Hanson & Mowshowitz Debate: COVID-19 Variolation

post by Ben Pace (Benito) · 2020-04-08T00:07:28.315Z · LW · GW · 3 comments

Contents

  Some of the topics discussed
  Highlights
  Links to Things Discussed in the Debate
  Transcript
    Introduction
  Debate First Half
    Basic Case for Deliberate Infection
    Critiques on effect size and timing
    Response to critique on effect size
    Agreement: In praise of experiments
    Response to critique on timing
    Robin asks for an alternative concrete proposal
    Zvi asks for next actions to take
  Debate Second Half
    Brief recap on assorted topics
    Ick reactions to the idea of deliberate infection
    Is deliberate infection worthwhile near the peak?
    Concretely discussing what deliberate infection looks like and comparison to work on isolation
    Things that can be done to help
  Q&A
    What could go horribly wrong? – from Oli Habryka
    Should effective altruists spend time on Covid-19? – From Daniel Filan
    What probability does Zvi assign to us needing to deliberately infect? – From Ben Pace
    How soon to herd immunity?
    Why isn't China doing deliberate infection?
    What does Zvi think the effects of this negative shock will be on our institutions? – From Alyssa Vance
    What specific information does Robin think we can get from deliberate infection that we cannot get from natural data? – From Lotus Cobra
    On what scale does Robin think deliberate infection will be needed? – From Elizabeth Van Nostrand
    Ending
None
3 comments

On Sunday 29th March, Robin Hanson and Zvi Mowshowitz debated Robin's policy proposal of deliberate infection (now named "variolation"), a proposal that Robin argues can reduce deaths from coronavirus by somewhere between 3x to 30x. 

They talked on a YouTube livestream for 90 mins, followed by a 30 minute Q&A moderated by me. Full recording on YouTube.

This is a proposal Robin wrote about first on February 17th. Since the debate he wrote a new full summary of his position in the post Variolation (+ Isolation) May Cut Covid19 Deaths 3-30x and continues to write on the topic. He has also now had a debate with Greg Cochran (link), plus a back-and-forth with Tyler Cowen about it (Tyler's critique, Robin's response). 

Zvi's post-debate posts include Taking Initial Viral Load Seriously [LW · GW], and On R0 [LW · GW] (in part a response to Robin's post on the topic) and continues to write on the topic.

This was an experiment in online events by the LessWrong team, with more to come.

Some of the topics discussed

Highlights

Where deliberate infection fits into the picture

Robin: The basic idea here is to have a better Plan B. Lots of people in the last month had been really hopeful that even though we really screwed up in the first month or two of this, especially in the United States compared to many other places in the world, that we're the dynamic powerhouse and we're going to fix that and we're going to, therefore, create unprecedented levels of testing and tracing and isolation and we're just going to do a heroic job. And hey, it's our moonshot, it's our World War II and this generation is going to do it. And you know what? That's a hope, it's real and it should be pursued. But we should also be planning for what happens if that doesn't work

On the difficulty of making policy proposals

Robin: One of the main problems with trying to make policy proposals, is that... the simplest thing to talk about is the thing that would ultimately happen. But if you're trying to talk about the thing that would ultimately happen, people immediately respond with, "But look at all the uncertainties. How could you possibly know how to do that right?" And of course what you really want to propose is that you start down a path that could get to that big thing, but that along that path you would be doing small things that would be telling you all the key things you need to know about how to do the big thing right. Or even whether to quit and don't do the big thing.

Robin: So I am talking about both. I mean, we can't talk about the short things without talking about the big things too, because the whole point of doing the small things is to prepare for, and to lay the groundwork for, the big things. But it's stupid to just pretend like we could know exactly how to do the big things right out from the bat. So in any policy area, including this, this is for all policy whatsoever, you're always talking at these two different levels of trying to imagine the thing you'd eventually do and how that would play out and whether that was a good idea. And then trying to think about what path you would use to get there in terms of what experiments you do, what permission you ask for, what trials you do, and what variations, so that you could learn not only how to do it right, but also create demonstrations and motivation and excitement and energy and attention, such that we might eventually get to the big things.

On the potential impact of reducing viral load through deliberate infection

Robin: And surprisingly, given how we have this huge historical precedent of smallpox, there's been very little discussion of this. I should also mention, I initially, six weeks ago, started writing about this in the context of flattening the curve just from the benefit of if you deliberately infect and isolate people, then you infect some people outside of the peak and that allows them to get more medicine. And that's roughly 20% reduction in death rates as I calculated. Which I was excited about and then last week I realized there's this 3 to 30 factor in death rate, which is a much bigger motivation.

On natural viral load potentially being quite low

Zvi: One thing that I noticed when I was thinking about this carefully was that it's very important to understand what kinds of infections you would get in the wild. Like how many people low doses versus high doses versus giving everybody a low dose. Because you don't give everybody a high dose, no matter what policy you're in, unless you're doing something very malicious or on purpose.

So one rule of thumb I asked myself was, what's the household size in the United States as an example, right? Just to think about the American situation. The average American household is 2.6 people. If you exclude children that can easily be held home, who also just don't have much risk because children under the age of 18 basically never die or get seriously ill from this to a first approximation, we’re left with a household size that averages about two. 2.02 I believe is what I got.

So at that point, if we're worried about high viral loads being modelled as coming from familiar interactions inside whatever your lockdown, quarantine area would be. If someone else in it got it and then you got it from them, you're assuming you'd have probably a high viral load. And then if you get interactions at the grocery store or otherwise, trying to make your way in the world normally under a quarantine, you would then probably get a relatively low viral load. Right? Like the study with the measles with the kid who got it outside the home was relatively safe.

On avoiding thinking about failure

Robin: But I actually agree with Ben that in many, even rationalists’ minds, they have hope. And this in some sense feels like giving up on the hope, that is once you just try to infect many people, then you're giving up on the vaccine that will save us all or the squash that will kill it all. And they put a substantial hope on those things. And this feels like giving up.

Zvi: Yes. And you're not allowed to give up. Like that's really terrible to give up on human lives.

Robin: Or to make a plan B in case you fail is even really terrible.

Zvi: Yeah. Making a plan B is already like escapism and pretty terrible but certainly actually doing it before we're sure we didn't have a way...

Robin: Not only making a plan B but exploring and putting resources into figuring out which plan B would be better.

Zvi: Well because like you're stealing, I can just imagine-

Robin: Stealing resources of plan A.

Zvi: Right. Like people are dying because you didn't give them tests. Like how dare you sir. You could have built a hospital instead of building this research clinic.

On the coronavirus and existential risk

Zvi: I think this is more of an x-risk or approaching x-risk, than you're giving it credit for. In the sense that for example, this could lead to a US–China war, which could be very, very bad. This could lead to lasting severe economic damage, which would be very, very bad. Potentially even collapse of civil order under some scenarios. And if we don't get immunity for very long, this could be an actual civilizational threat. If this comes every year and is this bad because we can't shut down the economy like this every year, and it's not clear that we can sustain ourselves. So there are risks.

On governance changing due to the virus

Robin: So, I wanted to mention that in most other areas of altruistic concerns, one of the big issues that come back again and again it's governance. People have various ways they hope the world can be improved, then they face governance limits. The governance gets in the way. Governance isn't there to solve the problem. 

Robin: We will be changing governance in response to this crisis. We already are changing governance in response to the crisis. How governance changes in response to this crisis will be crucial in terms of whether it opens it up or shuts down ways that we could coordinate on other important questions.

On doing the basics first

Zvi: Yeah, I think basically we are failing really hard on a lot of very simple things right now that we could push on that would greatly improve our chances of successfully squashing this, at least for a while. And that only after we've done those things and ramped up, so a few months from now, we should absolutely look at our data from deliberate infection tests and look at where we are and how likely we are to be able to contain this in the long run and economic costs of that and ask ourselves, should we be deliberately infecting now?

On the most likely future

Robin: The scenario that I find most likely and disturbing is that all the hope that's generated lately for a lot more testing and a lot more tracing and a lot more isolation, the same government agencies are going to be implementing that, that already implemented the failure before, and they just won't pull it off. And that'll consistently happen all through the West and therefore this will just get really big. And then it'll go back and re-infect even China and Korea, et cetera, who had been keeping it out until a much larger way of comebacks comes back to hit them. And then there'll be this vast amount of death and all this finger pointing and all these recriminations and all this blaming. And that's a pretty bad scenario, but it seems to me the most likely one. I'm happy to support, hope and to help the hopeful scenarios try to develop, but I still can't put a huge probability on them.

Zvi: I think the reason I'm more hopeful than you are is because I don't see this as primarily resting upon the government being competent because I think that the people and the corporations and the non-governmental resources will pull this together and get us through this in a lot of scenarios, despite the government, right? Through the government's resistance. That's kind of a lot of what's happening already.

Next steps for testing deliberate infection

Robin: We're talking right now about whether to focus on doing this. So we're asking, this relative to other things we could focus on, its relevant potential and priority and what we could be doing. So, I want to make the argument that even if you're not ready to pull the trigger, and I'm not clear that you shouldn't pull the trigger, we'll talk about that in a moment, that there's a lot to do here with high leverage that somebody, a foundation or philanthropists or whatever, could go a long way and do a lot to help with this just by first investigating legal permission and finding a way that somebody somehow could do it legally. And then finding a way to set up a place, even offshore, where you go do it for a small group so that we get the sort of data and the demonstration.

Links to Things Discussed in the Debate

Transcript

Introduction

Ben: And we're live. Cool.

Ben: Thank you very much Robin, thank you very much Zvi. Welcome to everyone who's joining, thanks for joining us. So this is the first LessWrong debate, and I've wanted to have a debate like this for a while. In particular, I've been interested in having Robin and Zvi have a debate, ever since I saw them writing various disagreements to each other on the internet on topics like blackmail. Robin's blog, Overcoming Bias, I've been reading for about 10 years, since I was like 13, so I've learned a bunch from that. And also I've seen Robin in various debates, and I can't think of a comparable reference class for a bullet of Econ 101 knowledge than when I've seen Robin do certain debates. There's a great debate Robin does with a guy who says we're going to colonize space soon, on YouTube, that I've occasionally watched with friends for fun.

Ben: And then Zvi's blog I've been reading for a couple of years. And in terms of effects size, well I guess you got me to stop using Facebook, and I've definitely learned a lot from that in particular. Yeah, I have not used Facebook for about nine months. I haven't posted and I haven't used a react, except for making the Facebook event for this debate. So, you should question the incentives you're putting on me.

Zvi: Throwing it all the way just for this!

Ben: I think a lot of people in the current world don't have many opinions and they're sort of scared for them to think for themselves in public, but wherever I see Robin's and Zvi writing, they're always very much themselves and very much having strong opinions. We were originally going to take this debate on a different topic, but in the current crisis it seemed like it was something Robin's been writing about for over a month now, about his particular policy proposal. And Zvi had a lot of opinions and it seemed like a good topic to discuss.

Ben: So I thought I'd be interested, just initially, to ask you guys how you're currently dealing with the current crisis and how much it's affecting you. I'm curious, Robin, how have the last couple of weeks been for you?

Robin: Well, we moved into this condo two years ago, so as a professor I'm still working, I'm just working from home, teaching classes over the internet here. I had a little more warning than most, so managed to stock up, so I'm relatively comfortable. We're mostly anxious about my mother who seems to want to go to church every two days and hang out with people and can't be talked out of that. And we have lots of food that isn't as tasty as the food we could get if we went and got fresh ingredients, and we have to decide whether to be tempted to go out and get fresh ingredients or make sure to stay home, and whether we should not let the maid come and let it get dirty or that sort of thing. But otherwise we're doing pretty well.

