Once More, Without Feeling

Fermi Estimate

I’ve created a spreadsheet that estimates the impact of various interventions on COVID-19.  My estimate is on the “Stuff I Sorta Think” tab.  Feel free to copy the spreadsheet and fill out the blue cells in the “Fill Me Out!” tab to create your own estimate.

Note that the numbers I put there are just estimates and many are not very well researched.  But I suspect that the conclusion is clear enough that it won’t have a huge impact.


Consider filling out the spreadsheet for yourself before looking at mine!

The conclusions of mine were roughly similar with my previous posts on COVID-19.  I found that doing nothing was massively better than various interventions, with the majority of the cost of doing nothing coming from the extent to which people don’t follow the plan and shut down things anyway.


I’ve already written a lot of what I have to say on this.  But to summarize:


  1. By far the dominant factor here comes from the economic impact of shutting down the economy.
  2. Compared to that, nothing else really matters, so the first order answer is “don’t shut anything down”.
  3. The second order corrections are mostly just mirrors of the first order fact.  Addressing it early is better than waiting and then shutting things down, primarily because it involves shutting things down for less time.  But it’s still significantly worse than doing nothing.
  4. The world’s current strategy is the really bad: wait until it spreads, and then mostly shut things down but not so completely or early enough that you actually stop the spread of the disease.
  5. IFR is really low, especially for young people–probably below 0.01%.
  6. The true mortality rate of COVID-19 might actually be comparable to the flu, though we don’t have enough precision to really know.
  7. In fact the IFR estimates aren’t substantially above background mortality rate.  I don’t actually think COVID-19 has no impact on mortality, but it is within error bars.
  8. The cost of shutting things down is MASSIVE.  All-told my estimate is a few hundred trillion dollars!  I’m guessing the last time the world made as large a mistake as freaking out about COVID-19 was WWII.
  9. By and large this is using a “generally make the world better” lens and not a particularly longtermist one.  That makes sense if you think the main flow-through effect of what we do about COVID-19 is generically how good the world is.  If instead you think it’s something specific about the impact this has on the odds of a future x-risk event, it probably makes sense to do a deep-dive on whether the “get used to dealing with pandemics” argument or “boy who cried wolf” argument is stronger from the perspective of preventing future bio-terrorism-related x-risk events.
  10. If you do think the main factor here is bio-terrorism then I think it makes sense to just focus on that and ignore all the other factors here; thinks like GDP and IFR probably don’t have huge impacts on that.

The Cracks Begin To Show

[For my previous COVID-19 posts, see here and here.]

For a while my social media was filled with nothing but COIVD-19 posts.  Well, it’s still all coronavirus, but the nature of the posts is slowly morphing.

To be clear, the majority of the posts are still the same; “As Coronavirus Cases in US Spike 40% in One Day, Medical Workers Plea for Key Supplies” bemoans a typical article.

But a new strain is seeping back into the national conversation after a few week hiatus.  Often it just appears as a scapegoat–“Don’t believe the myth that we must sacrifice lives to save the economy”, writes The Guardian.  But the fact that if even needs to be posted shows a growing worry among some that the coronavirus narrative might be in danger of slipping out of their grasp.  Few are bold enough to quite say it yet, but some toe in that direction.  The generally raise the question of what a risk-reward tradeoff would look like, but stop short of answer it.  Trump, ever the tactful speaker, spoke confidently about packing Easter pews as we move on from dealing with COVID-19.

This new crop of thought, that maybe, I don’t know, if you look at the cost-benefit analysis, I mean, like, well, obviously it would be horrible to suggest, you know, but….  It’s not coming out of nowhere.

The collective confidence in taking extreme measures to combat COVID-19 has always seemed a bit quixotic.  First, it was important to do things to stop the spread.  Then, the narrative seamless turned over to flattening the curve–accepting that shutting down the economy wouldn’t stop COVID-19 from spreading but suggesting we do it anyway–without missing a beat.  The arguments for extreme suppression have always been light on the numbers and facts, other than pointing to an exponential graph and saying “look it gets big!”  Instead, they’ve been scolding, self-riteous, and condescending.  “I’m A Doctor. The U.S. Response To Coronavirus Has Been Nothing Short Of Criminal.”, writes HUFFPOST.  Any tradeoffs between approaches are a “myth”.

The Washington Post runs an op-ed: “The media must stop live-broadcasting Trump’s dangerous, destructive coronavirus briefings”.  The institution’s new slogan looms overhead.


There remains a decent argument for the world trying to coordinate on taking the Singapore/Taiwan/Hong Kong approach–half-shut-down for a week, tons of quarantine for possible cases and good hygiene, and then let life go on; by aggressively using the most cost effective measures early, we may have been able to kill off COVID-19 at a reasonable price.

But each day the case for suppression weakens, and as the media hunkers down into its collective hard-line stance the suppression movement threatens to jump the shark.  NYC mayor Bill de Blasio calls for more suppression while saying that over half of the city will get it–defeating the main purpose of the lockdown.  While earlier arguing that the economic hit wouldn’t be that bad, pundits now have to grapple with historically large jobless claims, stock market crashes, and forcasted GDP decreases; a $100T loss now looks like if anything optimistic given the dire forcasts banks are pushing out.  China gave up on its lockdown after a month and cases are coming back; even Hong Kong, Taiwan, and Singapore and starting to shut down to prevent imported cases from running rampant, making true suppression look more and more like a pipe dream for most of the world.

