I also had too-strong priors and "expert" ideas to be properly fox-like in my predictions, and not quick enough to update about how things were actually going based on the data. Because I was slow to move from the base-rate, I underestimated the severity of COVID-19 for too long. I'm unsure how to fix that, since most of the time it's the right move, and paying attention to every new event is very expensive in terms of mental energy. (Suggestions welcome!)

Bottom line: in order to outperform base-rate, somebody somewhere has to do the expensive updates. No way around that. So the options are (a) do the expensive updates yourself, (b) give up and go with base rate, (c) find someone else who's doing the updates. (c) is the obvious answer, but bear in mind that recognizing real expertise is Hard [LW · GW], and you've already noticed just how questionable the "expertise" of many supposed experts actually is.

In the easy case, the nominal experts are in a position where we get reasonably-frequent feedback on their performance (which obviously is usually not the case for rare events). In principle, one might get around that by an expert specializing in noticing rare events across domains.

people find it far easier to forgive others for being wrong than being right

Harry Potter and the Half-Blood Prince

First of all, I really appreciate this postmortem. Admitting times when you were wrong couldn't have been an easy task, particularly if/when you staked a lot of your identity and reputation to being right. As EA and rationalist individuals and institutions become older and more professionalized, I'm guessing that institutional pressures will increasingly push us further and further away from wanting to admit mistakes; so I sincerely hope we get in the habit of publicly recognizing mistakes early on. (Unfinished list of my own mistakes, incidentally[1]). I hope to digest your post further and offer more insightful thoughts, but here are some initial thoughts:

Addendum on masks:

Another consideration about masks is that masks turn out in practice to be very reusable, a fact we (or at least I) should have investigated a lot more in early March.

On hospital-based transmission:

I don't know how much you believed in it, but as presented, this appears to be merely (ha!) a forecasting error rather than a strategic error. In the absence of a clear counterfactual, I don't think you were obviously wrong here, since it's quite plausible that if a lot of people like you ignored/downplayed the role of hospital-based transmission, it'd have gotten a lot worse.

On being a jerk re Jim and Elizabeth's post:

For what it's worth, I also (privately) asked them to take it down because I had similar considerations to you and thought the thing they wrote about masks was unilateralist-y and a bit of an infohazard. I think I was wrong there. But I think I mostly was object-level wrong about the relative tradeoffs and harms. To the extent I updated now, a) I updated object-level on how much I should cooperate or desire others to cooperate with specific institutions, and b) I updated broadly (but not completely) in general favor of openness [EA · GW] and against censorship.

I continue to maintain that if I (and possibly you) had the same object-level beliefs as before, it was not incorrect to consider it an info-hazard (but not all object-level info-hazards are worth suppressing! Particularly if release promotes the relevant meta-level norms more than it harms), though of course not an existential one.

On superforecasting:

You said you think superforecasting is

materially worse than [you] hoped it would be at noticing rare events early.

I don't know how high your hopes were, but for what it's worth, I think this proves too much. I'm not sure about the exact aggregation algorithms that the Open Phil Good Judgement covid-19 project was running, but I feel like all I can realistically gather was that "of this specific set of part-time superforecasters that were on the Open Phil-funded project, more than 50% of them were way too optimistic."

While it's certainly some evidence against superforecasters being good at noticing rare events early, I don't think it's sufficient evidence against superforecasters being able to do this, and I definitely don't think this is a lot of evidence against superforecasting as a process.

As you weakly allude to, if you were on the project and paying attention more, you would probably have done better. Likewise, I know other superforecasters who I think were much more pessimistic than the GJ median. I suspect superforecasters who regularly read LessWrong and the EA Forum would have done better; and if I were to design a better system for superforecasting on rare events, I'd a) prime people to pay attention to a lot of rare events first, and b) have people train and score on log-loss or some other scoring system that's more punishing of overconfidence than Brier.

(All that said, I think Metaculus did okay but not great on covid-y questions relative to what someone with high hopes for prediction aggregation algorithms might reasonably expect).

On US Gov't Institutions:

I think there was a bunch of insights that your policy research experience has colored. For example, you mention how you trusted the FDA to have done a lot better under Scott Goettlieb. This might be obvious to you, but it's something I didn't even really think about until you highlighted this point. You also highlight a lot of useful specific uncertainties about whether the issue was political directors under Trump or nonpolitical directors of specific institutions. I think all of these things are very useful to know from the perspective of a policy researcher like yourself (and for students of US policy), since how to reform institutions is very decision-relevant to you and many other EAs.

That said, at a very coarse level, I think I'm a lot more cynical than you are implying with regard to how well US institutions would have handled this pre-Trump. It's possible we're not actually disagreeing, so I'm curious on your counterfactual probabilities on things being an order of magnitude better (<20,000 Americans dead of COVID-19 by now, say) in the following two worlds:

a) Clinton administration continuing all of Obama's policies?

b) Clinton administration continuing all of Obama's policies except for US CDC in China being equally understaffed as they are in our timeline.