Zvi: Yeah, it's amazing the different orders of magnitude of risks that you're debating, and we have the same thing where Laura's and my parents we're very worried about this whole time. I was trying to convince my parents to leave New York City and they briefly went to Riverdale and then they decided to come back, and literally we'll just never go outside of the apartment. We'll have our friends bring us packages and we'll leave them outside for three days and it'll be fine, or something like that.

Zvi: But I still worry a lot. But no, it's been a very hectic two weeks for me and my family because we have two young children. Alexander is five and Gideon is two. And so we normally have a two bedroom apartment in Manhattan, and we very quickly dawned on us that being stuck there, even with a lot of food in the pantry, was going to rapidly drive everybody involved crazy. The nanny would have to either be with us or not be with us and both options seemed terrible, and we wouldn't be able to get any work done and it would be a complete disaster.

Zvi: And then the doubling time turned out to be two and a half days instead of what I thought it was, which was four or five. So I thought I had an extra week or two to do things that I thought I was safe and then I wasn't, just like everybody else. So we went down to see houses here in Warwick, New York, and then we decided to stay an extra day and actually just rent one immediately. So we managed to close a day after we started looking at places, signed a lease, got out of our old lease, all of our stuff was shipped here. We went back to New York, actually, three times because our awful, awful Mayor started talking about potential lockdowns and quarantines. And the last thing you do, right before a lockdown, is say you might have a lockdown. You idiot.

Zvi: So instead we went in the car and got a friend to bring another car and we just took all of the stuff, Passover style, we just couldn't live without, and we just ran away with two cars and some stuff. Then we came back with a van the next day and got the stuff that we really didn't want to live without. And then we got movers for the furniture, which we did in two parts because I didn't want to be in New York three days later to let the house be ready, among other reasons. Because the infection rate was doubling every three days and it seemed really like a bad idea. And on top of all of this, at the same time, we get news that my son's special education teacher is sick and she's getting a test and she's positive.

Zvi: So we get a test for our son, who is five, who of course is asymptomatic because he's five, and it's inconclusive because he wouldn't cooperate with the swab, we apparently couldn't get enough material. And so then we have to go back and get tests for us because we now had long enough to do that. And luckily we figured out how to do that and we got tests and we're both negative.

Zvi: What's weird is that like, this is actually relevant to our discussion, we were hoping it would be positive. Because by that time Laura had minor symptoms and they had gone away and I had not had any symptoms for a while. I'd be like, "Well, if we're both positive now then we're just immune. Isn't that just better? Like yeah, we suffered a bit but what's done is done and we got through it before the hospital system got overwhelmed and it wasn't our choice but great."

Zvi: But alas, we now still have to debate exactly how we're going to de-sanitize our packages and our takeout and so on, and try to get delivery slots for the groceries. But we're settling in now to rural house life and it's very different in many ways where you can't go anywhere. But it has advantages, and I think life is mostly returning to normal at this point and that allows me to then, of course, spend a lot of my time obsessing over the situation of Covid-19. Which is not what I was hoping for, I was hoping to get back to work, but we'll make do. So, here we are.

Zvi: So that's my situation. I've also, my dad is an immunologist, so I've been trying to consult with him to get more information and apparently he was at least helpful in contacting Andrew Cuomo and getting him to do a convalescent serum antibody testing.

Ben: Nice. Nice. I'm glad to hear that.

Zvi: I'm happy we've at least done something.

Ben: Yeah. So yeah, just my update, we've been quarantining for a couple of weeks now and I was successful in getting my mum to isolate and stockpile a couple weeks early, which was pretty good. We had one person renting in our house and they still had been renting for a while and been looking for a place. We managed to help them get rent on a place for a month and they moved out the day before the lockdown and they were someone who had kept going into work and using public transport. So we're lucky to finally have the place to ourselves and just people who are down for isolating and so on. But yeah, otherwise we've been working on it pretty non-stop at LessWrong. We’ve been trying to summarize a bunch of the advice online and build some databases and doing a bunch of stuff. It's been a full time focus of the team now. Just late nights and so on.

Zvi: Yeah, it's weird getting half of my world-news from rationalists over the last month, between LessWrong and rationalist Twitter it seems like a lot of the best information is coming forward.

Ben: I feel like I learned a lot from Robin cause I think Robin tweeted… you were writing about this early February, mid-February?

Robin: Mid-February.

Ben: And then like a week later you were like "Ah, this has become normal now. This is time for me to leave the conversation because I have done all that I can do." I it like "Oh wow, you're trying to be that ahead of the curve?” Okay, that makes sense why you're not getting...

Robin: It didn't really work, I'm afraid. I want to make clear to everyone, we don't think we are suffering, especially compared to other people and especially what's going to be happening over the next few months. So we just thought it would be good to relate everyone our personal stories just so you could–

Zvi: Yeah. Look, we had a freak out set of two weeks that were very, very stressful and we were briefly worried that maybe Laura's parents were positive, and any number of things could happen. But yeah, it's nothing compared to people who are actually sick or actually on the front lines of this thing. Like we have enough spare cash to just go rent a place and move there and work from home and hide out. So yes, we're going to be fine and other people might not be. It would be, obviously if our parents got sick that would be very bad and very sad. That's the only thing that we're personally worried about. That and our stock portfolios, I guess to the extent that we still have them.

Debate First Half

Basic Case for Deliberate Infection

Ben: So moving onto the topic of discussion, which is the policy proposal from Robin. Robin, do you want to just take five 10 minutes and just like outline your position?

Robin: I will try to briefly summarize, and Zvi feel free to butt-in. So I don't know if you've told them, but we've decided to do a relatively flexible debate format where we'll just interrupt each other and go back and forth. And if the moderators decide, somewhere along the way, that one of us has been talking a lot more than the other, they might intervene. But probably that won't happen. So we're just going to talk.

Ben: Great. Go for it.

Robin: All right. And you guys have some links, probably up, with some of my blog posts presumably. All right, so the basic idea here is to have a better Plan B. Lots of people in the last month had been really hopeful that even though we really screwed up in the first month or two of this, especially in the United States compared to many other places in the world, that we're the dynamic powerhouse and we're going to fix that and we're going to, therefore, create unprecedented levels of testing and tracing and isolation and we're just going to do a heroic job. And hey, it's our moonshot, it's our World War II and this generation is going to do it.

Robin: And you know what? That's a hope, it's real and it should be pursued. But we should also be planning for what happens if that doesn't work. Because you know, up until a few weeks ago at least, most experts in this area, most public health people were saying it's over. It got out, at this point was where we usually give up. So what if we can't contain this thing? What if it basically spreads to most everyone? And so most everyone gets sick and therefore, even if the estimates of half a percent of the population who are infected die, and half the world is infected, you're still talking 20 million people in the world dying. So what could we do about that?

Robin: So the usual story is flatten the curve, try to lower the infection rate so that even though just as many people eventually get infected, they're spread out more over time so that they each have more medical resources available to them. That's the usual story. And it's a fine story and we should be willing to pay substantial, though not infinite, costs to spread out the curve. And that's why we are doing a lot of lockdown sorts of things.

Robin: But that policy has some, perhaps, neglected costs. So obviously there's a huge economic cost and social cost just from messing up the whole system. And then a lot of people will be depressed and commit suicide and not could access to medicine, et cetera. But in addition, there's this key effect of dose size in infections. So when you are home, locked down, you're much more likely to get infected in the home or get infected from some close contact you have. And those sorts of infections would tend to come up with a lot of copies of the virus in the first, if you kissed somebody, for example, in the first bunch. And at that point, it's a race in your body between the immune system, which is trying to grow up fast, and then the viruses which are trying to grow fast.

Robin: And the bigger that initial load, the harder it is for your system to beat it. And in the three other data points that I've been able to find clearly for death in humans, this death rate as a function of this dose varies from a factor of 3 to 30. So we're talking a really large factor here. For example, people are really obsessed with ventilators. Like do we have enough ventilators, et cetera. I'm going apparently 85% of people who get a ventilator die anyway, so ventilators can, at most, say 15% of them.

Robin: So a factor of 3 to 30, it's just bigger than that. So that makes you worry a bit about the lockdowns, although if you are going to be able to suppress it and then only a small fraction of people ever got it, that that factor of 3 to 30 could still be worth it. But if we're actually going to go everywhere, it's not clear it's worth paying that factor of 3 or 30, but that raises another exciting prospect which is, we could go the other direction. Instead of accepting an average infection rate that you'd get if everybody was just mixing, which is lower than if you lock everybody down and make sure they have high contact doses, what if we go out of our way to do low doses and deliberately infect people with low doses.

Robin: And you know this has historical precedent. This is basically the treatment used for smallpox up until the vaccine in the late 1700s. And reportedly, for back then, it was the difference between a 20% or 30% mortality rate if you got infected naturally, and 1% or 2% rate if you got infected through the low dose. So, we could start a policy of infecting people on purpose and then immediately isolating. Now if you just go infect people on purpose but you don't isolate them, you just speed up the whole thing and you make the flatten the curve thing worse. You're sharpening the curve. But since you're deliberately infecting somebody, you can go out of your way to make sure that they're isolated right at that point. And so the idea here would then be to deliberately infect people.

Robin: Now of course, you could imagine a mandatory program and telling who to get infected and people could be really upset. I would more imagine a voluntary program where the first thing to do is just make sure there is a place you're allowed to do it and a place you could go if you wanted to do it, where maybe you'd have to pay to go but it's like a hotel or something where there's some medical services there if you turn out to need it. But otherwise you're in your hotel room and you're infected.

Robin: And in order to make that happen, we just need legal permission which is some work because it's not clear that you'd have legal permission here. But if you could just have some group of people that have legal permission to do this, then you can set that up. And then the first few groups that did this, they would try to learn exactly what the right way was to give you a low dose and whether it should be a prick on your skin or through your mouth or whatever, what should it be as a way to get to you. And we would figure out exactly how far we could go in terms of lowering mortality rates.

Robin: And then, of course, once it was successfully demonstrated, then you might consider a larger policy, which mainly I would think would be able to form subsidizing, i.e. paying people to volunteer to do this so you get more of them. And preferentially paying some groups over others. Obviously young people and healthy people are going to face a lower death risk if they do this, but they will contribute roughly the same amount to lowering the overall degree of infection in society and getting to the point of, what they call, herd immunity. So it would make more sense to infect the young and healthy. Also make more sense to infect critical workers who you would need at key times.

Robin: And so that's basically the idea. And surprisingly, given how we have this huge historical precedent of smallpox, there's been very little discussion of this. I should also mention, I initially, six weeks ago, started writing about this in the context of flattening the curve just from the benefit of if you deliberately infect and isolate people, then you infect some people outside of the peak and that allows them to get more medicine. And that's roughly 20% reduction in death rates as I calculated. Which I was excited about and then last week I realized there's this 3 to 30 factor in death rate, which is a much bigger motivation. And so that's my basic case. Of course, we can go farther into the discussion, and Zvi knows this, but the rest of you may not have so.

Critiques on effect size and timing

Zvi: Right. So just going to outline. I do think that this effect is almost certainly real. We have low confidence in its size obviously, the effective of low viral versus high viral loads initially. But it's almost certainly real. And I do think that the initial proposal of Robin's to deliberately infect people, to get it outside of the top of the curve, that proposal has expired just because we are already too late. By the time we ramped up deliberate infections, we wouldn't be able to flatten the curve that way anyway.

Robin: Maybe in New York City. There's a whole rest of the world that is not anywhere near their peak yet.

