I was going to write this as a Slate Star Codex comment, but I’m going to make it a tumblr post tagging @slatestarscratchpad instead, since experience suggests it’s likely to be more widely and carefully read in this form. (Crossposting to LW too, so you may be reading this there, possibly with mangled formatting.)
The idea frontier
I am getting more and more concerned about the “information epidemiology” of the public conversation about Covid-19.
Here are some distinctive features I see in the public conversation:
1. Information intake must be triaged.
There is a very large amount of new publicly available information every day. There are no slow news days. “Keeping up with the story” in the way one would keep up with an evolving news story would be a full-time job.
Many of us do not have time to do this, and I imagine many of those who do have time cannot tolerate the experience in practice. In fact, there can be a tradeoff between one’s level of personal involvement in the crisis and one’s ability to “follow it” as a news story.
(I work for a telemedicine company, and after a day of dealing with the ever-changing impacts of Covid-19 on my work, I have relatively little patience left to read about its ever-changing impacts on absolutely everything else. That’s just me, though, and I realize some people’s mental bandwidth does not work like this.)
2. Abstractions are needed, and the relevant abstractions are novel and contested.
Crucial and time-sensitive decisions must be made on the basis of simulations, abstract mental models, and other intellectual tools.
In some sense this is true of everything, but in most cases we have a better sense of how to map the situation onto some large reference class of past intellectual work. When there is an economic downturn, the standard macroeconomic arguments that have existed for many decades pop back up and make the predictable recommendations they always make; even though there is no expert consensus, the two or three common expert stances are already familiar.
With Covid-19, this is not so. All the intervention types currently under discussion would be, in their own ways, unprecedented. As it struggles to follow the raw facts, the general public is also struggling to get its head around terms and concepts like “suppression,” “containment,” “contact tracing,” etc. which were (in the relevant senses) not part of our mental world at all until recently.
Thus, relative to most policy debates, this one has a strange frontier energy, a sense that we’re all discovering something for the first time. Even the professional epidemiologists are struggling to translate their abstract knowledge into brief-but-clear soundbites. (I imagine many of them have never needed to be public communicators at this kind of scale.)
3. There is no division of labor between those who make ideas and those who spread them.
There is a hunger for a clear big picture (from #1). There are few pre-established intellectual furnishings (#2). This means there’s a vacuum that people very much want to fill. By ordinary standards, no one has satisfying answers, not even the experts; we are all struggling to do basically the same intellectual task, simultaneously.
None of us have satisfying answers – we are all the same in that respect. But we differ in how good we are at public communication. At communicating things that sound like they could be answers, clearly, pithily. At optimizing our words for maximum replication.
It is remarkable to me, just as a bare observation, that (in my experience) the best widespread scientific communication on Covid-19 – I mean just in the sense of verbal lucidity and efficiency, effective use of graphs, etc., not necessarily in the sense of accuracy or soundness – has been done by Tomas Pueyo, a formerly obscure (?) expert on … viral marketing.
(To be clear, I am not dismissing Pueyo’s opinions by citing his background. I am hypothesizing his background explains the spread of his opinions, and that their correctness level has been causally inert, or might well have been.)
The set of ideas we use to understand the situation, and the way we phrase those ideas, is being determined from scratch as we speak. Determined by all of us. For the most part, we are passively allowing the ideas to be determined by the people who determine ideas in the absence of selection – by people who have specialized, not in creating ideas, but in spreading them.
4. Since we must offload much of our fact-gathering (#1) and idea-gathering (#2) work onto others, we are granting a lot on the basis of trust.
Scott’s latest coronavirus links post contains the following phrases:
Most of the smart people I’ve been reading have converged on something like the ideas expressed in […]
On the other hand, all of my friends who are actually worried about getting the condition are […]
These jumped out at me when I read the post. They feel worryingly like an “information cascade” – a situation where an opinion seems increasing credible as more and more people take that opinion partially on faith from other individually credible people, and thus spread it to those who find them credible in turn.
Scott puts some weight on these opinions on the basis of trust – i.e. not 100% from his independent vetting of their quality, but also to some extent from an outside view, because these people are “smart,” “actually worried.” Likelier to be right than baseline, as a personal attribute. So now these opinions get boosted to a much larger audience, who will take them again partially on trust. After all, Scott Alexander trusts it, and he’s definitely smart and worried and keeping up with the news better than many of us.
What “most of the smart people … have been converging on,” by the way, is Tomas Pueyo’s latest post.