My reasoning for why I'm generally pretty cynical (at least conditional upon this pandemic spreading at all, maybe a larger international presence could have helped contained it early) in those counterfactual worlds[2]:

1) There's sort of an existing counterfactual for preparedness of governments with a broadly American/Western culture but as competent at governance as a typical European country. It's called Europe. And I feel like every large geographically Western country was pretty bad at preparedness? People are praising Germany's response, but when it comes down to it, Germany has 9000+ confirmed covid-19 deaths in a population of 83 million, or >100 deaths/million, despite taking a large economic hit to suppress the pandemic. Japan had <1000 confirmed deaths in a population of 126.5 million. Now Japan was bad at testing, so maybe Japan actually had ~4000 deaths. But even at those numbers (~31 deaths/million), Japan still had <1/3 the number of deaths per capita as Germany. And object-level, Japan seemed to have screwed up a bunch of important things, so there's a simple transitivity argument where if a high-income country did worse than Japan, their policies/institutions couldn't have been that great.

Maybe I'm harping on this too much, but I really don't want us to succumb to the tyranny of low expectations here.

Now some culturally Western countries did fine (Australia, New Zealand). I'm not sure why they did well (maybe it's because they're islands, maybe seasonality is bigger than I think so Southern hemisphere had a huge initial advantage early on, maybe because they're around 10-15% East Asian so people had enough ties to China to be worrying earlier, maybe low population density, maybe their institutions are newer and better, maybe just luck), but regardless, I'd counterfactually bet on the response of Hillary's America looking more like a slightly less competent Europe or maybe Canada and less like Australia/NZ.

2) I didn't look at it that much, but at the high-level, the US response to 2009 H1N1 looked more competent, but ultimately the response didn't seem sufficient to have achieved containment if the mortality rates were as high as people thought it'd be? (Not sure of this, willing to be convinced otherwise on this one).

3) Some inside-view reasoning about specific actors.

___

Anyway, all these gripes aside, thank you again for your thoughtful (and well-written!) post. That couldn't have been easy to write, and I really appreciate it.

[1] Your post actually me to thinking about how I should be more honest/introspective about my strategic and not just predictive mistakes, so thanks for that! I plan to update the list soon with some strategic mistakes as well. For example, I considered myself to be on the "right" side of masks epistemically but not strategically.

[2] I'm maybe 35% on a) and 30% on b). A lot of the probability mass is considerations on there being enough chaos/sensitivity to initial conditions that this pandemic maybe wouldn't have happened at all, rather than Obama's or Hillary's response being an order of magnitude better conditional upon there being an epidemic.

I have lots of thoughts, but most of all I am just really grateful for you writing this retrospective. I think it's a really great public service and I've already been thinking a lot about this post since I read it the first time a few hours ago, and expect to continue thinking about it a lot more.

Good post David!

I applaud posts like this. It's a great tradition that we do them.

I believe the conversation about border closures was with me on Twitter?

Are we at a stage yet where we can answer counterfactual questions about what would have happened if all international flights had been shut off in January (Assuming away the political problem)?

"Rely more on other people's views in the rationality community" is inherently a bias-variance tradeoff. It's also collectively self-defeating, with a potential to become a death spiral. As such, it may be a reasonable thing to do, but it is potentially very dangerous as a principle to state for others to follow. The selfishly-first-order-rational thing to do would be to do more of it yourself, and encourage others to do less of it; of course, this is free riding, and may not be meta-rational.

My point is: this may be wise, but you should beware, and you should definitely not encourage others to do this without cautioning them against the risks.

One way to beware about this is to ask the question: "which of the biggest differences between rationality community conventional beliefs, and general 'smart non-explicit-rationalist' conventional beliefs, have the weakest support?" I have personal answers to that question but in the spirit of this comment, I won't share them here.

This postmortem is so impressive. Someone should collect all the pandemic related postmortems. I'd be particularly interested in those written by people in the field (broadly construed).

Strong-upvoted! I admire you for writing this.

I said we should be very concerned in January, albeit not very publicly.

The link says:

Initial (naive) estimates of CFR are always overstated because of selection bias for the most serious cases. So our bayesian prior should be a non-trivial proportion of asymptomatic cases. And ignoring this is why we routinely overestimate severity of new outbreaks.

That's not to say that we don't need to be very concerned, but policymakers and public health officials need to be cautious about damaging credibility by repeatedly crying wolf. But the balance between avoiding alarm and ensuring sufficient response is a very difficult one.

I definitely would not have read this as saying "we should be very concerned", if that's one of the things you meant to communicate.

I also followed the herd too much from expert circles, and my twitter feed from infectious disease epidemiology circles was behind even my slow self in recognizing that this was a incipient disaster back in March.

Woah, this is interesting and really alarming.

Because I was slow to move from the base-rate, I underestimated the severity of COVID-19 for too long. I'm unsure how to fix that, since most of the time it's the right move, and paying attention to every new event is very expensive in terms of mental energy. (Suggestions welcome!)