Zvi:I think they're a few weeks behind at most in most places. But certainly in New York, the peak is here right now. But my understanding is that right now we're a few weeks away from peak, hospital systems are going to be overwhelmed in most places unless we had extraordinary success with the social distancing.

Robin: But if it was a factor of 10, you might still be tempted to use it even at the peak. If it was really a factor of 10 reduction death rate.

Zvi: If you were confident it was a factor of 10 versus what you would get otherwise. So one thing that I noticed when I was thinking about this carefully was that it's very important to understand what kinds of infections you would get in the wild. Like how many people low doses versus high doses versus giving everybody a low dose. Because you don't give everybody a high dose, no matter what policy you're in, unless you're doing something very malicious or on purpose.

Zvi: So one rule of thumb I asked myself was, what's the household size in the United States as an example, right? Just to think about the American situation. The average American household is 2.6 people. If you exclude children that can easily be held home, who also just don't have much risk because children under the age of 18 basically never die or get seriously ill from this to a first approximation, we’re left with a household size that averages about two. 2.02 I believe is what I got.

Zvi: So at that point, if we're worried about high viral loads being modelled as coming from familiar interactions inside whatever your lockdown, quarantine area would be. If someone else in it got it and then you got it from them, you're assuming you'd have probably a high viral load. And then if you get interactions at the grocery store or otherwise, trying to make your way in the world normally under a quarantine, you would then probably get a relatively low viral load. Right? Like the study with the measles with the kid who got it outside the home was relatively safe, right?

Robin: Sure. But I've only been able to find these three data points so far. Maybe there are more, and I wish somebody would go look for them. These three data points are about ‘in-the-wild’ infections and there are differences there. So the smallpox example is in-the-wild infections across a wide range of degrees of contact, was producing 20% to 30% mortality and then the deliberate ‘variolation’ they call it, was one to 2%. Yeah.

Ben: Can I just, as Zvi joins, to take a minute or two to outline all the main points of your disagreement, as well as this one.

Robin: Please.

Zvi: Yeah, in just a minute, but basically my opinion is that we are currently not in a position, I think Robin would largely agree, to just do this on a massive scale. We have an idea of “What if we did this?”, but we have neither the permission nor the logistical capabilities nor the understanding of how to safely give people reliable low dose. Nor do we have the extra tests to verify that they actually got infected, because if we do it right most of them will be asymptomatic and then we would have to hold them until we reliably thought they were immune. And then to have this work, they would have to be confident they were immune, which probably means getting antibody tests for them, which we also don't have. And all of this on top of having actively malicious FDA and CDC and WHO people trying to stop even ordinary mitigation and reasonable effects, let alone something like this.

Zvi: So I don't think that this is particularly actionable right now. Plus the problem of with a lot of low doses already, we're just not going to get, 3 to 30 is a misleading win. We're talking more like 1.5 to 10. And then you have to consider the fact that not having hospitals available to save you which, if we did this at scale, would just be true if we did over the next month or two in most places, would mean a factor that wasn't that much lower than that working against you.

Zvi: So I just don't think this is a practical thing to do right now because we just can't spare the resources in various ways to do it properly and we don't know how to do it. But I do think that we should absolutely run this and many other experiments that involve infecting people on purpose, or accepting that a lot of people who experiment on will be infected, in order to learn much more about the virus in various ways. And I think we'll probably have broad agreement about those policies when we discuss them.

Ben: I think the UK maybe had an experiment for some period where you'd get paid money, you could take the virus and they would do certain tests on you to try and test some of these variables.

Zvi: Yeah, I saw a little bit of that. I haven't seen any results. Do you know if there was any results?

Ben: I have also not seen any results from that.

Robin: I think people were paid to pick a different but similar coronavirus, so people weren't actually paid to take this coronavirus.

Zvi: Yeah. I believe that that was not the most helpful thing for us to do right now. I mean it might be better than nothing, but it didn't teach us the information we want to know. Like we could learn so much from so few infections, most of which would almost certainly end up being low dose in healthy people, and we're just not doing it and–

Robin: But the latest Nature issue has an article about, what they called, human challenge: what are the ethics of actually allowing people to be directly infected with the virus rather than infecting the whole population to get faster vaccine data. And they were not so sure it was ethical.

Zvi: Right. Which obviously makes both of our heads explode. But the word biological ethics, medical ethics, is almost an oxymoron when used in the wild. Just the things that they think about just don't make sense from the perspective of what happens to people in many cases. At least that's my perspective and I think it's probably–

Robin: That subject could take us the entire time if we–

Zvi: Yeah. I didn't mean to get off on a rant here, as they say.

Ben: And then my model is you're also under the impression there are a bunch of other low hanging fruits that should be a primary next steps as opposed to exposing the youth?

Zvi: Yeah, I think basically we are failing really hard on a lot of very simple things right now that we could push on that would greatly improve our chances of successfully squashing this, at least for a while. And that only after we've done those things and ramped up, so a few months from now, we should absolutely look at our data from deliberate infection tests and look at where we are and how likely we are to be able to contain this in the long run and economic costs of that and ask ourselves, should we be deliberately infecting now?

Zvi: And it's not obvious to me we should start with the young also. A lot of my intuitions say we should actually start with the old, because if we're accepting that everyone's going to get infected, the primary win comes from the infection being mild and therefore you want to save someone's life, you do it in an old person. A person who's 23 gets a mild infection, doesn't save that much life cause they were probably fine anyway. A person is 87–

Robin: So I did an initial analysis in a spreadsheet simulation and in that thing it turned out infecting the young was better. But of course, a more complicated model, including more things might have a different effect.

Zvi: I agree.

Response to critique on effect size

Robin: So it sounds like your main objection is that it can't be done now and or that it would have less effects that I'm claiming. So that second thing seems like easier to start with, but my-

Zvi: Yeah, we were getting into it and then we decided to outline things.

Robin: Well sure. But from my point of view, I just say we have limited data, but all the data we've seen has tended to see pretty large effects of initial dose and what they call a viral load. That is, there's a very strong correlation between viral load and symptoms and a great many viruses. And so there are these three data points I could find and they are a factor of 3 to 30. And if you're saying 1.5 to 10, [I assume you’re] just trying to say there must be a selection effect there so that I'm going to discount it on the basis of if people saw a smaller effect, maybe they didn't publish it or something. Is that-

Zvi: Well, no. Let's be very concrete about the example. I remember two of them very distinctly. The third one was the actual coronavirus. So there's that one, so my opinion-

Robin: No. One was smallpox, one was measles, one was SARS.

Zvi: Oh SARS. I forget what the details of the SARS one were. So the years of the measles one, I believe, was at a 14 to 1 difference between the first person to get infected in a home and everybody else?

Robin: Yep.

Zvi: And what I was saying was, you have to discount that by a factor of two because half of people will get it outside the home anyway.

Robin: Not clear.

Zvi: Well, I mean just because of household size. Like American, if you discount children, just the average household has two people-

Robin: But if we're all locked down then-

Zvi: You think we increased household size substantially?

Robin: I mean it's about close contacts. I mean a lot of people aren't going to be all by themselves, they're going to make visits once in a while to see a close friend or family. And those are the times when they also might get it close.

Zvi: Well, it's not obvious to me that visiting a friend is the same level of high dose as living with someone. I don't know. Obviously that's not the way it was done before. Certainly there's some possibility of that, but it's also possible that a lot of interactions will be less intense than they would have been otherwise if people were taking a lot of precautions. Like certainly a lot of places that people–

Robin: Okay. If we take a factor of 14 and divide it by 2 and get a factor of 7, I mean that's still huge, right?

Zvi: It's still huge, but the issue being that if we have a factor of five from the hospital not being available, those two almost cancel out.

Robin: Do we in fact have a factor of five for a hospital not being available? I mean that's less clear to me. This seems to me the biggest factor I can think of. What other factors are a thing?

Zvi: We don't know the number for sure. I know it was a factor of at least several. So smallpox was a difference between a deliberate minimal infection and the infections from the wild. Not just between a low dose, but like a deliberate low dose. And I strongly agree that if we can get to a super low dose through experiments that reduce our findings even below current low doses by a factor of this kind of level, then we should strongly consider doing it. The other question is do we think to a first extent everyone's going to get infected or not, obviously. But–

Robin: Right. I started out my presentation with that is the plan B sort of thing. We need–

Zvi: Right. We both agree that this is a plan B. We shouldn't just jump to it if we don't have to.

Robin: But we should go as far along to the path as needed so that we would be ready to invoke plan B. If we're not even ready to invoke it then we need to make sure that we get it ready to invoke.

Agreement: In praise of experiments

Zvi: Right. So I think the points that we violently agree is that you–

Robin: I'm saying if, for example, we allowed 10,000 people to be deliberately infected and then allowing people to vary the dose and delivery size, that thing, that would probably be plenty enough to learn about how to get a low dose and the right method, you would think. And 10,000 people is a pretty small number here in comparison to the size of the problem.

Zvi: But from the perspective of “if we didn't have people literally standing in our way with guns”, we should absolutely do this. We kind of do, right? We have people who are telling us we can't. This is very much not considered an appropriate thing to do and anybody who tried to do this in any kind of controlled setting, in any kind of public way, is gonna have problems.

Robin: Well, so I think in some sense, civil disobedience should be left on the table as an option. Not yet.

Zvi: If it would work.

Robin: But we should try to do this legally. We should try to do this with permission. But if we're talking these large factors of gain, at some point you might want to allow, or even encourage, people to do things that aren't permitted.

Zvi: Yeah, I think that if we can run the experiment such that we will actually get the data we want, not just haphazardly infect people, but like infect people carefully, collect the data, do the testing. Of course, we still have to get the resources necessary to gather the data, which would have to come from somewhere. But my understanding is at least some amount of testing is just not being allowed by the FDA. Like there's capacity to produce a lot more tests than we are and we're just not being allowed to do so because they weren't sufficiently accurate or whatnot. And we could potentially use those.

Robin: Although there could be a difference in permission in some university or lab trying officially to do a test, and a bunch of us just infecting ourselves but then collecting the data together.

Zvi: That would be the hope. But basically this is one of many experiments we are not doing, right? Like one–

Robin: We're talking right now about whether to focus on doing this. So we're asking, this relative to other things we could focus on, its relevant potential and priority and what we could be doing. So, I want to make the argument that even if you're not ready to pull the trigger, and I'm not clear that you shouldn't pull the trigger, we'll talk about that in a moment, that there's a lot to do here with high leverage that somebody, a foundation or philanthropists or whatever, could go a long way and do a lot to help with this just by first investigating legal permission and finding a way that somebody somehow could do it legally. And then finding a way to set up a place, even offshore, where you go do it for a small group so that we get the sort of data and the demonstration.

Robin: Cause I think, if you could even do it at the small scale and then publicize the results, that could go a long way toward enticing a lot of other people to consider it. So the limiting factor is that first demonstration.

Zvi: Right. I didn't mean that we shouldn't do it because we have so many other experiments we should run first or anything. We should run all of them, right. They don't preclude each other in any way. One of the few studies we have was a reasonably terrible, in many ways, study of 30 people in France, and that caused medications that people need to be just sold out everywhere, almost immediately, because it looked promising. So that's how much we... That's how desperate we are for data, right. Whether or not it ends up working.

Response to critique on timing

Robin: So there's this other point you made about it's too late. And so let's go through different scenarios and ask in which scenarios are [we] too late. So one scenario would be the current doubling every three days just considers until the entire United States population is done, and well, I guess, at that point, we're talking another four weeks, right.

