled incentives lead us there.
Branches of the military want high budgets.
Budget decisions are made based on intelligence from those branches.
Branches compete with each other for funding from Congress and other officials.
↓
Various branches hugely overestimate enemy capacities.
E.g. the army reports extremely high Soviet ground force numbers.
The Air Force reports extremely high Soviet ICBM capacity.
Inter-branch coordination gets trampled.
There is no incentive for estimates or behavior that aligns with strategy or reality.
All military branches want to get in all-out war if/when it happens.
↓
The Pacific Navy basically insists on attacking Asia alongside Russia in all cases, because they want to be involved and don’t just want to attack minor Siberian targets “on the sidelines” of The Big War.
Nuclear plans have Moscow area getting blanketed with hundreds of nuclear bombs from all sides. “Hundreds of nuclear bombs” is a phrase that here and elsewhere means “calamitous overkill”.
Military branches don’t want to listen to civilian politicians.
Civilian politicians are powerful decision-makers.
↓
Information is concealed, including from the president (for instance, the JSCP, which is the detailed plan for all-out war).
Military leaders just don’t listen to civilians who outrank them (e.g. in moving ships with nuclear warheads illegally stationed in Japanese ports).
Civilian President Kennedy is politically obliged not to override poor decisions made by President Eisenhower, the famous military general.
Nuclear bomber pilots need to receive an authorized signal to enact plans for bombing Russian and Asian targets.
Air force planners want as little delay as possible in executing war plans once they get the order.
Air Force planners want to save time and effort.
↓
Authorization codes are stored in plaintext in envelopes in each plane, are the same between every plane, and are rarely changed.
Any pilot who realized this could easily lead their base in a nuclear strike, and almost certainly trigger all-out nuclear war.
There’s no way in the target database to easily distinguish Russian and Chinese targets, so everyone at Air Force bases assumes that if they get the war order, they’ll just drop nuclear weapons on everyone. All Chinese cities were going to be destroyed under every nuclear attack plan, throughout the entire early Cold War.
Communications systems with Washington DC might be destroyed if Russia attacks the US with nuclear weapons first.
Communication systems between bases might be destroyed during a Russian attack.
Communications in general are pretty unreliable.
Everyone in the military chain of command, including the President, wants the US to be able to respond as quickly as possible to a Russian first strike.
↓
Ability to initiate a nuclear war is secretly delegated down the chain of command in cases where bases are not in touch with Washington DC.
Contact with Washington DC is often unreliable – for hours every day on some bases in the Pacific.
Basically anyone in the chain of command is not just capable of, but entirely authorized to, declare total nuclear war most of the time.
This are not even every example. A story retold in many different forms throughout the entire book goes like this:
Daniel Ellsberg learns about one of these outcomes.
Ellsberg talks to some relevant officials and outlines a possible catastrophe.
The officials go still, think about it, and say with concern, “That seems entirely possible.”
Nothing changes, ever.
A possible solution for most of these spiraling incentives is a countervailing force, balancing the dynamic back away from “total catastrophe”. An actor, or an incentive, or something. Often, that does not exist – in the veil of secrecy surrounding nuclear war, any party with an incentive to care about the implied risk isn’t aware of the entire situation, and can’t unilaterally fix it if it exists. Ellsberg tries to repair these flawed systems when he notices them, but has little power to do so.
He talks about how he suspects that some leaders, including President Kennedy, never had real intentions of using nuclear weapons, but even if that’s true, the scenarios above suggest that presidential intent may have had little to do with the outcome.
Ellsberg’s knowledge of the situation drops off in the 70’s or so when he started doing other work. Are all of these nuclear war and control systems still like this?? Maybe??!! Certainly nobody was rushing to reform them throughout his long tenure with the government.
I don’t know what to do about any of this. This book illuminates the number of needles we somehow threaded in avoiding catastrophe since the start of the Cold War. Here’s where you can get it.
John F. Kennedy was a decorated military officer who served in World War II. He was vastly outranked by Eisenhower, of course, but it’s still frightening to imagine a political system where a President’s implicit authority on defense issues was so strongly related to their military rank that even another President who served in wartime couldn’t override them.
--
Also, having read the book, do you have any thoughts on how we got so lucky?
There’s always anthropic reasoning, but given the number of terrible systems in place and the non-disastrous result, I wonder whether there were more anti-war incentives than we realized.