Is Tomas Pueyo right? He is certainly good at seeming like a “smart” and “actually worried” person whose ideas you want to spread. That in itself is enough. I shared his first big article with my co-workers; at that time it seemed like a shining beacon of resolute, well-explained thought shining alone in a sea of fog. I couldn’t pull off that effect as well if I tried, I think – not even if the world depended on it. I’m not that good. Are you?
My co-workers read that first post, and their friends did, and their friends. If you’re reading this, I can be almost sure you read it too. Meanwhile, what I am not doing is carefully reading the many scientific preprints that are coming out every week from people with more domain expertise, or the opinions the same people are articulating in public spaces (usually, alas, in tangled twitter threads). That’s hard work, and I don’t have the time and energy. Do you?
I don’t know if this is actually an effective metaphor – after all, I’m not a viral marketer – but I keep thinking of privilege escalation attacks.
It is not a bad thing, individually, to place trust some in a credible-sounding person without a clear track record. We can’t really do otherwise, here. But it is a bad thing when that trust spreads in a cascade, to your “smartest” friends, to the bloggers who are everyone’s smartest friends, to the levers of power – all on the basis of what is (in every individual transmission step) a tiny bit of evidence, a glimmer of what might be correctness rising above pure fog and static. We would all take 51% accuracy over a coin flip – and thus, that which is accurate 51% of the time becomes orthodoxy within a week.
Most of the smart people you’ve been reading have converged on something like …
#FlattenTheCurve: a case study of an imperfect meme
Keeping up with the lingo
A few weeks ago – how many? I can’t remember! – we were all about flattening the curve, whatever that means.
But this week? Well, most of the smart people you’ve been reading have converged on something like: “flattening” is insufficient. We must be “squashing” instead. And (so the logic goes) because “flattening” is insufficient, the sound byte “flatten the curve” is dangerous, implying that all necessary actions fall under “flattening” when some non-flattening actions are also needed.
These are just words. We should be wary when arguments seem to hinge on the meaning of words that no one has clearly defined.
I mean, you surely don’t need me to tell you that! If you’re reading this, you’re likely to be a veteran of internet arguments, familiar from direct experience and not just theory with the special stupidity of merely semantic debates. That’s to say nothing of the subset of my readership who are LessWrong rationalists, who’ve read the sequences, whose identity was formed around this kind of thing long before the present situation. (I’m saying: you if anyone should be able to get this right. You were made for this.)
It’s #FlattenTheCurve’s world, we just live in it
What did “flatten the curve” mean? Did it mean that steady, individual-level non-pharmaceutical interventions would be enough to save hospitals from overload? Some people have interpreted the memetic GIFs that way, and critiqued them on that basis.
But remember, #FlattenTheCurve went viral back when fretting about “coronavirus panic” was a mainstream thing, when people actually needed to be talked into social distancing. The most viral of the GIFs does not contrast “flattening” with some other, more severe strategy; it contrasts it with nothing. Its bad-guy Goofus character, the foil who must be educated into flattening, says: “Whatever, it’s just like a cold or flu.”
No one is saying that these days. Why? How did things change so quickly? One day people were smugly saying not to panic, and then all of a sudden they were all sharing a string of words, a picture, something that captivated the imagination. A meme performed a trick of privilege escalation, vaulted off of Facebook into the NYT and the WaPo and the WSJ and the public statements of numerous high officials. Which meme? – oh, yes, that one.
We are only able to have this conversation about flattening-vs-squashing because the Overton Window has shifted drastically. Shifted due to real events, yes. But also due to #FlattenTheCurve. The hand you bite may be imperfect, but it is the hand that feeds you.
Bach, the epidemiologists, and me
Joscha Bach thinks #FlattenTheCurve is a “lie,” a “deadly delusion.” Because the GIF showed a curve sliding under a line, yet the line is very low, and the curve is very high, and we may never get there.
Is he right? He is definitely right that the line is very low, and we may not slide under it. Yet I was unimpressed.
For one thing, Bach’s argument was simply not formally valid: it depended on taking a static estimate of total % infected and holding it constant when comparing scenarios across which it would vary.
(This was one of several substantive, non-semantic objections I made. One of them, the point about Gaussians, turned out to be wrong, in the sense that granting my point could not have affected Bach’s conclusion – not that Bach could have reached his conclusion anyway. This argument was my worst one, and the only one anyone seemed to notice.)