Zvi: It depends on exactly where you think we are right now. My best guess is we're at several million. We're at 1% infected at this point in the US, so one, two, three, four, five, six. Yeah, we've seven doublings left then everyone's infected, so that's seven times two-and-a-half. Yeah, over two weeks.

Robin: Right, so that's one case. So in the case that basically most everyone gets infected within a month, it's very hard to do much of anything, of course, in terms of policy. And then you would have to face the question of do you want to try deliberate infection near the peak under the hope that it has a lower rate, on the basis of data that isn't as direct as you'd like, but it's still relevant data, then you'd have to be making a gamble. And so, even for those people, if we could just get data in the next few weeks that helped pin down – even just doing better literature review to find other historical cases. I mean that's a really cheap low bar.

Zvi: I mean, obviously we should do a literature review, right. If there are people who are just sitting around at home doing ordinary academic stuff, and you hire them to do a literature review – some billionaire should hire them to do a literature review. That's a slam dunk. I don't see how you could possibly disagree.

Robin: Right. But another scenario is the lockdown works temporarily. It doesn't squash it and make it disappear in a month, but it holds the line, or holds the line in some places, grows in other places, declines in other places. And now you're looking at a many month, perhaps even year long scenario of alternate intermittent lockdowns. And in that scenario, it seems like there's plenty of opportunity for deliberate infection, as a policy, even four months from now, to show up and have a big effect.

Zvi: Yes. So when I say it's too late, I meant it's too late to do it sort of to lessen the initial peak, right. The initial peak is going to happen, whatever it is, without any chance of–

Robin: If the initial peak is as fast as we fear. If we successfully lockdown enough to spread out the peaks substantially.

Zvi: Right. I mean I think the scenario where it ends up being sensible to deliberately infect people is in fact that scenario where we are able to somewhat contain it, buy yourself some time, ramp up our available resources, but it's not good enough to squash, right. And it becomes clear that we can't, right.

Robin: So, I mean, I'd say... Well, among the things we could do in the next month, while we're trying to squash, getting ready to have this available, should we partially squash, successfully seems a pretty high priority, the only other things would be is what would it take to actually make us successfully squash, but I mean most of us can't personally work on testing, and the government is supposedly going to set up a big agency to do tracing, and I'm very afraid that they will just be incompetent at that and make a bad job of it. But if we could help with that, that would help under that scenario. Because basically to do this successful squashing we'll have to not only lock people down long enough, we're going to have to set up a regime of very high levels of testing and very high levels of tracing, which probably includes a fairly high level of privacy invasion.

Zvi: My understanding is one of these things where you have a variety of things you can do. You sort of start with this R0 of around 4, which varies a little bit by place, and then you can engage in a variety of different actions to reduce that to help you stop the spread. You successfully go well below one, you win, you squash, as long as you keep it below one, you get to keep squashing. One of the things you can do is test a lot. One of the things you can do is do more aggressive tracing on people you've tested and invade people's privacy. You can also do things like get people to wear masks, get people to continue to socially distance to some extent, take other proper hygiene precautions, et cetera. You can also do deliberate infections, obviously. One of the things it does is it helps with that.

Robin: Right, but I'm asking specifically, well what would you put up against this research program as something that you think is something people who are listening could get involved with and do something with in the next few weeks.

Robin asks for an alternative concrete proposal

Zvi: Right. So personally what I have been involved with is actually trying somewhat to increase capacity/push the government towards taking better actions to increase testing capacity and actions towards treatment. And that has had some success because I happen to have some channels, but that's not–

Robin: So we're talking testing capacity or other medical capacities?

Zvi: We actually set up a lab to do tests and actually got people running tests that wouldn't otherwise have been run. So we did in fact ramp up our testing capacity somewhat. And then the FDA said you can't run that test. That is, of course, how it works. We also managed to get Cuomo to pursue antibody testing and convalescence here, but that's obviously uniquely good success. So a number of people in our circles have been pursuing messaging, trying to get the word out about various low hanging fruit that people could pick in the hopes that people would pick them. We also obviously could try to get word to people with resources to try and pick some of that fruit themselves. There are a lot of people listening, I think, including Ben, who have primary or secondary context of people with a lot of resources.

Robin: So I'm hearing a lot of meta things, but I'm less hearing specifics. But first you just make some tests. You say okay, you could be making... So lots of people are out there trying to make more tests right now. There's a world full of people trying to make more tests, and obviously, as you say, regulation is more the barrier.

Zvi: Well I mean I strongly agree. If it is feasible to run experiments on COVID-19 to find out more about how COVID-19 works, what leads to better or worse infections, what leads to what infection rate on what actions, so people can know what to do. That will almost certainly involve a deliberate infection incidentally, even if that wasn't the intended purpose. I do think that is all very high priority. I think that finding out information through experimentation seems to be by far the most efficient local thing that anyone can do.

Robin: Okay, again I'm saying, you're drifting into abstraction, and so I'm trying to pull you down to concrete. So, messaging and networking and experimenting, which is all at the abstract level, but this proposal of mine is relatively concrete. Yes. These meta things do help add leverage, but it's about a thing you can concretely imagine them. And I think one of the advantages is that, instead of just trying to encourage the government to do better, which is this amorphous thing where you're trying to pull all these levers and there's all these intermediate variables you can't see and that matter a lot, this is a relatively simple, straightforward thing to implement. If you say deliberate infection means you have a place, they show up, you infect them, you keep them there until their symptoms are subsided, or two weeks plus later, and then you let them go. And so it seems to be harder to mess up as a method.

Robin: And it's obvious that even if you do a bad choice initially about exactly how to initially infect, or even where to have the location, you would quickly learn that people don't want to go to that place, they want to go somewhere else. They want you to have better curtains, or beds, or whatever else it is. You would learn in the process of doing that and you would get data on which methods are better, most effective. So part of the advantage I'm suggesting here is yes, you could message about it. Yes, you could network about it. Yes, you could do all these abstract things that you can do for anything, but here is a concrete, simple proposal that's hard to mess up.

Zvi: Yeah. So in terms of concrete, simple things, I hear people just trying to physically get equipment to the right places and stuff where we certainly have much more leverage than people would think, because I've seen a number of–

Robin: So we're talking ventilators then?

Zvi: I've seen ventilators, I've seen masks, I've seen, in my case, tests. Actually some physical tests. We've gotten people tested who needed to be tested, out of thin air. But yeah, I agree that there's a lot of people doing that. That's ramping up on a logarithmic scale on....

Robin: Right. So in those categories, can you say how much effort is there and what percentage effect am I adding to efforts to get masks to people, or efforts to get tests to people? I mean you got to say you personally, or you, whoever's listening, is a small fraction of that, because a lot of people are...

Zvi: Yeah. It is a small fraction. I don't want to discourage people if they're considering doing those things, I think your leverage is better than you think it is.

Robin: Okay, but I'm talking about a thing that almost nobody's doing anything about, right.

Zvi: No, I'm going to get to that in a second.

Robin: So your personal leverage is much higher on this, I would say.

Zvi: Absolutely. So to get to that, so when I said experimentation, I was trying to be concrete. So you're proposing you're infecting people and then you see what happens to them as your primary thing. So if I was trying to just use what resources I had and was able to do things that might get people infected, I would be doing something that was trying to kill more birds with that stone. I would be saying let's study practical types of interactions that happen in real life but see what causes people to get infected, and what doesn't, at what rates, and then, once they're infected, let's not waste this opportunity to study what happens to them exactly afterwards based on those means.

Zvi: And then if we thought we had a way to introduce very, very small infections to them, and maybe get a good prognosis, and we had anybody run that experiment, then we should absolutely do that once we have the mechanical idea how to do that. I don't know how hard it is to try and get smallpox levels of exactly the right amount. My understanding is that it was an art form that was developed over a very long time exactly how much you needed to get the person to be infected at all and actually have an immune response, because if you do too little, nothing happens.

Robin: Sure, but with enough cases that would be an obvious parameter that you'd experiment with. I mean people wouldn't forget to get to look at that, of course. And within 10,000 cases, I'm pretty sure you'd get that parameter down.

Zvi: If you can get 10,000 cases. So the question then is do we have a reasonable path, right, to actually cause this to happen?

Zvi asks for next actions to take

Robin: So moderator, I thought we were initially thinking about 45 minutes and then a break. Is this about when you were going to do the break?

Ben: I was going to maybe head for another five minutes.

Zvi: I think we should keep going, given where we are in the conversation. But I'm very curious to hear how you think that we would do this.

Ben: Yeah, I know. That seems solid. Yeah, let's go for at least another five minutes. And yeah, go ahead.

Robin: So, I mean the fact about this proposal is that a lot of people hate it.

Zvi: Yes.

Robin: So that's both an obstacle and an opportunity in that if you understand that you don't hate it, you understand that a lot of other people are put off from it, therefore your participation and effort could have much higher leverage. This is something that I mean I found five other people who have written on the idea in the last four weeks, who have posted something on it, and they're not doing that much with it. So there are things you personally could do which would have especially high leverage, just because almost nobody is working on this. Now obviously one of the things to do was to focus on how to present it and how to get it legally and culturally framed and engaged so that it has the best chance, but you also have a high leverage with respect to that because so many people initially react to this so badly that almost nobody wants to talk about it. 

Robin: There's a research institute that I have an association with, and they were calling for proposals, and they were handing out lots of money to people who wrote proposals. They want to do something. But they rejected mine without comment because I presume they said, "No, we don't want to be associated with that." That's the kind of thing that's going on all over the place, which means that if you are a sort of rationalist who says, "No, but this makes sense," you are an unusual resource, and therefore your efforts could have unusually high leverage relative to masks, or ventilators, or tests, which lots of people get are a good idea, and for them it's just more a matter of how to do it. So that's my basic pitch is that, because the current effort is so small and the potential is so huge, you have a better shot if you are the sort of person who doesn't throw up when you hear this, of jumping in on this and trying to work with me or others and figure out a way to make it work. And then you asked me many questions about how would we do this, I don't know because there's almost none of us working on this. Come help. Come help figure out how to–

Zvi: So the proposal is not to spend a large percentage of the nation's or world's, or efforts on this problem. The proposal is to use our specific resources to try and just make it happen on the side while no-one is looking, kind of.

Robin: In the hope that, with a concrete enough demonstration and a big enough effect, you could then entice lots of other people to try it. I mean I think getting a large fraction of the population to do it isn't crazy, but it wouldn't happen on the basis of merely the evidence or arguments we have now, it would have to be mediated by concrete demonstrations. So then the question is how can we get those demonstrations going? And of course that's what happened with smallpox. Smallpox variolation, it was initially very controversial, and many people thought it was a terrible idea. Just the same sort of emotions that create anti-vaxxers, the horror at being infected with something that might be real was even more than, because this was in fact the real thing they were infected with.

Robin: And imagine people in 1700, the sort of ordinary people back then and how they would react to these proposals, a lot of them really hated it. Nevertheless, it became adopted in that standard practice. And in part because elite intellectuals, including famous mathematician Bernoulli, went and did a lot of careful intellectual work to demonstrate that it not only empirically worked out well, but in abstract modeling and analysis it would be good overall for society.

Zvi: Yeah, it was a very over-determinedly correct thing to do once they got–

Robin: Which doesn't mean it actually happens in our world, unfortunately.

Zvi: And it still was very hard, but eventually a lot of people ended up doing it, although definitely not all, true. And right, so the question is is there a practical path for us to pursue? And so your answer is that just you haven't gotten that far. You wrote a grant and clearly they were the wrong people to ask.

Robin: But I mean just more people thinking about this, even doing simple things as a literature review, doing some legal research to find legal scenario options, even just knowing what it would take to set up a facility like a hotel with medical facilities. Are there things like that? Could you use an existing elderly care facility for this purpose? There's just a bunch of things to just go do to try to figure out to see how we can make this work.