Meanwhile the evidence mounts that the expected DALY loss from COVID-19 is at the low end of estimates.  Iceland leads the world in tests per capita, but more importantly it’s done randomized testing.  This lets us address one of the central uncertainties around COVID-19: what’s IFR, the infection fatality rate?  We know roughly how many deaths there are but need to divide by the total number of cases.  Well, in Iceland, 48 tests out of 5571 have been positive, for a rate of about 0.86%.  The population of Iceland is about 360k, meaning that roughly 3,600 people likely have COVID-19 in Iceland.  There has been one death, with another twelve in the hospital, giving an IFR somewhere between 0.03% and 0.33%, way lower than the CFR-based estimates floating around.  South Korea’s tests might not be random–so take this with a grain of salt–but the same math there would imply that around 2.5% of the population has COVID-19; 131 deaths out of ~1m people would imply an IFR around 0.01%.    Korea and Italy have both yet to have a death below the age of 30, bounding CFR at less than 0.10% for young people.  Given that IFR generally seems to be a factor of 10 below CFR, a < 0.01% IFR for people under 30 would put their mortality rate over the last few months in line with background mortality; it’s not even statistically significant that COVID-19 increases death rate for young people.  The average age of death from COVID-19 remains in the 70s and 80s in various populations.  Put all of this together–a young IFR below 0.01%, an average age of death around 75-80, and an overall IFR around 0.10%–and the average DALYs lost per infected person from possible death by COVID-19 is around 5 days.  It’s worth noting here that these IFR estimates are in line with those of the flu.

The math looked extremely lopsided a week ago; with IFR estimates coming down more and economic costs continuing to increase, it’s not close to being close.

And I suspect that world leaders are beginning to realize this–or perhaps knew it all along.  Trump is targeting a few weeks for lifting all social distancing in the US; China reopened after a month or so, restarting its COVID-19 spreading; and the UK fought hard to stay open before decisively losing the PR battle.  The world has pumped around $7T into the economy via stimuluses–an impressive amount, if still an order of magnitude below what’s needed (never mind the impact on global debt or inflation).  The truth is that not only is it a bad decision by orders of magnitude to shut down, but countries just can’t shut down for more than a month or so without sinking into a possibly fatal cycle of recession and debt.

We’re all used to seeing a lot of poorly thought out shit on social media.  What’s been a bit shocking to me, though, is exactly how bad the EA writing on COVID-19 has been.  It seems to be just as non-quantitative, emotional, judgmental, and wrong as the median content I see.  This really should be right up EA’s wheelhouse; I think a decent Fermi estimate does a pretty good job here.  And there were a few early Fermi estimates!  But when the assumptions they made started to seem likely to be wrong–so wrong that the conclusion flipped from pro-suppression to anti-suppression–rather than update towards being anti-suppression, EA seems to have instead kept its conclusion.

My best guess, now, is that the EA echo chamber of social media is just not representative of core EA thinking.  That, just like the rest of the world, most EAs who are anti-shutdown have decided it’s bad PR for them to say so.

Which would be comforting–at least the people who are really in charge might be right!  Except that I worry we might just be caught in a PR feedback cycle.  I don’t think it was obvious, ex-anti, that it would be political suicide to be against shutting things down.  So how did we get here?  Well, my guess is that it started off as a trickle–but the more publicly expressed opinions from thought leaders came out in favor of shutting things down, the more scared anyone who disagreed was to voice their opinion, and so we spiraled into an echo chamber.  Public opinion caused and reinforced itself, and we ended up destroying 30% of the economy because we let this spiral start.

Collateral Damage

[Edited on 2020-03-22 based on some feedback]

I wrote an earlier piece on COVID-19, arguing that the costs of intervention might outweigh the benefits.

Over the past few weeks we’ve begun to see the ramifications of both COVID-19 and the measures taken to deter it.  Overall I think the case has become significantly stronger that interventions do more harm than good, so much so that the harm done by attempting to stop the spread of COVID-19 might rank among the most destructive decisions made by society in decades.

The Cost

Let’s start by estimating the cost of our reaction.  In particular, I’m focusing on:

  • Social Distancing
  • Shutting down businesses
  • Closing borders
  • Inciting fear
  • Overwhelming hospitals

Although the majority of the focus will be on the impact of shutting down businesses.

There are a lot of threats here and many different ways to estimate each; my goal is just to get a very rough approximation.

Stock Market

Stock markets are down, a lot.  The S&P 500 is down about 27% from before COVID-19.  Other markets are down even more; non-US markets are down about 33%.  Let’s use 30% as the approximation.

Listed company market cap is around $100T, so this is a $30T loss from them.  But public companies are less than half of all profit–so we’re actually looking at something in the $50T-$100T range.

What does it mean exactly for a stock price to decrease?  Stock prices are the value of the company–not annual revenue.  They’re meant to represent the risk-adjusted net-present-value of the company, i.e. what the total future value of it is.  So loss in market cap is meant to roughly correspond 1:1 to loss in value.


Another way to approach this would be to assume that stock price declines represent a decrease in NPV of future GDP.  GDP is roughly $100T.  Using a standard interest rate of ~5% per year (to approximate inflation, interest rates, real estate mortgage calculations, etc.), that means an NPV of roughly $2,000T.  A 30% decrease in that would correspond to about $600T.

Why is this so much larger than the first estimate?  I think the answer is that profit is less than GDP: in addition to making things companies have costs.  So companies are roughly leveraged GDP: super long GDP against short costs.  That means that the anticipated % decrease in GDP should be less than the % decrease in stocks.