This could be at a system level; for example, maybe we don’t hear about the military’s unusually good security practices, only the bad ones. Or it could be at the individual level; even if military leaders didn’t know they were at risk of actually destroying the world, “not killing millions of people” might still be a more powerful motivation than we realize, leading to increased personal caution and a general reluctance to escalate. Maybe we’d have been safe even with a few more Petrov-like incidents, because almost no one would ever take that final step...
...or maybe we did just get lucky. Any thoughts?
There has been too little focus on nuclear risks compared to importance in EA, and I think that this post helps ameliorate that. In addition to the book itself, which was worth highlighting, the review also functions as a good review of how systemic incentives create issues up to and including existential risks, and allows us to think about how to address them.
I do recall Ellsberg claiming that the US plans were based on striking first. But interestingly, in a summary of key points from another book on nuclear war (The Dead Hand), Kit writes:
So now I feel unsure whether to trust Ellsberg or trust The Dead Hand (assuming Kit’s summary accurately reflects its claims). (Or maybe there’s some way to reconcile the two sets of claims which I’m not currently seeing?)
I’m not an expert on the topic and don’t have sources on hand that would make the argument in greater detail, but I did take a course on ‘The global nuclear regime’ (broadly about institutional developments surrounding nuclear material and weapons control since 1945) and based on my knowledge from that, I’d suggest that there is a way to reconcile the two sets of claims.
First, I think it’s important to distinguish between ‘surprise attack’ and ‘first strike’. The former is obviously a subset of the latter, but there are also other conceivable kinds of first strike attacks. Surprise attack, to me, sounds like an attack that is launched without an immediate trigger, with the purpose of hitting (and eliminating or severly weakening) an adversary unexpectedly. A nuclear first strike might, instead, be considered in a situation where a conflict is escalading to a point that a nuclear strike by the other party seems to be growing more likely. It might be considered as an instrument to prevent the other party from launching their missiles by hitting them first (e.g. because the costs of waiting for them to launch before counter-striking are considered unacceptable). This comes down to definitions, ultimately, but I think I wouldn’t describe such a first strike as a surprise attack.
Second, there is not necessarily a contradiction between there being plans for first- rather than second-strike attacks and US officials expressing doubts about the USSR’s belief in US willingness to actually conduct a first strike. The US figures you mention might have thought that in that moment, the likelihood of a US first-strike was really low and that hence it would’ve been surprising for the USSR to start the detection project at that moment. These US figures might also have been disingenious or biased when assessing the honesty of the USSR leadership (I would argue that the tendency to attach hidden, often propagandistics, motives to ‘enemy leaders’ - without strong evidence base or even a coherent plausibility argument as support—is fairly common among US ‘hawks’). Debending on who the key US figures mentioned in your summary are (unfortunately, I haven’t read The Dead Hand), it might also be that they just weren’t aware of the first strike plans of the US. Lastly (and I don’t consider that one super likely), it might be that the US figures just thought that the Soviet leadership wouldn’t expect a US first strike in spite of the plans for it (either because the Soviets didn’t know about the plans, or because they didn’t think the US was likely to act on them).
Good points, thanks!
I’d guess that this is probably the main thing that explains the contrasts between the two quotes.
I doubt that this is can help much in explaining the contrast, since the quote from Kit’s summary of The Dead Hand sounds like it applies across a large time period, rather than just at a smaller handful of key points. (Though of course, Kit’s summary may be very slightly misleading on certain nuanced points.)
I also don’t think this is likely to be a major part of the explanation, as I think it was pretty well known among US military and political leaders that the US had first strike plans. I’m not sure of this, though.
[Arriving very late to the party]
I also just finished reading this book, and likewise found it quite interesting. Though I do find myself somewhat skeptical of some of the claims made in the book. E.g., as you note, there’s uncertainty around the specifics of nuclear winter—yet I seem to recall Ellsberg writing as if this wasn’t a topic of major debate, and as if it was near-certain that there’s be an extremely catastrophic nuclear winter in a wide range of nuclear exchanges.
Related to the uncertainty about the likelihood and impacts of nuclear winter conditional on nuclear war, I think “the death toll is actually 90% of humanity” may overstate things. That may have been a likely death toll (conditional on full-scale nuclear war) during some periods of the Cold War, when nuclear arsenals were much larger. But I think that was probably unlikely (though plausible) both before and after those periods. (Some analysis on this here. Also, of course, even just a relatively small chance of a >90% death toll can still be a very big deal!)