Ben: If I could jump in there, this seems like a good time to take a 10-minute break. Hopefully when we'll come back, we'll chat a bit more concretely about how we expect it would actually go down, and how likely we expect it to fix into the various win conditions. See you guys then. Thanks.

[Editor’s note: Robin lists things to do in more detail in the end of the second half.]

Debate Second Half

Brief recap on assorted topics

Ben: Welcome back. 

Ben: So I will just try and summarize what I understood both of you to be saying, and then we'll see where to go from there. Robin, I think you opened by saying two of the main things that seemed relevant to you were that infections in young people tend to be much less severe, on the road to herd immunity, and in general, this means that infecting them is going to have much less bad outcomes than infecting old people. And the other important part was that you realize this important point about the viral load hypothesis, that it seems quite plausible that ways that many people are getting infected is being infected with a very large viral load, perhaps in the house with someone they live with. And if you can infect people with a very small viral load, this tends to have much less severe outcomes. And so we could systematically make sure that everyone who gets it has a much less severe version of it. So then you started with that.

Ben: And then I think Zvi came back with sort of three main points, which were that we're heading towards the peak sort of continuously, or to a major peak, and the last thing you want during the peak is to have more people infected, just when the hospitals are overrun, that's just going to be the worst time to have more people infected. I think the second point was that we already have a fairly limited medical capacity, and it is unclear that we actually have any slack at all to execute this sort of thing. And then the third point, which was discussed which is slightly meta, which was that people are currently coordinating around important low-hanging interventions, like more testing and using masks and so on, and that has currently not successfully been effected, and that's where he feels much more effort should be put.

[Note from the editor: I failed to mention the important discussion about the effect size. Oops.]

Ben: And to which I think Robin replied talking about this is a sort of weird idea that if you're the sort of person who can deal with weird ideas, it is quite plausible, but actually your leverage here is just way higher compared to those other things that a lot of people in general seem to be able to get behind. That's my current summary. I'm curious, were there any key points that I missed or did I mischaracterize some of the things you guys said?

Robin: So just at the beginning, my initial point was just deliberate infection plus isolation allowed you to move some of the cases away from the peak. It was less about young versus old. Given that you're going to do it, it turned out in my analysis to be better to do the young, but the main point was that you could flatten the curve via deliberate infection, even if there were no dose effect. But then I recently realized the dose effect. Anyway, that was the one correction I'll make.

Ben: I think so. Zvi, did that seem accurate?

Zvi: Yeah. And I respond to that with basically that we maybe could have done that if we had done it early and done it properly – as in the quarantine didn't break it all. But I find it hard to model, that by doing it before we hammer the current peak, it doesn't seem like it would be useful in that light.

Ben: So your lesson was that… Sorry, I just want to finish something first. Sorry this is a dumb thing, but your lesson is we should've listened to Robin earlier when he wrote the blog posts.

Zvi: Oh definitely, we should listen to Robin when he says something unique and new, and think about, because again, we're the people who are capable of thinking about such things. I like to think that when I say something, or many rationalists say something unique and weird, it's worth spending your time to think about it, even if most of the time you think oh, that's wrong, and go about your day.

Ben: Sorry, I just wanted to add a few more things. So we will move to a Q&A shortly, in maybe somewhere in the next 15 to 30 minutes. And if you want to have a question, the correct place to put that question is either in a comment section on LessWrong, on the event for this post, or to PM user Jacob Lagerros, his username is JacobJacob. And if he picks your question he'll send you a Zoom link and you can join in on this call and chat with us.

Zvi: Okay, cool.

Ben: Yeah, that's modern technology. It's alright.

Zvi: I thought we'd just be reading off the questions. This is much cooler.

Ben: Yeah. We’ll try it out. I think it's going to work all right. It's like radio. 

Ben: So my main three senses of places that we could go further, which I want to say, and then you guys can figure out what seems juicy to you. And I think there probably is disagreement between you about how likely the win condition is going to involve substantial amounts of deliberate infection, whether it looks like herd immunity or whether alternative win conditions are necessary. 

Ick reactions to the idea of deliberate infection

Ben: I think another thing… so Robin talked about the ick reaction and I think one of my main guesses is for why people have an ick reaction is due to the potential for major downside from deliberate infection, the potential for either we do it in a bad way that it causes containment failure or just we infect a lot of people when there in fact was an alternative method that did not involve herd immunity and did not involve causing everyone to actually go through the disease.

Ben: And so there would just be a massive number of deaths that would not necessarily need to happen. So that's my sense and – Zvi?

Zvi: I mean I would say that, and I think Robin probably echos this based on what I've seen him write and say mostly, that those are not the main places the ick factor is coming from. Like those are the places that, sort of if you're trying to come up with reasonable objections, reasonably reasons people might be icked out, as opposed to the actual reason people get icked out.

Robin: There is a place for responding to excuses even when they aren't the real reasons. They often force people to give a next excuse and a next excuse, and you finally maybe eventually get to the real reasons.

Ben: But I do expect with some people, I agree that many people's reactions are not very thoughtful, but I think there's still pretty good solid reasons for some people’s responses.

Robin: But I actually agree with Ben that in many, even rationalists’ minds, they have hope. And this in some sense feels like giving up on the hope, that is once you just try to infect many people, then you're giving up on the vaccine that will save us all or the squash that will kill it all. And they put a substantial hope on those things. And this feels like giving up.

Zvi: Yes. And you're not allowed to give up. Like that's really terrible to give up on human lives.

Robin: Or to make a plan B in case you fail is even really terrible.

Zvi: Yeah. Making a plan B is already like escapism and pretty terrible but certainly actually doing it before we're sure we didn't have a way...

Robin: Not only making a plan B but exploring and putting resources into figuring out which plan B would be better.

Zvi: Well because like you're stealing, I can just imagine-

Robin: Stealing resources of plan A.

Zvi: Right. Like people are dying because you didn't give them tests. Like how dare you sir. You could have built a hospital instead of building this research clinic.

Ben: I see, so I do want to check, so that is an important question, whether it is giving up hope or whether it's a serious argument that this will not be necessary and it will on net just actually be damaging to go down that route.

Is deliberate infection worthwhile near the peak?

Robin: So I had a couple of things that I thought we might be able to fit in. One is I just wanted to make a list of all the concrete things we could do because I realized we didn't include all of them. Another is we might be able to go over the bad arguments we agree are bad and that might be a connection we can make. 

Robin: And then the other thing I was thinking was the argument that... I say the data range is a factor of 3 to 30, so say it's only a factor of 3 in mortality reduction and say we're at the peak. Well this mortality reduction goes with all the other symptom reductions. That is all the way along the chain to death, you're also having weaker symptoms and less problems. So I would say if you take three people and you deliver the infectum, that's the equivalent of one ordinary person in terms of medical demand. So if you think most of those people are going to be infected near the peak anyway, it's not clear to me that even near the peak you shouldn't want to do deliberate infection in order to get that factor of 3, to cut down the medical demand by the factor of 3.

Zvi: If those resources don't have a better use somewhere else, including isolating people, my model says even if we're at the peak it's very possible to actually protect yourself if you're willing to take proper precautions.

Robin: Well, so my assumption is that you infect plus isolate. So my general policy is that in order not to accelerate and make the peak sharper, whoever you deliberately infect you immediately isolate. And that, it's in my calculations, my simulations, I accepted that displaces one other person who might be otherwise isolated. So then I'm not just assuming more isolation.

Zvi: Right, the alternative is to isolate somebody who is not infected at all and have them just stay safe.

Robin: Right, but I was focused on the medical resources argument that if the argument is that at the peak you don't want to be doing this because you're taking away, you're adding to the load on medical resources. I'd say, well no you're not. That is, deliberate infection doesn't actually accelerate, doesn't actually sharpen the peak. It actually can flatten the peak and again you'd be swapping three people who will soon get infected anyway with one person who's now infected.

Zvi: Well, it depends how long the peak lasts, right. If the peak is only a month, then by the time you let this person out, they could have just never been infected and the peak is already over and you've taken some amount of resources to help them. Whereas if the peak is going to last for longer, you can rotate people in and out and you can actually make progress, right.

Robin: I mean, there are different kinds of quarantine or isolation resources depending on whether you're trying to prevent infections in or out from it. People you deliberately isolate, and in fact you're not so much trying to prevent infections coming in, you're trying to prevent infections from going out.

Zvi: Right, after the first day or two where it could increase the viral load, you're like, okay, it's done, you're infected.

Robin: Right, so that's a different kind of resource.

Zvi: No, it's a cheaper isolation is the argument because normally you have to–

Concretely discussing what deliberate infection looks like and comparison to work on isolation

Robin: You can basically have a hotel of people where they all come in on the same day and they all get infected the same day. They can wander around and meet with each other and do everything they want. They can all be in contact with each other, you can even have a university campus that way, armed guards around the edge. But everybody comes in on the same day.

Zvi: It's a little trickier. You still have to verify they're infected, right, because otherwise they'll get high viral during the secondary infection.

Robin: Right. So, the first day or two is the, they're isolated and individually infected. But then...

Zvi: Well, how fast do the tests work? My understanding is the tests take several days before you will get a positive swab from them because you haven't had the virus multiply enough yet. Is that right?

Robin: Well, I mean in this case we could just say it's a week of isolation plus a month perhaps of sharing isolation.

Zvi: Right. The first week is full cost, then you have to burn some tests and then you have a cheaper isolation after that.

Robin: Right.

Zvi: Sorry, it's not clear to me... Like my isolation for example, right, how expensive you should consider it? If you actually know how to isolate like how hard is it? Just like...

Robin: Well, but under your assumption that isolation isn't that hard, then by assumption this peak is not going to be all in one month. It'll be spread out perhaps over two or three months. Right. Your assumption about the feasibility of isolation is directly connected to your prediction about how sharp the peak is.

Zvi: Right. It depends on... It's possible for some. Certainly one way one could invest if one had a lot of resources would be to actually improve isolation procedures to allow more people to be isolated and potentially flatten the curve.

Robin: You know, for example, effective delivery of groceries or mail or something, if somehow it was easier to make that more clean then that could be a–

Zvi: I certainly think there's, I certainly think that's another place that we could talk, we could think about, because nobody's really looking at it very hard right now as far as I can tell. And they're not scaling up very fast and obviously it's not the same thing where like we could convince everybody to suddenly do this giant thing that would have a huge impact. But certainly it's a possibility.

Robin: I mean, again, I'm certainly not going to say the thing I'm proposing is the only thing worth doing, that certainly sounds worth doing. And in fact, in my experience I see all these firms telling me that they want to deliver me food and things and telling me they're being clean and just having some third party independent verification or certification of clean delivery could be a huge contribution. Maybe it's not so much some clever way to do it but just certifying that it is clean.

Zvi: Yeah, that would certainly be good. Also, just ramping upscale. Just to speak from personal experience, we can't get delivery here because all the delivery slots have already been taken and as fast as they're appearing, they're just gone because of course price mechanisms, who knows what those are. So it's just a matter of furiously trying to click, so we're going to get you to pick up instead in the hopes that we'll actually be able to do that and then that's still, an order of magnitude better than going into the store, maybe two but like not great.

Robin: Right. Well, we can expect ordinary interest and market mechanisms to, over the next few months, to be honing and improving isolation mechanisms, right. And so if they succeed enough, then that will flatten the peak and that means we're not talking three weeks until the peak here. Maybe we're talking three months.