Something also seemed fishy about Bach’s understanding of “flatten the curve.” The very expert from whom he got his (misused) static estimate was still tweeting about how we needed to flatten the curve. All the experts were tweeting about how we needed to flatten the curve. Which was more plausible: that they were all quite trivially wrong, about the same thing, at once? Or that their words meant something more sensible?
The intersection of “world-class epidemiologists” and “people who argue on twitter” have now, inevitably, weighed in on Bach’s article. For instance:
And I can’t resist quoting one more Carl Bergstrom thread, this one about another Medium post by a viral marketer (not the other one), in which Carl B’s making the exact same damn point I made about the static estimate:
Like me, these people make both substantive and semantic objections. In fact, theirs are a strict superset of mine (see that last Bergstrom thread re: Gaussians!).
I am not saying “look, I was right, the experts agree with me, please recognize this.” I mean, I am saying that.
But I’m also saying – look, people, none of this is settled. None of us have satisfying answers, remember. We are all stressed-out, confused glorified apes with social media accounts yelling at each other about poorly defined words as we try to respond to an invader that is ravaging our glorified-ape civilization. Our minds cannot handle all this information. We are at the mercy of viral sound bites, and the people who know how to shape them.
What is it the rationalists like to say? “We’re running on corrupted hardware?”
Carl Bergstrom championed a meme, #FlattenTheCurve. He believed it would work, and I think it in fact did. But Carl Bergstrom, twitter adept though he may be, is still someone whose primary career is science, not consensus-making. In a war of memes between him and (e.g.) Tomas Pueyo, I’d bet the bank on Pueyo winning.
And that is frightening. I like Pueyo’s writing, but I don’t want to just let him – or his ilk – privilege-escalate their way into effective command of our glorified ape civilization.
I want us to recognize the kind of uncertainty we live under now, the necessity for information and idea triage, the resulting danger of viral soundbites winning our minds on virality alone because we were too mentally overwhelmed to stop the spread … I want us to recognize all of that, and act accordingly.
Not to retreat into the comfort of “fact-checking” and passive consultation of “the experts.” That was always a mirage, even when it seemed available, and here and now it is clearly gone. All of us are on an equal footing in this new frontier, all of us sifting through Medium articles, twitter threads, preprints we half understand. There are no expert positions, and there are too many facts to count.
Not to trust the experts – but to exercise caution. To recognize that we are letting a “consensus” crystalize and re-crystalize on the basis of cute dueling phrases, simplified diagrams and their counter-simplified-diagrams, bad takes that at least seem better than pure white noise, and which we elevate to greatness for that alone. Maybe we can just … stop. Maybe we can demand better. Wash our minds’ hands, too.
Our intellectual hygiene might end up being as important as our physical hygiene. Those who control the levers of power are as confused and stressed-out as you are, and as ready to trust viral marketers with firm handshakes and firm recommendations. To trust whichever sound byte is ascendant this week.
Thankfully, you have some measure of control. Because we are all on flat ground in this new frontier, your social media posts are as good as anyone’s; you can devote your mind to making ideas, or your rhetorical skill to promoting specifically those ideas you have carefully vetted. You can choose to help those with power do better than the status quo, in your own little way, whatever that may be. Or you can choose not to.
Okay, words aside, does the right strategy look like the famous GIF taken literally, or like a feedback system where we keep turning social distancing on and off so the graph looks like a heart rate monitor, or like a “hammer” reset followed by a successful emulation of South Korea, or
I don’t know and you don’t know and Tomas doesn’t know and Carl doesn’t know. It’s hard! I’m hadn’t even heard of “R_0” until like two months ago! Neither had you, probably!
Marc Lipsitch’s group at Harvard has been putting out a bunch of preprints and stuff that look reputable to me, and are being widely shared amongst PhDs with bluechecks and university positions. Their most recent preprint, from 3 days ago, appears to be advocating the heart rate monitor-ish thing, so yay for that, maybe. But … this sounds like the same information cascade I warned against, so really, I dunno, man.
However, I will suggest that perhaps the marginal effect of sharing additional reputable-seeming takes and crystalizing weekly orthodoxies is negative in expectation, given an environment saturated with very viral, poorly vetted words and ideas.
And that your best chance of a positive marginal impact is to be very careful, like the people who won’t trust any medical intervention until it has 50+ p-hacked papers behind it, has been instrumental in the minting of many PhDs, and has thereby convinced the strange beings at the FDA and the Cochrane Collaboration who move at 1⁄100 the speed of you and me. Not because this is globally a good way to be, but because it locally is – given an environment saturated with very viral, poorly vetted words and ideas.