Generally recessions seem to imply a roughly 5% hit to GDP (split between a few quarters).  That would instead give us an estimate of $40T-$100T–back in the same range as our first one.

Costs of Quarantine

Social Distancing has lots of negative effects other than just reduced economic output.

One strong predictor of lifespan is social network; social distancing certainly can’t help there.  I’d guess we’d see a significant uptick in depression and anxiety.

There’s also likely to be exercise, which should have a similar impact.

Furthermore, shutting down the world for a few months turns that stretch of people’s lives into a non-productive, socially isolating, stressful period–which is itself a ~2 month cost.

International Tension

COVID-19 has probably been bad for international relations.  There’s animosity between countries, closed borders, and general tension and stress; this likely increases the odds of international conflict and x-risk.  I’m not really sure how to estimate that.  I think a 0.5% increase in x-risk seems in the right ballpark.

Summary Of Costs

Putting this all together, the costs of shutting down the world because of COVID-19 seem to be on the order of $75T + 2 months + 0.5% increase in x-risk.



What do we buy ourselves by reacting to COVID-19?

COVID-19 Left Alone

To get a baseline: what would happen if the world treated COVID-19 the same as the seasonal flu?

Numbers here are all over the place, ranging orders of magnitude.  In general, though, the following seem to be true:

  • The average age of death is just over 80 years old
  • The mortality rate for healthy young people is something like 0.03%
  • The overall mortality rate, if there are no control measures and hospitals get overwhelmed, is likely to be in the range of 1%.

Death Rate Among Young Healthy People

The central challenge here is estimating how many people have actually gotten COVID-19.  We can then divide the deaths by that.  However, many people have gotten COVID-19 but not reported it, either because they’re asymptomatic or because it’s mild enough that they don’t need to go to a hospital.

This study claims that for people under 30:

  • 1.2% of symptomatic cases require hospitalization
  • 5% of hospitalizations are severe
  • 0.03% of the infected die

If we assume that about 30% of cases are symptomatic the first two bullet points would imply 30%*1.2%*5% = 0.02% of infections are fatal, lining up with the last bullet point.

For what it’s worth, this is in line with the fact that in Italy and Korea, there have been zero deaths under 30, among roughly 10,000 reported cases in each; that puts an upper bound on the rate of roughly 0.10%.

Other studies attempting to estimate death rate among young healthy people have been seriously confounded by the underreporting of non-life-threatening cases.

There are pretty large error bars on the death rate–probably a factor of 4 or so.  But one thing worth noting: the overall death rate for people in their 20’s is about 0.1% per year.  If one reported all deaths from someone with COVID-19 as COVID dealths, and could detect COVID-19 in peoples’ bodies for about 30 days, then that would result in about 0.01% of infected people in their 20’s showing up as dying from COVID-19.  Again, pretty wide error bars–but in the median case this explains half of all COVID-19 related deaths for young people–deaths from other things reported as COVID-19 deaths.

Death Rate Overall

Right now the overall deaths / reported cases is around 4%.  This is likely to be a significant overestimate, though, because many cases go unreported–mostly non-fatal ones.

In general, almost all of the country-by-country estimates are going to be higher than the true deaths / infections, but by looking at the lower bound of countries we should get a sense, at least, of deaths / symptomatic cases.

The fatality rates among countries with more than 1,000 cases range from a high of 8% in Italy to a low of 0.27% in Austria.  Among high-profile countries with robust testing where COVID-19 has been around a while, the low end seems to be around 1% in South Korea, USA, and Canada.  China, which had significant hospital overcrowding, reported around 4%.


(data from https://www.worldometers.info/coronavirus/)

However this corresponds to a total number of world cases around 250,000.  That’s less than 1 in every 10,000 people.  This is likely a gross underestimate.  It seems like, for instance, 5% of NBA players may have COVID-19, and significant numbers of athletes in general.  These people are selected, among other things, for all getting tested.  It’s pretty hard to know the true infection rate, but I’m guessing it’s around 5-10 times the reported ones.  That means that the true death rates are probably around 0.03% – 1% depending on the country and its response.

I’m guessing that 1% is roughly the correct number for deaths / total cases (not just reported or symptomatic ones) if there is basically no healthcare.  But again, this will skew heavily in age.


What’s the DALY impact if COVID-19 were to spread uninhibited?

I’m guessing we’d look at an overall fatality rate of roughly 1%, with an average age of roughly 80 and a young/healthy fatality rate around 0.03%.  Assuming the average person lives to 90, this would be an average life expectancy hit of 1%*10, or about 0.1 years per person.  Given the selection effects on health, it’s likely to be a marginally smaller DALY hit–maybe something like 0.05 years–about 0.05% of a life, or a few weeks.

Again, these numbers aren’t exact, but they’re the right ballpark.

Lost Time to COVID-19

Even if COVID-19 isn’t deadly, it still costs people a week of their life.

Economic Impact

I think that this is mostly rolled into the life cost and that also counting for the lost working time would be double-counting.  Given the demographics involved I also think the economic cost is going to be more in the “time lost to COVID-19” section than any others; so feel free to add another week of cost per person who gets it if you think it won’t double count.

In Practice

So the difference between completing curing COVID-19, and completely letting it run its course, is about 0.05% of lifespan plus 1 week–up to 0.07% or so

In practice, we won’t hit either extreme.  For one thing even if we do go all-out in trying to prevent the spread of COVID-19, we might fail; enough people might disregard the controls that eventually it keeps growing.  We also likely won’t hit 100% of people infected no matter what.