Zvi: Right. I have a model where super spreading and different people taking different behaviors matters a lot, and so there sort of will be a peak amongst people who can't isolate. Where in the isolated people will actually ride it out. And then the people who aren't isolating will largely have recovered or have gotten sick and then situation changes.

Robin: So the more this is spread out, the more an opportunity it is, at least for the last third of the distribution, for it to get deliberate infection.

Zvi: And then they can certainly consider it. So I think we strongly agree that if there was a practical way for us to be allowed to and actually run this experiment, it'd be much better... It would be a very good use of resources. I think we both agree there’s also a lot of other similar things you could do if you had the ability to engage in medical experiments on humans with live COVID. That would also be well worth doing and we can talk about exact procedures but...

Robin: Well, so I mean this is simple enough that the question might be: do we necessarily need an officially sanctioned large and sponsored and run by a big major organization sorts of experiments or what sorts of smaller scale, more amateurish experiments could be done here that might be sufficient.

Zvi: Yeah, I mean I'm not a legal scholar, I imagine that if anyone gets sick, you're in a hell of a lot of trouble, and you might even go to jail even if no-one does, because this is just...

Robin: I think you're especially probably in a lot of trouble if you're trying to sell this. If you're a for profit business and then you would be sued to death.

Zvi: Oh, no. Yeah, a for profit business, no way.

Robin: But if you were simply amateur people helping each other, then it's less clear that there would be much... That it could be done. Like if you help, showed somebody a kit for how to do deliberate infection and you posted the kit on the web and then some people decided to use the kit. Could you really be sued for them deliberately infecting themselves. Taking that risk might be worth it if you're just one person who– what can they get if they sue you?

Zvi: I mean, I would not be worried about being sued, I would be somewhat worried about being sued, I'd also be worried about being criminally arrested.

Robin: Okay, but if we're talking, even the United States, I guess I recently did a bet with someone about a quarter million people in the US dying. If you could actually make a substantial cut in that, is that worth civil disobedience at some point. To hell with the rules.

Zvi: I'm not going to tell people not to do it if they're willing. I agree. I just, I'm trying to find a practically... Obviously also if you want people to report on and believe and use your findings, you really want to be able to do it as above board as possible, obviously

Robin: Of course, but there's a trade-off here, but again, if there's a choice between doing something soon and fast and getting it done versus waiting six months for regulatory approval…

Ben: So Zvi, I would like to just ask if you want to just take a minute to reflect and make sure you've hit all the key disagreements you had to bring up with Robins proposal.

Zvi: I would say I certainly had additional concerns about the proposal.

Robin: As do I.

Zvi: Yes, but I don't think it'd be necessarily a good use of time to bring them up right now. I think going to Q&A.

Ben: Yeah. And again, we'll go to Q and A in a minute. I will wait for Jacob to tell me that he's got two people ready, or got someone ready in the room.

Zvi: Right.

Ben: So we'll probably just go for another couple of minutes.

Things that can be done to help

Robin: I'd like to just list the things that can be done here.

Zvi: Yeah, let's do this.

Robin: Just in the hope of enticing someone to doing them. So I didn't mention before computer modeling, so a lot of people we know love to do computer things. So my little computer sim is badly done because I'm not in the habit of doing that a lot. But even just the question, would it be better to deliberately infect the older, the young if there is a large reduction due to dose is not obvious. And so it'd be worth doing some modeling to figure that out so people we know could do computer modeling of that.

Robin: Even coming up with a simple sort of dose experiment plan, if you were going to deliberately infect some people, just what would be your plan in terms of who gets what when and what you check. Somebody with competence in doing that should just figure out a plan. Those are two other things that I didn't mention before, but of course I mentioned before, some sort of facilities planning either for a legal thing you're doing above board to figure out what place you buy and who do you hire and what do you staff it with and what's the plan there, or some perhaps amateur coordination thing and what would be the plan there. Legal research about figuring out what versions are legal and then some sort of literature review to better pin down what our reference points for other data on low dosage effects in viruses like this are.

Zvi: Yeah, I would add to that. So we need people who actually know biology and immunology and how to work with viruses who can potentially explore how to do it, physically, if we actually-

Robin: How to create a low dose, what would be the best mechanical way to isolate it, save it, and then deliver it.

Zvi: If anybody knows how to potentially find leverage with people who are in regulatory agencies, potentially get approval, that would be potentially very helpful. One thing I was thinking was we could try and actually just gather data via, I mean it's not as good as doing experiments, but certainly we could try to use surveys and asking questions to try and find out more about the people who actually just naturally got infected and how they got infected and how that compares to how their cases went and just try to see if we can establish something that way.

Robin: Well, in fact there was a study I saw on China recently which looked at how people got infected in terms of the chance they got infected, but it just didn't look at death rates. So maybe we could just get their data and re crunch it and figure out how death rates relate to the mode of infection.

Zvi: Right. You look for more natural experiments, look for places where we can ask these questions.

Robin: But quite often you don't need a new experiment. You just need to go find old experiments people didn't analyze right and just get their data and re crunch it.

Zvi: Yeah. I mean that sounds promising too, and obviously if we wanted to do this with building facilities and such, having somebody who's willing to back it on a philanthropic basis would be wonderful. I think going to a grant is hopeless in this situation for obvious reasons. I don't think that-

Robin: Just, I wasn't applying for a grant, I was just applying for them to put my proposal in text on their website.

Zvi: I think even that's pretty hopeless. I think what you're hoping for is specific people who have... Specific people who have a lot of money, billionaires and multi-millionaires who-

Robin: Opportunistically find people who can help, obviously. Yes.

Zvi: Yeah.

Q&A

What could go horribly wrong? – from Oli Habryka

Ben: All right. I think we have a first questioner. So I'm going to invite them in the room and ask them to tell us who they are and what their question is. I think I maybe know who this person is.

Zvi: Oh, that guy.

Ben: Good afternoon Oliver Habryka.

Oliver: Hello.

Ben: Yo, what's your question?

Oliver: Great. So my question is, my current emotional status is something like this all... Something about this seems really good, but what I really want is two of you giving a bit of background on what could possibly go horribly wrong with this. Some of my intuitions are that this seems good, but a part of me just assigns some probability of this going horribly wrong. A case might be that our viral load… it turns out we mis-measure drastically and we increase it drastically. We completely block a containment measure. What are the guesses that you have for just what is the worst case that could happen? What is the probability of that and what are some easy ways we could mitigate that?

Robin: I'd say obviously in the amateur scenario, if you are trying to create samples and store them and deliver them, you could accidentally infect people with other things by doing a bad job of that whole process. So there's a reason why professionals produce medicines in the way they do, in the way they store them, and the way they deliver them. Amateurs trying to produce substitute, that would be a bad scenario, right. So I'd much rather have a professional do this than amateurs to avoid those sorts of scenarios, right.

Robin: You could imagine, they just infect it with some other random thing in the process of packaging it together and poking you with a needle or whatever they're going to do. You can just screw that up. That would be one. And obviously another thing is, say some church likes this idea, their pastor goes wild for it. They do a big thing and then they screw something else up and they all commit suicide and that's in the news... And this gets associated with some weird disliked group and then that's the death of it, right.

Zvi: I mean, certainly we have to worry about facing giant social backlash. Whoever tries it facing potential legal backlash, stuff like that. Shut down attempts. But when we ask... I think bigger, what could possibly go horribly wrong. So one thing that could go horribly wrong potentially is that the selection pressures of the procedure somehow cause the virus to mutate in a very nasty way that causes it to be harder to detect or last longer or something like that. Because you're quarantining everybody for... You're infecting tons of people and then quarantining them for just long enough that you're expected to die and then the version of it that can live for 50 days and still spread without being symptomatic suddenly gets out and now everybody has to quarantine for twice as long and it's a disaster.

Robin: Just to be clear, we're already risking those sorts of things with lock down. I mean people have suggested that...

Zvi: Potentially you're increasing the probability of that happening. I brainstorming things that could go horribly wrong. Right. Second thing that could go horribly wrong is as was discussed on Twitter this morning, if people aren't immune for that long, I think we both agreed that it's going to be more than a few months almost all the time, but if we infect everybody and a year from now a variation comes out because it's still endemic no matter what we do. And then it comes back and previous SARS in style, coronavirus style infections, in some cases prior immunity to other strains has made infections worse. So by infecting a larger percentage of people, even though they recovered, you could set yourself up for bigger problems down the line.

Robin: But that's a problem with any way in which most people eventually get infected. That's not particular to deliberate infection mechanism. It's just, if we let most people get infected then we risk that scenario.

Zvi: I mean I'm assuming with deliberate infection we end up infecting, I was assuming we ended up infecting more people than we would have otherwise.

Robin: I'm not assuming that.

Zvi: Well, right, I was thinking once we have the ability to infect people relatively safely, versus taking that risk, we kind of would keep going. But I don't know.

Robin: I mean there would be a moment in time where you could pull the trigger and then deliberately infect everybody, say within a month, if the vaccine showed up in the first week, then you would say, oh my goodness...

Zvi: Right. The obvious other thing that could happen is the vaccine shows up a month later or a treatment shows up later and then everybody goes, the horror, the horror–

Robin: We should've waited.

Zvi: And also everything like this has a bad name forever, right. The next time this happens that we have to do this, they don't do it. They probably wouldn't have anyway, but they even more won't do it next time.

Robin: That's actually a good point to make here, which is that the biggest payoff I think from doing this isn't for this, it's as a practice for the next time when it matters even more. That is, we will eventually have even worse pandemics and one of the biggest benefits of this pandemic is that we are forced to take this seriously and think through some of these things such that we will be more ready next time.

Ben: Did that answer your question as you wanted it Oliver? Is there any follow ups you'd like to ask, or any more specifics?

Oliver: I think that basically answered most of it. I think I would be interested in a rough probability estimate or something like that, or just an overall judgment. Like “Yep, I recognize that those are some downside risks, overall I think they're sizable, or overall I think they're not really worth exploring more, investing more.” Just to get a sense of where currently modular intellectual progress can be made in this topic.

Robin: Yeah. I'm just focused on the factor of 3 to 30, gain and reducing death rates. I mean that can buy you a lot of risk. You should be willing to take a fair bit of risk to buy that.

Zvi: Yeah. I think that the social slash legal risks of whoever tries it are very high probability, especially if not done very carefully or slash with official approvals. But I don't think that the downsides that I was talking about are very likely. I think it was just a matter of what might go horribly wrong and those are the things that I can imagine going truly horribly wrong with it. Or I think some kind of backfire that does reputational damage is not that unlikely either, but probably pretty big underdog relative to just not happening.

Zvi: The other question is how probable is, if we do this plan B, it's going to work and be a good idea, et cetera. I think that in the scenario where we are stuck in some form of lockdown for six plus months effectively, where we're doing real economic damage the whole time, the chances it's a good idea is pretty high. Certainly more than 25%, quite possibly more like half. I think there's a decent number of reasons why... That's assuming we could actually test it and then have a place where we could implement it without just being told no. Right. I think the most likely scenario is that–

Robin: Well, I should mention, it's a big world. There's over a hundred nations out there.

Zvi: Yeah. We might convince, like, Singapore to do it.

Robin: And just getting someplace in the world to try it would be a substantial gain from the point of view not of just saving lives but creating a precedent for the future.