That you should sit down, take the outside view, think hard about whether you can make a serious independent intellectual contribution when literally everyone on earth, basically, is trying to figure out the same thing.
And you know, maybe you are really smart! Maybe the answer is yes! If so, do your homework. Read everything, more than I am reading, and more carefully, and be ready to show your work. Spend more time on this than the median person (or me) is literally capable of doing right now. This is the value you are claiming to provide to me.
If you can’t do that, that is fine – I can’t either. But if you can’t do that, and you still boost every week’s new coronavirus orthodoxy, you are an intellectual disease vector. Don’t worry: I will hear it from other people if I don’t hear it from you. But you will lend your credibility to it. Whatever trust I place in you will contribute to the information cascade.
This work, this hard independent work collecting lots of raw undigested information, is actually what Tomas Pueyo seems to be doing – I mean, apart from framing everything in a very viral way, which is why you and I know of his work. We are saturated with signal-boosts of the few such cases that exist. We do not need more signal-boosts. We need more independent work like this. Please do it. Or, if not that, then be like the lady in that very problematic GIF: don’t panic, but be careful, wash your mind’s hands, and (yes) flatten the intellectual curve.
When you’re dealing with a threat that doubles in size every few days, you do not have the luxury of excess caution. Inverted pendulums have an exponentially growing error as well, and no matter what you do (or don’t do) to react, if your control system doesn’t do it faster than the instability grows, you lose. Period. If you try to move slowly in the act of balancing, then you will fall off the tight rope no matter how sure you later become of what the right action would have been. It is fundamentally necessary to be able to react and then correct for errors later (so yes, pre-frame this in your communication so that you don’t over-commit to something you will need to later change).
It’s also worth noting that “literally everyone on earth” only starts trying to solve the problem once they know that it’s a problem, and at the time that Pueyo’s first essay came out, that was absolutely not the case. At that time, I was still scrambling trying to figure out how to best leverage my credibility and communication skills to convey the exact same point about “Why you must act now” because people around me had not yet come to realize how serious an issue this was going to be. Sure, they’d have heard it without me too. But they would have heard it later with less time to act, and might not have taken it as seriously without my credibility behind it. If enough people like me took your advice they might not have heard it at all because it could be out competed by other less useful memes.
It’s just that, as you manage to find alternative takes (perhaps by credentialed experts, perhaps not) that find flaws in the memes you’ve been spreading, you spread those too. I would say “correct for your mistakes” except that it’s not even a “mistake” necessarily, just “a clarification of oversimplification” or “the next control input given the newest estimate of the state”.
As we get deeper into this mess and people start mobilizing, then “do something in this general direction!” becomes less important. At some point you have to wonder whether the pendulum has swung too far, or if we need to be acting in a different direction, or something else. When everyone in the world is thinking about it we now have a very different problem and instead of simply requiring an ability to take back of the envelope models seriously when they are outside the “socially accepted reality”, you actually need more detailed analyses.
Still, public opinion will need to get on board with whatever is necessary, and in the absence of your input the memes don’t just stop and wait for science, and neither does the coronavirus. If you try to say “but I can pick nits! This isn’t credentialed and perfect!” and try to replace useful first steps with inaction, then you blow your credibility and with it your ability to help shape things for the better. Let’s not do that.
Yes, it is important to not initiate or signal boost bad information at the cost of good ones. Yes, it is important to look for people who are (actually) experts. But it’s also important to provide a path from the real experts to the layfolk, since that doesn’t and cannot happen on its own. The public in general not only can’t evaluate the object level arguments about epidemiology and must defer to authority, they can’t even evaluate object level arguments about who is the real authority—that’s why you get antivaxxers listening to crackpots. It’s appeals to authority (mixed in with justifications) all the way up. If you can’t create the best ideas but you can distinguish between the best ideas and those which merely look good to the untrained eye, it is your job to pass the best ideas down to those who are less able to make that distinction. If you can’t make that distinction yourself but you can at least distinguish between people who can and posers, then it is your job as the next link in the chain to pass this information from those more able to discern to those who are less able to discern than you. This goes all the way down to the masses watching the news, and you better hope you can get the news to get their shit together. I still know people who are in denial because mainstream news told them to be and then failed to appropriately correct for their earlier mistakes. Let’s work to fix that.
Exponential memetic spread does not pathology make. Yes, it’s possible for overactive or mistargeted immune systems to fail to prevent things or to do more harm than good. Yes, Dunning Kruger applies and humility is as necessary as ever. However, so is the courage to be bold and to take action when it is called for instead of hiding in false humility. This “intellectual curve” is a part of our collective immune response to an actual virus which is killing people and threatening to kill exponentially more. Do not flatten the wrong curve. Find a role that allows you to guide it in the right direction, and then guide.