I think a reasonable estimate here is that, if we compare a more consistently applied version of the current world’s response to just not responding at all, we’re likely looking at a swing of roughly 25% of the population.  So taking the DALY hit and dividing by 4 we get roughly 0.02% that we’re buying ourselves with the reaction.


The Tradeoff

So how does the tradeoff look?

On the one hand, it’s $75T + 2 months of life + 0.5% x-risk.  On the other hand, it’s 0.02% of DALYs–roughly two weeks.

Well, that’s not very close.


We have to have some way to convert all of these into each other.  If we just scale x-risk to current population levels then that’s another 6 months or so per person.

If we look at the US as an example, it’s likely about $15T hit to the US, which would be around $50k per person–roughly the corresponding to 1 year of VSL.  (It’s not clear exactly what the correct VSL to use here is; given the demographic in question I’m guessing $5m is the right ballpark.)  Other countries are likely to see similar hits measured in GDP % decrease * GDP per capita / VSL.

So I’m guessing that the total here is roughly 1-2 years per person, or about $150T.

On the other side, we have two weeks per person, or about $3T.

Summary of Tradeoffs

So I think the two sides here are off by roughly a factor of 50.  The cost to the world’s current approach, compared to just ignoring COVID-19, seems to be two orders of magnitude higher than the benefits.


In the end, I think the decision isn’t close.  The world is making a mistake by reacting as strongly as it is to COVID-19.

One could argue that the best approach would instead have been to do what Hong Kong, Singapore, and Taiwan did: to stop the spread early and aggressively, and then relax a bit.

But I think that’s missing the forest for the trees.  If the cost of intervention is 100 times the cost of the disease, then the advantages of early action are just a distraction from what really matters, which is not overreacting.

These numbers are all really rough, and I don’t fully believe them.  But I do think they’re close enough to convey the correct message.

The health impact of COVID-19 will be relatively small.  The collateral damage done by the world’s overreaction to it looks like it will be about as bad as the 2008 financial crisis, and much worse than nearly any other thing to happen to the world since World War 2.

I think this has some pretty damning implications.  I think that generally promoting social distancing is extremely damaging to the world, and that governments are creating a recession by their response to COVID-19, and that the most detested responses are, while unempathetic, probably correct.

What Are People Thinking?

This is starkly different from the prevailing narrative.  I think the prevailing narrative is mostly just wrong.  I think there’s been a huge positive feedback loop: once it became clear that the politically correct response was to mandate extreme anti-COVID measures saying otherwise became taboo.

I think one of the most compelling cases here is “flattening the curve”.  As far as I can tell, the idea is really bad.  Essentially it involves accepting that most people will get COVID-19, and then closing down the entire country for months anyway, and letting the nightmare extent potentially for years, all for relatively modest wins.  It seems like an extreme case of people being backed into a corner, trying to salvage the “we should quarantine” narrative in spite of the evidence that at this point anything short of a full lockdown will be both very expensive and not very effective.

Deceptive headlines abound.  Alarms are raised because more men die than women (alternatively fewer women die), The US has a higher mortality rate than some other country (alternately that country’s rate is lower), and that young people aren’t invincible (actually they basically are according to data from South Korea, Italy, etc.).  Companies are lambasted for mandating unpaid leave even as countries are praised for mandating those companies to close their businesses.  Airlines are ridiculed for asking for bailouts from governments that made their business temporarily illegal.  We praise doctors who advocate lockdowns and blackball those advocating normalcy, and then remark that most public calls are for lockdowns.  (Off the record many epidemiologists will call the world’s response a massive overreaction, but few are stupid enough to say so publicly.)

We are constantly calling public opinion an echo chamber.  Why is the current consensus exempt?

I think there’s one more piece to this.  The “chattering class”–the people whose opinions fill the center-left space on Facebook and Twitter–tend to be upper middle class, and sit in a sweet spot where COVID-19 affects them less than just about anyone else.  It’s largely people with enough personal runway and a strong enough CV that there isn’t any real fear of serious consequences from losing a job or missing a few paychecks; simultaneously it’s people who don’t tend to have huge stakes in businesses that are tanking.

What Should We Do?

Well, my first answer is: nothing is 90% of the way to the right response, so I’d settle for that.

But in terms of effective interventions:

  1. A $1T prize for the first company to create a COVID-19 vaccine that works for at least 1 year and costs no more than $100 per person.
  2. If flattening the curve really is the way to go, a $250b package towards building more ventilators and hospital beds and training people to use them.
  3. Massive funding for hand sanitizer, masks, COVID-19 home test kits, etc.

And I do think those would probably be good!  But more important is not shutting down the world.

What Is Dead May Never Die

I think the world is overreacting to nCov, and using a sledgehammer instead of a scalpel.

I should say that the most compelling reason to think this would be if you fully understood the science and incidence and preparations. I don’t. This is instead a different sort of argument: that my prior is the response makes no sense, and that the nature of the response makes me think that it was not designed rationally.

To be clear, I’m not confident I’m right here; it’s just my best guess. But it remains my best guess even after others have disagreed.  This post will also probably come across more strongly than I actually feel!  And to be clear, I do think that we should ramp up anti-nCov measures if/when its incidence grows, though only to a moderate extent.

Furthermore, I think our reaction as a society has done an order of magnitude more damage than nCov could.

I should also make it clear that for the purposes of this, I’m only talking about young healthy people who aren’t near Wuhan. If you are over 70, in poor health, or in Hubei province, you should ignore this.


My Prior

I guess I’ll start with my prior for why I’m not too worried about it right now, for healthy people.