Zvi: That's a fair point. I certainly think that if you got anyone to implement it, if it was right, it would be a pretty big deal assuming if it worked out and proved to be a win. I think that the scenarios where it's a good idea potentially are an underdog versus the scenarios where it's not. Where either it's all over too fast or we manage to control it, but I have very high uncertainty about all of that and certainly I don't think any of these numbers should discourage people from thinking about the problem because again, if you think of the upside is as big as Robin thinks it is. And I think that's a reasonable thing to think it might be, even if I think it's somewhat lower, it's still pretty big. And you think that you have a chance to move that lever somewhat, then unless you have a better idea, if you want to do something relevant to the situation, it's a reasonable thing to work on from a perspective of someone like us. That sound about right? Even if it's likely not going to work. A better question is how do you make it likely to get a chance. How do we make sure that if it does work, that we tried it.

Ben: All right. Thanks Oli. Thank you very much and I'm going to kick you out of this call.

Zvi: No more Oliver.

Should effective altruists spend time on Covid-19? – From Daniel Filan

Ben: Remove. All right, next questioner, enter. Hey Daniel.

Daniel: Hello.

Ben: Which country are you calling from? I'm going to guess Berkeley.

Daniel: I would actually consider Berkeley to not be a country and say that I'm calling from the United States of America.

Ben: Okay. You win that one. What's your question?

Zvi: I'm not sure he's right. Go ahead.

Daniel: My question is, effective altruists up until recently spent a bunch of their time worrying about the world ending or about saving however many animals from dying per year of factory farming and all sorts of things. This disease sort of, if you look at it, it seems like maybe a couple percent of the world could die from it. That'd be bad, but it doesn't seem like, naively, it doesn't seem to be on the same scale as everybody in the world dying, or on the scale of other things that effective altruists typically worry about. So should we really be spending our time thinking about this and debating and such.

Ben: Okay. Thanks, thoughts guys?

Zvi: Yes.

Robin: Was it a yes or no question? Yes.

Daniel: Maybe why is the question.

Zvi: Yeah, do you want to go first Robin?

Robin: I mean obviously the trite answer is you should multiply the probability times the leverage you have and stuff that you can do, that this is right in front of you and so it makes sense to... Even effective altruists shouldn't just focus on ending the world scenarios, they should look at the wide range of large versus small scenarios and look opportunistically where they can help. But certainly one of the biggest ending the world scenarios that people have focused on over the years is pandemics and this is a pandemic. This isn't a world ending pandemic, but the habits and approaches that are legitimized in this case and that are developed in this case will be the main resources we have if and when a really bigger one shows up. So, it's well worth making this one go right just so we could be better prepared for a much bigger one.

Zvi: And also, so other things to point out, one of which is I think this is more of an x-risk or approaching x-risk, than you're giving it credit for. In the sense that for example, this could lead to a US–China war, which could be very, very bad. This could lead to lasting severe economic damage, which would be very, very bad. Potentially even collapse of civil order under some scenarios. And if we don't get immunity for very long, this could be an actual civilizational threat. If this comes every year and is this bad because we can't shut down the economy like this every year, and it's not clear that we can sustain ourselves. So there are risks.

Zvi: But more than that I would point out, x-risk is a small percentage of what effective altruists worry about. Some of them worry about animals, some of them worry about malaria nets and giving directly to poor people to make sure they can start small businesses. And those are two of the big networks, both of which get a lot more money than x-risk, as far as I... If my model is correct as concerned. So I don't see why this is remotely inappropriate and in fact, I have been very disappointed and puzzled by the failure of effective altruists to move a bunch of resources and money towards these problems when people in our community were very, very good about sounding the alarm that this was coming. We had a lot of opportunities, still have a lot of opportunity and where is Open Phil?

Ben: Let me briefly mention to listeners, please go to LessWrong and Jacob is pinging a bunch of you. Please refresh the page to check whether you have notifications, et cetera. What did you want to say Dan? Was that, are you now convinced that folks should now be spending a bunch of time on the pandemic thing?

Daniel: Kind of. I just wanted to say OpenPhil really is funding some forecasting work and, I don't know, they're not doing nothing.

Robin: So, I wanted to mention that in most other areas of altruistic concerns, one of the big issues that come back again and again it's governance. People have various ways they hope the world can be improved, then they face governance limits. The governance gets in the way. Governance isn't there to solve the problem. We will be changing governance in response to this crisis. We already are changing governance in response to the crisis. How governance changes in response to this crisis will be crucial in terms of whether it opens it up or shuts down ways that we could coordinate on other important questions.

Ben: Yeah. I wanted to briefly add that we're currently finding out how good our institutions are at dealing with a pandemic and a lot of people are currently quite surprised that the CDC is not preparing well enough and encouraging people to just do basic things in a timely enough fashion.

Zvi: The CDC, the WHO, the FDA, have all done really badly. Another thing before we get to that, I want to point out that a lot of EA's often talk about meta EA work about raising the profile of an attractiveness of and status of EA to get more resources, to get more people working. And if EA proved instrumental in helping with this crisis, then that would be a tremendous boost to EAs profile, and EA's ability to acquire resources and goodwill and so on. And so, even if you didn't value the lives of the pandemic, I think you should still work on this.

Daniel: Yeah, I would say that checking how good our governance institutions are is like, maybe we don't need to do it right now, but yeah, the raising-EA's-profile by helping out in a thing everyone wants help with, that seems pretty plausible.

Ben: Cool. Thank you very much Daniel. I am now going to kick you out.

Daniel: Okay. Bye.

Ben: Farewell. All right. Next question is still coming in a few minutes. Is there any other points you wanted to come back on? 

What probability does Zvi assign to us needing to deliberately infect? – From Ben Pace

Ben: I guess I still was interested in trying to get your sense Zvi of the probability that the successful outcome here involves a herd immunity strategy that is greatly informed by doing a bunch of deliberate exposure versus the likelihood we will end up with a different sort of win condition that did not require that.

Zvi: Also, I say that sort of you have to separate it out into, conditional on the world stays otherwise unchanged and we find a way to give the world that option and they would take it, if it was the right thing to do, versus what's the chance in practice that it happens? I think that second number is very, very low. I think that in practice everybody hates it. We do not have that many resources and we are unlikely to be able to make this happen. Not that we shouldn't try, but that conditional on it being right, it's still a large underdog, like 10% or less to happen.

Ben: What is your answer to the first question?

Zvi: Well, the answer to the first question... So I'm very much less confident in that answer, but I would say maybe roughly 30% that we end up in a scenario where deliberate infection of non-trivial numbers of people is the right thing to do. And not just people in key specific conditions, but like a large percentage of the population, would be the right thing to do if we could pull it off. I also worry that just our civilization, the same way it wasn't able to stop the pandemic in the first place, just doesn't have the coordination to pull this off.

Robin: The scenario that I find most likely and disturbing is that all the hope that's generated lately for a lot more testing and a lot more tracing and a lot more isolation, the same government agencies are going to be implementing that, that already implemented the failure before, and they just won't pull it off. And that'll consistently happen all through the West and therefore this will just get really big. And then it'll go back and re-infect even China and Korea, et cetera, who had been keeping it out until a much larger way of comebacks comes back to hit them. And then there'll be this vast amount of death and all this finger pointing and all these recriminations and all this blaming. And that's a pretty bad scenario, but it seems to me the most likely one. I'm happy to support, hope and to help the hopeful scenarios try to develop, but I still can't put a huge probability on them.

Zvi: I think the reason I'm more hopeful than you are is because I don't see this as primarily resting upon the government being competent because I think that the people and the corporations and the non-governmental resources will pull this together and get us through this in a lot of scenarios, despite the government, right? Through the government's resistance. That's kind of a lot of what's happening already.

Robin: But in that scenario, counterfactually we have a flatter peak than we otherwise would have. But still, most people get infected and an awful lot of people die, basically. You're cutting down the extra deaths that would've happened if everybody got sick in the same two weeks because maybe now everybody gets sick in the same two months or three months. But still, basically, most everybody gets sick and half a percent of them die and–

Zvi: I mean when people isolate, mostly the government can tell them to isolate and it seems to be pretty bad at getting them and forcing them to isolate when they don't want to. But it has the persuasive ability of the media and the people and whatever, how did that convince a large number of people to isolate and that number will go up. And I do think the people who isolate will be mostly safe.

Robin: I mean they'll be safe for a few months, but that's not the same as safe. If a vaccine takes two years or five years to come, I don't see how they're safe for five years.

Zvi: Yeah. I agree that five years is a long time, but–

Robin: Two years. I don't see how you can keep them safe for two years. All these people with, even yourself, isolated. Two years of staying isolated, the whole economy adjusting to isolation... That is a really tall bar.

Zvi: So when we talk about testing capabilities, right, that's a key part of this. I see the government's role in testing as not stopping it from happening, as what they have to do. The government doesn't create tests.

Robin: Okay. But in equilibrium, on average, how often is each person tested, do you imagine?

Zvi: So I imagine–

Robin: Once a week? I mean, come on.

Zvi: I think maybe more–

Robin: Once a year maybe.

Zvi: I think maybe a lot more than that.

How soon to herd immunity?

Ben: All right, the people coming on is kind of slow, so I'm going to ask a few questions that people submitted. And then if anyone else joins in we'll add them, but otherwise hopefully we'll have it more streamlined for next time. So I've been sent a few comments. So someone has said, "I think we are unlikely to hit herd immunity levels of infection in the US in the next two years. I would like to see Robin and Zvi discuss whether they also think that or not, since this bears on the value of Robin's proposal and lots of other things."

Robin: Well, that's what we were just talking about. How plausible is it that people can just stay isolated for the next two years? Well, a majority of the population stay isolated enough for the next two years not to catch this?

Zvi: Well, I mean the scenario that I think is most likely is that a significant number of people get infected but not enough to achieve herd immunity, if we completely returned to previous activities that people still attended their sporting events and their weddings and their buffets and so on, at the same frequency as before and shook hands all the time and so on. But we will make a significant number of adjustments on those levels, probably including getting behind wearing masks. I think that's probably a few weeks away at most at this point. It seems like people have gotten over their initial, "Wearing masks makes you a pariah," thing. And together with a lot of people in potential super-spreader positions being immune because they already got infected, and people taking various degrees of precautions, I think we can get R0 below one. And then it's not that hard to contain outbreaks as they happen. I think that's the most–

Robin: That just seems to me a crazily ambitious scenario. I mean, I'm just very impressed by how complicated the world is, how uncoordinated the world is–

Zvi: Well, the reason why I'm confident in it is because I don't see us as having to check off the following six boxes. I see it as a mathematical formula, where you have to do enough things to get below R0 = 1.

Robin: So say some places are below one and others are above one, unfortunately the nature of exponential weighting is that even if 85% of things are below one and 15% are above, on average you still have growth. So it's that last percent of the deviance to really put most of the weight on, because this is an exponential growing thing. So it's not enough that half the people, half the places have an R0 below 1. That doesn't do it.

Ben: All right. 

Why isn't China doing deliberate infection?

Ben: Let me jump in and ask one more question before I think we're going to get another audio person. Someone's there saying, "As an authoritarian country with not much consideration of medical ethics, China is able to try the deliberate infection rate. They also have plenty of medical expertise, so this idea was probably considered. Furthermore, if it was successful, I would expect them to announce that or at least for the information to be leaked. Is there a plausible story here for why we haven't heard about it yet?"

Robin: Well, I mean, they're humans with the same sort of human reluctance to consider various areas as we are, so the average Chinese isn't that different from the average person of us. It's more, does their structure select out and generate it more than ours might? So I could believe that somebody in the Chinese government has generated this option; there's a white paper or something that's submitted to some higher levels sitting on somebody's desk. But then I presume the reaction above is, "Well, at the moment we've locked this down. This isn't ..." I would think the moment when they were considering that would be when they were facing a wave from across the shores, where they were feeling like, "Either we lock down our borders and then we're really hurting our economy or we expose ourselves," and that's the moment when they would consider deliberate infection. But they, like us, probably would find it looks bad PR to talk about it, so they would probably be privately setting it up as an option to consider, but not publicly talking about it. Because if they weren't going to do it soon–

Zvi: They have less barriers in some ways, but definitely not no barriers. And I think they'd have to consider whether or not trying to do this would already be a disaster for them because I think my motto of–

Robin: Their story is they beat it, right? Their story is they fought it and they beat it, so they have to give up on that story to go with deliberate infection. At some point maybe they'll do that, but they don't want to do that now. They like this story.