Excellent comment, thank you! Don’t let the perfect be the enemy of the good if you’re running from an exponential growth curve.
No, it is really easy. If the society is capable of executing the feedback system, then it has a method of driving R0 below 1. If it can do that, it should execute the hammer. This requires less aggregate time under extreme measures than the feedback system. If the society fails to execute the dance, then it can try the feedback system, but that sounds a lot harder to me than the dance.
But you were wrong about Gaussians. So much the worse for the experts.
There are only two choices: herd immunity, or the dance. Herd immunity can be accomplished by vaccine, by permanent change to society to reduce R0, or by having 1-1/R0 proportion of people survive the disease. Without a vaccine and without permanent change to society, the area 1-1/R0 is fixed, although a full-speed epidemic could infect even more.
Counterpoint: most people who will read your post are already better than average at vetting-memes-before-spreading. If you succeed at making these folks even more cautious, everyone else in the world will still keep spreading unvetted memes, so worse memes will win.
The problem isn’t spreading memes in and of itself. “Flatten the curve” is the same kind of statement as “we stay here for you, you stay home for us” or “we will defeat this invisible enemy”; it’s a motivational poster, a compact way for people to remind themselves what’s happening and why their efforts are important. The problem is when people start trying to interpret the memes as *policies*, and argue about whether the curve can indeed be flattened or declare that we’ll be in lockdown for N months because otherwise the curve won’t be flat.
It seems to me like the role of LessWrong should be to produce new ideas and be at the cutting edge. It’s not a place that’s valuable as a venue for hearing the same ideas from elsewhere repeated without much filtering.
But we already filter more than the reference class of smart Internet people, that’s the point. cousin_it argues, and I agree, that this community may already be on the extreme of “filters too carefully in the face of the need for urgent updates”. We did well by taking COVID-19 seriously before it was proven, and we could have done still better on that front.
Agreed that thinking about informational ecosystems is very high leverage. It’s not clear to me whether it’s better to call for greater individual virtue (as you do here) vs trying to rework the underlying incentive structure (some posts of mine that fall into that general category: 1, 2).
One notion I’ve found helpful, which is one of the things you’re getting at here, is that of “ideological Matthew effects”. Basically, the more dominant an ideology is, the less receptive people become to competing ideologies / the more people that ideology has available to shout down critics. Under this framework, contrarians can be really valuable even if they are usually wrong, because they inject diversity into the informational ecosystem. However, unless you’re in a betting context like the stock market, contrarianism has an incentives problem—from an individual incentives perspective, a contrarian risks opprobrium/ostracization/etc., whereas society being marginally more correct about topic X is generally a public good.
Recently I was toying with the idea of publicly shaming any well-funded group that doesn’t offer cash prizes for the best arguments that it’s misguided (possibly inspired by this post). Didn’t seem to take very well though.
I suppose it will be too late, but one way to test the “flatten the curve” idea is to note how many countries go to saturated populations while still keeping the need for ICU/ventilators always within capacity. The advertised claim for FTC was that it would allow everyone who needed treatment to get it.
That is how many countries do not overload capacity, while not crushing the virus.
My claim is that no-one will achieve this. That is they will either crush the virus (CCP China ex Wuhan etc, Taiwan) or have a blowout (North Italy).
I spent several days modeling this and regardless of how complex the models, the window to achieve this is very narrow. Models done by experienced professionals produce much the same result.
The intuition is that exponential growth either blows up or collapses. The zone where it grows very slowly is narrow (typically 1<R0<1.08). This happens even when you take into account the fact that fractions of the population become immune over time and so the growth rate slows as you approach saturation.
It is true that reducing the raw R0 (based on pristine population) does reduce the total number infected, because the effective R0 falls under 1.0 sooner and the virus dies out. But this is still in a world where there are huge capacity overloads.
Are you accounting for efforts to increase critical care capacity? Most Western governments seem to be acknowledging that cases will blow up, and the only question is whether we can get enough ventilators in time for the peak.
Thanks for sharing this. I wish more people would share their reasoning about how to model rather than just their finished models.
I think Scott linked to Pueyo’s essay as an illustration of the ideas, not as the source from which the smart people got the ideas.
Which means that this post’s attempt to track & evaluate the information flows is working off of an inaccurate map of how information has flowed.