Currently around 4,000 people have been reported dead of nCov. If that were distributed randomly, each person would be losing on average around 0.02 days of life from nCov–about the amount of time it’ll take to read this post.

Furthermore, the cases are not random. It seems like the old and sick are way more likely to die of nCov than healthy middle aged people. This probably takes another factor of 10 or so out of the death rate for the healthy–so maybe about 5 minutes of life each.

And I’m also going to assume that you’re not in Hubei. If you are, ignore this and keep safe. But if you’re not–well, about 80% of reported deaths are from near Wuhan, so that takes us down to my prior–that nCov is right now costing the average healthy person not in Wuhan roughly 1 minute of life in expected value.

On the other hand, what’s the cost of our reaction as a society?  Well, one place to look is stock markets, which are down around 20%.  You’ll get similar results if you look at US stocks, foreign stocks, oil, or basically anything else.  And to be clear, that’s not 20% of GDP; that’s 20% of the net present economic value of the entire future that we’ve burned to the ground fighting nCov.  Even if you think nCov will eventually spread to infect 10% of the world, given the age distributions of its impact and its mortality rate the cost of just letting is spread is likely to be around 0.1% of the world–more than two orders of magnitude lower than the cost we’re on track to pay fighting it.

That’s where my prior comes from.

What follows is a lot of attempts that I’ve seen to argue that no, we should still be freaking the fuck out, and my responses to them.


Maybe Cases are Underreported

Yeah, they probably are! But how underreported?

Sure, maybe I’m off by a factor of 10–and my prior should instead be about 10 minutes (the time it takes you to read this piece). But it can’t be that much farther off.

I don’t know anyone who’s died of nCov. I don’t know anyone who’s gotten nCov, and in fact I think I only know one person who knows one person who’s gotten it. That would imply an incidence rate of roughly 0.01%, or something roughly seven times the officially reported rates. And, not surprisingly, that person recently came from central China.

And underreporting is a double-edged sword: cases are likely way more underreported than deaths (because only sicker people bother going to the hospital), and so the true death rate is likely below the reported one.

It’ll Spread

Yeah, it probably will spread.

The numbers are a bit all over the place but nCov still seems to be spreading globally, though it may have tailed off in China, at least according to some numbers.

The first thing I’ll say here is: if and when nCov does end up spreading a lot, it might then make more sense to take drastic action. It might very well be correct to ignore it now and then make extreme decisions in a month.

But this brings us back to the death rate. Currently the healthy death rate is estimated to be a fraction of a percent for people below 50, even before filtering for healthy people. My guess is that the true healthy person death rate right now is probably below 0.10%. Which means that, even if everyone got nCov, I’d probably still only loose about a month of life in expectation–meaning that I shouldn’t be spending a month of my life avoiding it!

And that’s if it maximally spreads. Almost nothing maximally spreads; the worst things generally affect ~⅓ of the population, and we don’t yet have evidence this is in that category. In expectation it’s probably more like 10% of the population, bringing the expected time lost to something like a few days.

And note that the first counterargument–that maybe cases are underreported–works against this one. If it’s true then the death rate is probably another factor of 10 or so lower than claimed, the worst case scenario is a few days, and the expected time lost is like ten hours.

We Don’t Have It Because We’re Reacting

In some ways this is definitely true–we’re quarantining people who have it, which definitely makes sense.

But again I think there’s a limit to how far this goes. Two nights a week thousands of people descend on a three block stretch of Hong Kong, packing a hundred people each into tiny rooms where they breath on each other, sweat on each other, and make out; all just miles away from China. As far as I know, no one has gotten nCov from LKF.

And, realistically, what would happen at, for instance, a conference? If you think about the types of interactions people would have–especially given that people are likely to be washing hands instead of shaking hands–I’m guessing that there’s an expected value of a fraction of a single transmission; given mortality rates, probably something less than an hour of life lost per participant.  Which isn’t to say we definitely shouldn’t cancel conferences–the cost of doing so isn’t super high.

Hospitals Will Fill Up

The newest incarnation of fear from nCov is the following set of claims:

  1. Eventually, half the world’s population is going to get nCov.
  2. The death rate is currently ~0.50%
  3. (2) is only true because of hospitals. But in a month a billion people will have it and hospitals will be overflowing; and the healthy person death rate without a hospital is 10%.
  4. Thus, 5% of the world’s healthy population will die from nCov.

On its face this is plausible, though it makes a number of claims I weakly disagree with. I’m pretty sure it’s wrong. But even more strongly, I’m pretty sure that even the people making the claim don’t really, truly believe it. Not that they’re lying, so much as that it’s an expression of fear rather than a rational position that’s been fully thought out.

Why do I think this? Because a lot of these people are damn smart, and if they really did think the above was true, then it’s pretty clear what you should do.

See, we already have an nCov vaccine, if you’re willing to give enough for it. And if you really think the mortality rate is 0.50% now but will be 5% in a month, and that you’re fairly likely to get nCov, then you should be willing to. But whenever I suggest this it’s met with abject horror; not what you’d expect for a strategy that would save someone 4.5% of their life.  And even if you think actually trying to get nCov is too extreme, this really does imply that you shouldn’t be at all trying to avoid getting it for now.

The Fatality Rate Is Higher

Some claim that the fatality rate is really higher than is claimed. I think these claims all fall apart when you look into them more, at least for healthy people.