Scott Alexander writes:

Zeynep Tufekci is an even clearer example. She’s a sociologist and journalist who was writing about how it was “our civic duty” to prepare for coronavirus as early as February. She was also the first mainstream media figure to spread the word that masks were probably helpful.

Totally at random today, reading a blog post on the Mongol Empire like all normal people do during a crisis, I stumbled across a different reference to Zeynep. In a 2014 article, she was sounding a warning about the Ebola pandemic that was going on at the time. She was saying the exact same things everyone is saying now – global institutions are failing, nobody understands exponential growth, travel restrictions could work early but won’t be enough if it breaks out. She quoted a CDC prediction that there could be a million cases by the end of 2014. “Let that sink in,” she wrote. “A million Ebola victims in just a few months.”

In fact, this didn’t happen. There were only about 30,000 cases. The virus never really made it out of Liberia, Sierra Leone, and Guinea.

I don’t count this as a failed prediction on Zeynep’s part. First of all, because it could have been precisely because of people like her sounding the alarm that the epidemic was successfully contained. But more important, it wasn’t really a prediction at all. Her point wasn’t that she definitely knew this Ebola pandemic was the one that would be really bad. Her point was that it might be, so we needed to prepare. She said the same thing when the coronavirus was just starting. If this were a game, her batting average would be 50%, but that’s the wrong framework.

Zeynep Tufecki is admirable. But her admirable skill isn’t looking at various epidemics and successfully predicting which ones will be bad and which ones will fizzle out. She can’t do that any better than anyone else. Her superpower is her ability to treat something as important even before she has incontrovertible evidence that it has to be.

The whole article seems worth reading, especially if it's true that epidemiologists under-reacted to this. It's clearly correct that most people shouldn't follow every pandemic closely -- even most epidemiologists shouldn't follow every pandemic closely. But it's important that we get the base level of alarm correct -- it might be correct to overreact somewhat to the vast majority of pandemics, if that's what it takes to avoid underreacting to the big one. And it's important that people be very explicit about how carefully they've been looking into this or that specific pandemic, so that we can collectively know which epidemiologists and other observers to pay the most attention to.

A thing I regret not thinking of is that ventilators aren't as crucial as was expected because they're dependent on the long tissue being healthy.

I'm not an expert, but it's so obvious. I don't know how to avoid making that sort of mistake. Maybe being careful about tracking chains of causation.

As mentioned in the post, I think it's personally helpful to look back, and is a critical service to the community as well. Looking back at looking back, there are things I should add to this list - and even something (hospital transmission) which I edited more recently because I have updated against having been wrong about in this post - but it was, of course, an interim postmortem, so both of these types of post-hoc updates seem inevitable.

I think that the most critical lesson I learned was to be more skeptical of information sources generally - even the most accurate, including superforecasters and the rationalist community, are fallible in ways which are somewhat predictable, and hard to evaluate prior to knowing the ground truth. This both highlights the value of staying uncertain and entertaining multiple hypotheses, and the importance of keeping diverse information sources available. The points made by John Wentworth in his comment about the need to do expensive updates was also very clear and valuable.

I certainly think additional posts of this type, by myself and by others, would provide value - and I could see it being its own genre. Unfortunately, there have been very few. I am happy to see several projects looking back at the community's reactions, successes, and failures, but they are still in progress. The 2020 Petrov Day postmortem [LW · GW] and similar are also evaluating community behavior, and some have evaluated failures in companies, but I see fairly few, and I would think we could use more, and more individual posts. (I'd hoped to write another actual after-action report, but I have been busy - an insufficient excuse - and we're unfortunately still not post-COVID-19.)

Thanks for this!

Paragraph with confusing wording:

In retrospect, I think it would have been better, consequentially, to push for cloth masks earlier, but current modeling and our understanding of spread make it clear that mask wearing by itself is only marginally effective.

Do you mean our present day understanding, or our understanding at the time? Do you mean that you still think masks are only marginally effective, or thought so at the time?

Very interesting/useful.

I suggest it is important to separate the desirability of a course of action and its political feasibility e.g. in relation to border closures.

In epidemiology it is a basic fact in the 101 textbook that slowing long distance transmission (using quarantines / travel restrictions) is very important. Unfortunately this got caught up in claims of xenophobia etc. Countries that have been relatively successful have implemented such restrictions.

I would be interested in some justification of the claim that face masks are not very useful. From all my reading, this seems to be false.

One mistake I made was not to aggressively look for countries that were successful (like Taiwan) and to enquire what they did (border closures/tightly enforced quarantine, face masks, isolating people with cold/flu/fever symptoms - even though this is not a "valid" test for CV it gathers and uses much useful information).

Like many I got caught up in the false dichotomy of lockdown=ruined economy versus no-lockdown=many will die.

This is usually correct, but here it was a mistake. (I now think that superforecasting is materially worse than I hoped it would be at noticing rare events early.) 

This seems to be an interesting point. If you do believe that relying on GJP forecasters was a mistake, what's the problematic heuristic? And how can you tell in future whether to trust them?