One often cited source uses the Diamond Princess cruise, a ship that had an nCov outbreak, as a data set. The abstract, which has been cited a lot, says:

Adjusting for delay from confirmation-to-death, we estimated case and infection fatality ratios (CFR, IFR) for COVID-19 on the Diamond Princess ship as 2.3% (0.75%–5.3%) and 1.2% (0.38–2.7%). Comparing deaths onboard with expected deaths based on naive CFR estimates using China data, we estimate IFR and CFR in China to be 0.5% (95% CI: 0.2–1.2%) and 1.1% (95% CI: 0.3–2.4%) respectively.

This was a bit higher than I had previously heard. Later in the article, they give estimated fatality rates by age:

Well that’s interesting; this source sometimes cited for high fatality rates still finds only 0.20% for people under 40. Or at least that’s what nCFR says, which is their attempt to estimate expected deaths from the data. What if we look at the raw data?


There Are Long-Term Effects of Surviving nCov

Most of the claims that nCov lingers on in survivors looks at studies of SARS.  And I do think that this is a real worry!  But I find the presented data pretty uncompelling.

For one thing, SARS was about 10x as bad as nCov is per person; it’s going to be pretty hard to read into nCov long-term effects from that.  But for another, there are significant issues with the studies.  For instance this one pre-filters the list of SARS survivors for the 25% who reported long-term effects, and then compares them to the general population, finding that…. they have long-term effects.  The control group there really should be taking a sample of people, filtering for the 25% who report feeling the worst, and then just comparing to those 25%.

There might be real things going on here but I also think there’s a bit of an isolated demand for rigor.  People aren’t looking at the negative effects of shutting down society for a few months–people losing their jobs, for instance–and asking whether there are long-term effects of that on health (I bet there are!).


What Do I think?

I don’t know exactly what I think! But my rough sense is some combination of the following.

How would I think about this?

I would probably take two different approaches here and compare them.

Shut Up and Multiply

If you just do a good-faith effort at estimating risk, what does that imply you should do?

Well, if you think the death rate is ~0.10% for healthy people then you should be willing to sacrifice about one month to go from 100% of getting nCov to 0%. That’s an upper bound.

Right now, your odds of getting it seem to be something around 0.01%, so you should be willing to sacrifice about an hour to prevent it. That means that it might make sense to wash your hands, and pretty much nothing else is worth bothering with.

Now, if the infection rate keeps going up, eventually you should be willing to sacrifice more. Realistically, if it starts spreading to a significant fraction of the population, you might be able to take moderately extreme actions to decrease your odds of getting it by 10% (e.g. from 15% to 5%). So that would make it worth a few days of your time. That makes it seem like, even if it spreads massively, it’s probably not worth avoiding the outdoors, but that washing your hands and trying to avoid touching your face are likely worth it. Wearing masks is probably worth it if it doesn’t bother you but nowhere close if it decreases your productivity at all.

Stocking up on food seems maybe worth it but just in case the rest of the world freaks out and makes it hard for you to later get food.

Anecdotes and Common Sense

Right now, I know one person who knows one person who got nCov from traveling to China, and is going to be fine. That makes me feel pretty good about my above estimates: washing my hands regularly seems like a fair price to pay for that amount of risk, and anything else seems like an overreaction.

If it massively spreads–getting the flu sucks! It happened to me this year, and I lost half my productivity for a week. So spending a few days to prevent that seems reasonable to me–worth changing your habits if it gets above 10% of the world infected but not worth e.g. working from home.

On the other hand, half the people I know’s lives are on hold right now because of nCov preparations.

Other Measures

Temperature checks seem pretty reasonable to me–they are a good way to decrease transmission without disrupting most peoples’ lives. Closing borders seems more questionable, though at least having temperature checks on the borders makes sense. Either masks are a scourge or I’m a wimp, or maybe both; I tried wearing one and it was clear my productivity was going to be in the gutter with it. Washing your hands seems low cost and generally effective. Working from home seems high cost and effective. Going to Wuhan seems like a bad idea, though I honestly think most of China is fine.

The Costs

I think the costs to the nCov preparations the world is making is really high.  Some examples:

  1. Markets are down about 20% on nCov.  That’s not 20% of GDP, it’s 20% of the net present value of the economic future of the world.  If you use a 5% annual economic time-decay that translates to losing roughly the next 4 years of economic production.  That’s probably more than the impact of nCov if we just let the entire world get it.
  2. This hits the whole population: yeah it’s primarily the rich losing the wealth, but it’s the worse-off who are probably hit the hardest: losing a few months of salary/school/etc. is really hard on a lot of people!
  3. My guess is that’s why China is slowly giving up on having the country shut down: not just because cases are starting to decrease there but because it quickly becomes clear that it’s just not sustainable to keep the world shut down for months.
  4. Shutting the world down for a month is much worse than just losing a months’ economic growth, because the world is still consuming during that month; similarly companies still have expenses but are losing all their profits, and families still have to eat but are losing their wages.

In fact I think this 20% estimate isn’t totally crazy.

If you shut the world down for a few months, a lot of business will go under.  A lot of people will lose their jobs, a lot of people will starve in poor countries, and we very well might see the kind of snowball effects we saw in the 2008 recession.  Betting markets are already implying a 57% chance of a recession in 2020, up from around 20% before nCov; that’s also in line with the market move.  If anything I think that small businesses not captured by public market indexes will do worse than large public ones because they have less runway, and that poor countries contributing to a low fraction of world GDP will do worse because they can less afford a few months’ isolation.

These costs could probably be lessened with a well-targeted economic stimulus package.  But we’re probably on pace to lose around $20T of GDP in 2020 from nCOV; a stimulus of that size would be about 30 times the size of the 2009 US stimulus package.  I think the would probably could do it but debts would balloon globally, and realistically I don’t think governments are going to do anything nearly that big.

The Tradeoff

Putting two of these together:

Right now the benefit to worrying about nCov is close to 0.  But if it does spread a lot, then eventually you might have a 10% chance of avoiding it when you otherwise would have gotten it.  So these preparations might save the world something like 0.05% of the population’s DALYs.

On the other hand, the cost of the preparations is huge.  Markets are estimating it to be something like 20% of the net present value of the economic future of the world.  This is going to flow through not just to economics but to people; the 2008 recession wrought massive damage to everyone’s lives, and to our potential as a civilization.  So that would imply a cost of about 40 times the benefit, and the world is in fact freaking out way too much right now.


What’s Going On With The World?

In my worldview, everyone else is crazy. I basically do think that.

I basically think the world only has two settings–laughing it off, or freaking the fuck out. As soon as nCov spread outside of China, the rest of the world decided it was time for door number two.

It’s not shocking, but it’s also not productive. People’s attempts to think about how to nCov seem really, really bad to me. The best of them try to find out what the expected mortality is, but never even bother tracing that through to the cost of interventions and what’s worth it before deciding that it’s Freak Out Time. The worst of them just say “something is worse than someone thought, fuck everyone who isn’t freaking the fuck out”. The median confuses GDP with stock market moves without realizing there’s a factor of ~20 difference between them. I haven’t seen a single attempt to go through the fermi estimates I did above of the expected costs of nCov, the impact of interventions, and the cost of interventions. Maybe they’re happening somewhere–but my non-work interactions with people, online and in real life, have been roughly 90% nCov for the last month. I have to scroll for pages on Facebook to see a post about anything else! Every conference I know of has been canceled, and the first question anyone asks me when they realize I’m in Hong Kong is how bad things are. (Fun fact–Hong Kong got its freaking out of its system a few weeks ago, got bored of it, and is slowly starting to get back to work. Also basically no one here has nCov.)

And generally it’s pretty clear from talking to people about this that the first thing they do is decide it’s Really Bad and they should start Freaking Out, way before they’ve even considered doing some fermi estimates. And by the time they do start looking at data, they already know what conclusion they’re going to come to; by then they have too much ego on the line to seriously consider the possibility that it’s not that big of a deal, all things considered.

And this is all really sad, because as a society we’ll face potential obstacles time after time, and we have to be able to respond proportionately. If we have only two modes, we’ll end up Freaking The Fuck Out every year and then when it finally does matter–when the fate of the world really is at stake–we’ll just see yet another round of Think Pieces and santimonious I-told-you-so’s.

An EA Addendum

I think there’s one more piece of this that bothers me.

The advantage of freaking out is clear–it probably has some small chance of saving your life. What’s the cost?

Well, the cost is that the world descends a little bit further into chaos. GDP crashes, the odds of a global depression increase, xenophobia rises as people become terrified of those foreigners and their germs, jobs are lost which will have a permanent negative impact on productivity, and the world resigns itself to treating future incidents with a sledgehammer rather than a scalpel.

I think the flow through effects to the future generally penalize overreacting here, and so there’s a bit of a potential conflict between self and utility, with freaking out being more selfish. I would have hoped that EA would see through that.

But on top of that, I think that freaking out probably does a better job of saving your life than it does of saving your utility or impact. The negative impacts of freaking out probably won’t kill you, even if they do turn out to be large. Instead they’ll just disrupt your life, potentially permanently; not all of the jobs and friendships and mental fortitude lost to the nCov response are coming back.

Every day some people’s lives end, even if they won’t actually die for decades. I think it’s easy to end up sacrificing what matters about your life–your happiness, and your impact–in order to stave off death; sacrificing upside for certainty. I think people generally err too far in that direction, but for us it’s not close, because it’s likely that all of our impact is in the tails.

The Summoning


Found on the whiteboard of the Computer Safety Research Group.

Berkeley, California

Kobe and Hunter sat hunched over a laptop in Kobe’s living room.  Black and white stones slowly popped up on the rectangular grid; the commentators tried to grasp the mind of the masters.

A black stone appeared, and Sin furred his brow.  It was man vs machine, and game one wasn’t looking so good for man.

“Go go BetaGo!”  shouted Kobe.

Hunter cocked his head to the side.  “Remember how we’ve spent the last year of our lives trying to avoid scenarios where an AI inadvertently destroys the world?”

Kobe looked down and mumbled something about sports.  They sat in a somewhat awkward silence for the next ten minutes, watching Sin’s carefully designed structure crumble.

After the game, Kobe and Hunter headed over to the offices of the Computer Safety Research Group (CSRG) for the weekly strategy meeting.  The topic of discussion for the day: what to do when the Big Brain Countdown reaches 11pm?

Their positions were mostly predictable.  Hunter, Svengali, and Clapperton though that if they could just deter Big Brain, the world would have enough time to come up with some better frameworks for AI.  Kobe, Wilbert, and Cherish were skeptical that Big Brain was really going to make much progress.  (Hunter was quick to remind them about the afternoon’s go match.)  Rushmore and Melanie thought that eventually, someone was going to try to summon Elua, and although Big Brain wouldn’t be the worst group to do it, why not CSRG?

Lumpkins stood up to speak and everyone quieted down.

“In case you guys haven’t forgotten, this is the computer safety research group.  Our goal is to make sure no one summons Moloch, and that’s exactly what’s going to happen if we rush into this.  Big Brain has to be stopped.  And if you don’t agree, then maybe this isn’t the organization for you.”

Continue reading

Big Numbers

Part I: The Number Line


The official estimate that the United States Department of Transportation uses for the value of a life when determining how to trade of between safety and cost is $9.4 million. GiveWell estimates that, through distributing pesticide-infused bednets in developing countries to fight mosquito-born Malaria, a life can be saved for $3,500. Those translate to, respectively, 0.1 and 285 lives saved per million USD invested, making the Against Malaria Foundation a much better investment than American automobile safety initiatives.

Continue reading

I, For One

For the past few years the cause du jour of the effective altruism has been existential risk reduction. There are a variety of forms of the argument, but the most pervasive is:

  • We will soon be able to create more and more powerful AIs
  • Currently the AIs mostly just do what we tell them to do, but even that can be worrying
    • Imagine that some government writes a Homeland Defense AI that has a connection to all of the country’s missiles and drones. It intelligently figures out what the most destructive attack it could conduct on an enemy regime is, and how to coordinate all of the weapons systems available. It now only takes one click of a button for a rouge actor in the state to launch a devastating attack.
  • Eventually AIs will become as intelligent as humans, and then more intelligent.
  • AIs will then be able to improve themselves better than we can improve them.
  • Because AIs can program a lot faster than humans, once they’re able to write their own source code they will do so quite quickly
  • The self-improving AIs will in fact improve at a rate that gets faster and faster as they get smarter and smarter
  • Thus in a very short period of time AIs will go from human-level intelligence to something much much more powerful–a superintelligence.
  • Just as we are much more powerful than Apes, this AI will be much more powerful than us. If we don’t work really hard to keep it under our control and make sure it does what we want it to do, it might quickly take over the world
    • It might just destroy the world in a runaway attempt to do more and more of whatever random thing it wants to do.
    • If we try to give it a utility function but mess up, it might interpret what we give it too literally, and end up turning the whole universe into something that fits what we said but not what we meant (canonically tilling the universe with paperclips), a la Midas.
    • If we get really unlucky, the AI might actively create a ton of suffering in the quest to create a bunch of paperclips (or if we forget a negative sign somewhere in its utility function).
  • Thus we should be really careful about developing powerful AI.
    • We should make progress on AI control work–research into how to make sure AIs do exactly what you tell them to
    • We should make sure that the AIs we create are tool AIs (which accomplish the tasks we tell them too) instead of agent AIs (which choose what to do on their own)
    • We should figure out what utility function the AI should have, and find ways to specify it in ways a Turing machine can understand
    • We should make sure to do all of these before we actually get a superintelligent AI, because once AIs start self improving we might not have much time before it exerts its well on the world
  • AI superintelligence is reasonably likely to come in the next century, and so this is both important and urgent

I’m less convinced than many that x-risk is clearly the most important cause right now, but that’s not what this post is about. This post is full on contrarian arguments against EA-consensus views on AI x-risk. I’m not particularly confident in any of them, and state them without the reservation they deserve because I want to keep the frequency of “maybe” below 50% of all worlds. But together they’ve made me uncertain about how much I agree with the conventional wisdom.

Continue reading

But For The Grace Of God Vote I

My favorite food group is meat.  Beef, in particular.  I’m not sure why, exactly; maybe it’s because I grew up eating a lot of it.  I like pretty much all forms–steak, burgers, hot dogs…

When I was young I read some article online about the environmental impact of factory farms.  It made be feel shitty–I ate a lot of meat, and thought global warming was really bad.  So did what everyone would do in that situations.  I found some corner of my brain I didn’t care about very much, stuffed the article within those idle neurons, and disconnected it from it’s neighboring lobes.  Lo and behold, I no longer felt shitty.

Occasionally that dormant area of my brain would have to take on a larger burden, like when I read this.  But it wasn’t so bad.  And one thing was clear to me–those fucking vegan activists were pretty incoherent.  The combination of anti-science naturopothy, backwards-looking desires to keep nature preserved in the pristine state of five years before whenever they happened to live, focus on a death which should by their own account be mercy, and condescending certainty made them easy to dismiss.

Continue reading

Trump voters

I went to campaign for Clinton in suburban Pennsylvania a few weeks before the election.  It was a suburban, upper-middle-class, swing district with Trump signs on one lawn and Clinton signs on the next.  The goal was for me to get out the vote with likely Clinton voters who were kinda flaky and might or might not actually make it to the polls.  It was clear pretty early on, though, that that didn’t even remotely describe the people whose doors I knocked on.  There were a few threads that ran through the responses I got.

The first, and most prominent, was “you’re the 5th person who’s knocked on my door this week, fuck off”.  Ok, seems pretty reasonable.

The second required a bit more reading between the lines.  But it was clear, from what people said and how they said it, that no one wanted to tell me they were voting for Donald Trump–and not just because I was with the Clinton campaign.  It was also clear that some of them were going to.  One woman saw I was with the Clinton campaign, said “I’m not voting”, and slammed the door shut.  Another game a long, tortured response about how she didn’t know who she was voting for; after a follow-up question she replied “I guess I’m not sure we’re voting for Clinton”.  The men, in contrast, all seemed to be voting for Clinton.  Which is not to say men in general voted for Clinton–they didn’t–just that the Clinton campaign’s model of the of the suburban upper-middle-class female vote seemed to have more misses than the upper-middle-class men.  I’m guessing may of their male counterparts were already thought to be likely Trump supporters.

Continue reading