Zvi: The story came out two days ago that the Chinese movie theaters got shut down and a lot of people were like, "Oh my God. Maybe China is just lying about everything and actually things are terrible." I mean the story about urns in Wuhan, it implies that maybe they were 46,000 deaths and-

Ben: All right, let me introduce another guest. You may have a sense of who this person is. 

What does Zvi think the effects of this negative shock will be on our institutions? – From Alyssa Vance

Ben: Hello, is Alyssa there?

Zvi: I've ever seen this person in my life.

Alyssa Vance: So, Zvi, you wrote in your Moral Mazes sequence about the impact that negative shocks have on society, and how in past social equilibria negative shocks were important, and in some sense necessary, to stop institutions from becoming sufficiently bad. Obviously, this is a large negative shock. Various people have said that it's the biggest shock to the world, the Western civilization, since World War II. Last month, various people were pontificating about how, "Oh you know, this might be the end of the Chinese regime." Now various people are pontificating about how it might be the end of the West. If people can switch their stories so frequently and not explain why the previous story was wrong, I'm not sure I trust them very much. Do you have any specific thoughts on how this particular negative shock will impact the particular institutions that we have? How this is an example of the previous theory?

Zvi: Right, so the thing I was saying when I got sort of cut off was that the Chinese might not be considering deliberate infection because they think that if it was so bad they needed to do deliberate infection, that their regime would end regardless of what they did, so why plan for it? I think sometimes authoritarians just sort of do things like that. "Well, if it's that bad, we just have to suppress it even harder/pretend it doesn't exist and hope for the best." But I just don't have that great a model of China. In terms of what might happen, a lot of companies are going to fail. That much is very obvious, right? A lot of small businesses will fail. A lot of big businesses will almost certainly also fail. The bailouts are not that big and won't save them if they're inherently unprofitable in any way.

Zvi: And others will take their place, and a lot of that could be very good, and there'll be a lot of degradation of trust in a lot of major institutions because they lied their asses off to us and proved themselves incompetent. And in some ways that's bad, but in most ways, in the long run, that's very good. And in general, it could be seen as creative destruction, the equivalent of urban renewal, just like a bunch of stuff got torn down and it's sad that a lot of people died, and sad that a lot of people are broke and don't have money, but when things come roaring back, there's an opportunity to do more innovation and change.

Zvi: Obviously, we'll be much more receptive to doing things remotely in various ways. Zoom is up 8x, not just because they think they'll make money in the next three months, but because once people will start doing Zoom, they'll realize we can talk like this and then maybe they'll keep doing it. I think that like the predictions of the universities dying are plausible because it's a lot of money to pay for something. If you learn, you can get it somewhere else. And similarly, a lot of people will realize that schools were a terrible idea when they don't have them for a while, I'm guessing.

Alyssa Vance: Do you think universities and schools will be... So summer vacation, or what would have been summer vacation, is coming up for two months. Do you think everything will be shut down past then?

Zvi: Yes, I think certainly there's almost no chance they'll reopen this semester. I think that is basically null. Or if they do, they'll be like... Yeah, I think Liberty University reopened because they're insane.

Robin: The question is August. Will universities reopen in August.

Zvi: I think there's a decent chance they can open in August. But there's also a decent chance that a lot of people don't want to come, even if they try to open in August because they've learned they can... People have summer breaks all the time, but when you actually try to learn from home, you realize that... A huge percentage of what I did in college was a mix of opportunities for socialization I didn't use because I was an introvert nerd and reading books and working through problems. And I can do that at home.

Robin: I think I disagree a lot with this one with Zvi, probably more than anything else in this entire thing. I think universities are going to do just fine. People will realize they missed their friends and hanging out with other people at school and they will want to come back and do it. The kids don't like staying at home and not seeing the other kids. They want to be with other kids, but that–

Zvi: I hated school so maybe that's biasing a lot of my perspective.

Robin: Oh, kids hate school, but they love to be with the other students. They want to hate school together. They hate to hate school alone.

Zvi: Yeah. It seems like you should be able to goal factor this though very effectively. College is very expensive.

Robin: I was going to make the comment about... we like to think that we're going to collect evidence here and then we'll find out better ways of doing things. But one of the big things that's going to happen is people will be desperate to construct stories about who wins and who loses and who was right and who was wrong through this whole thing, and those stories don't have to be accurate. And there'll be a lot of collateral damage from the stories that when crushing some true things that aren't convenient. So even if, for example, deliberate infection showed that it worked, if the best story that came out that won didn't like that, it might well say it was bad. And if you just look at history of past crises and see what actually happened versus the stories people tell afterwards, it's not that strong a correlation.

Zvi: So I mean there's also some other economic impacts. I expect VIX to be permanently higher. I expect loan options to be permanently more expensive or at least for a decade or two. I expect companies-

Alyssa Vance: VIX is the Volatility Index.

Zvi: The Volatility Index, like the people's predictions, to be higher. I expect options that are well out of the money to be more expensive for at least a decade, probably longer. I expect our companies to be much more cognizant of their supply chains and how to work from home and what would happen if something similar like this happens again. In general, I expect people to actually prepare for such a thing. They were prepared for the last war, even if they don't prepare for the next one. Right? Yeah. So there'll be a lot of that, and probably a lot of international trade.

Robin: And travel.

Zvi: And travel. Yeah.

Ben: Okay. Alyssa is that a solid answer? Is there anything in particular you wanted to push back on?

Alyssa Vance: I mean, you could go on forever, but that was great. Thank you.

Zvi: Yep. All right.

Ben: Yes, we could go on forever. Cool. All right. Thanks a lot, Alyssa. All right, I'll invite the next questioner in and then probably this will be the last question and we'll wrap up. All right. 

What specific information does Robin think we can get from deliberate infection that we cannot get from natural data? – From Lotus Cobra

Ben: All I have down for this person is Lotus Cobra, which, if it's a name, is awesome. Hey, are you there?

Zvi: Three two, one.

Lotus Cobra: Hi. I can't hear you so well.

Zvi: Try playing a land.

Lotus Cobra: Sorry. Hi, can you hear me now?

Ben: Yeah, I can hear you.

Zvi: Fantastic.

Ben: Feel free to say who you are and where you're from, insofar as you would like to give identifying information.

Lotus Cobra: All right, I am Lotus Cobra. Some people might know me as Bar Fight. I'm calling in from Canada right now, and I've been following LessWrong for about 10 years now, but I created an account like 20 minutes ago just to ask this question.

Ben: Wow, it's a momentous occasion. All right, please tell us your awesome question.

Lotus Cobra: Yes, I felt like this warranted it and, like Alyssa, I had to dig out an old computer to get this running so I couldn't get the video working. Sorry. This isn't a privacy thing. I literally couldn't turn the camera on.

Ben: No worries. It's totally cool. So what's your question?

Lotus Cobra: All right, so my question is, this one's more for Robin, although everyone feel free to join in. The deliberate exposure plan has evolved pretty substantially since it first came out in the middle of February. But right now, what advantage or what information do you expect to gain from deliberately infecting say 10,000 people with COVID-19, that we're not going to get from longitudinal studies, the natural experiments you were talking about earlier? And do you think this information is going to be worth all of the risks that we've gone over so far? The reputational risks potentially, but also the worst case scenarios. What if infection does not provide protection in the future? That kind of thing.

Ben: Cool. Thanks. Robin, what are the main benefits?

Robin: So, I mean, if you start with the cost. If you took 10,000 people, then the half a percent mortality means you're talking 50 people, right? So right there, if you screw it up, you will have killed more or less than the 50 people you could have, so that's a huge personal responsibility to take, no matter what. But of course–

Zvi: To be fair, if they're all 23, you [will only] kill one or two.

Robin: Right, exactly. So which case is more of a problem? You need a bigger experiment with younger people to see what affects the mortality rate, although you can track other symptoms. The main thing you hope to gain ... There's two main things you hope to gain. You want to know specifically how to produce a low dose infection reliably. You want to know how to take the original infectious virus, how to store it, how to transform it, how to deliver it to the body, how to do that reliably, and what the dose is that puts you in the right range of most people get infected but they're not infected very much. And then how can you tell us who is. So the other sorts of data is just not going to tell you that. It would take a long time with other sorts of longitudinal data to get you that sort of data about exactly how to infect people so that they have a low dose.

Robin: And the other thing you would get, which is perhaps even more important, is most of the world just does not believe abstract arguments. Most of the world needs to see a concrete demonstration of something before they're going to take it seriously. If we actually had it happening at an actual facility, even with only a hundred people at a time, you'd have newspaper articles about it, you'd have people talking about it, they could imagine that happening, and then they could debate whether that was a good idea, in a far more involved way than they would if we just have abstract blog posts. That would probably be the main benefit is just to say, "Look, this is concrete and it's real and it's a sort of thing you could do."

Ben: Cool. Thanks. Lotus, do you feel that was an answer to your question?

Lotus Cobra: Yeah, I think it does. I don't really have any other questions except to thank both of you for coming out. Take care.

Ben: Thank you. Cool. Thanks, Lotus. I'm going to send you out of the room now. Thanks a lot. And yes [Robin and Zvi], thank you for coming out, and by coming out we mean sitting in your homes and not leaving in any way. 

On what scale does Robin think deliberate infection will be needed? – From Elizabeth Van Nostrand

Ben: All right, I'm going to invite the last questioner who is on the call, which is Miss Elizabeth Van Nostrand. Hey, Elizabeth. Are you there?

Elizabeth: Yeah.

Ben: What's your question?

Elizabeth: So I don't have the sense of what scale Robin is talking about deliberately infecting people on. Is it like we do a couple of brave volunteers and they staff things? Is it we gradually do half the population?

Robin: Right, so that's exactly in a sense one of the main problems with trying to make policy proposals, is that that distinction gets muddled up. The simplest thing to talk about is the thing that would ultimately happen. But if you're trying to talk about the thing that would ultimately happen, people immediately respond with, "But look at all the uncertainties. How could you possibly know how to do that right?" And of course what you really want to propose is that you start down a path that could get to that big thing, but that along that path you would be doing small things that would be telling you all the key things you need to know about how to do the big thing right. Or even whether to quit and don't do the big thing.

Robin: So I am talking about both. I mean, we can't talk about the short things without talking about the big things too, because the whole point of doing the small things is to prepare for, and to lay the groundwork for, the big things. But it's stupid to just pretend like we could know exactly how to do the big things right out from the bat. So in any policy area, including this, this is for all policy whatsoever, you're always talking at these two different levels of trying to imagine the thing you'd eventually do and how that would play out and whether that was a good idea. And then trying to think about what path you would use to get there in terms of what experiments you do, what permission you ask for, what trials you do, and what variations, so that you could learn not only how to do it right, but also create demonstrations and motivation and excitement and energy and attention, such that we might eventually get to the big things.

Elizabeth: So are there things where if we ran a couple of small experiments, that would convince you we shouldn't continue?

Robin: Well sure, like neighbors come with shotguns and shoot everybody because they're really offended by the whole thing. I mean, it means you don't do it that way. You have to go much more isolated, for example. Obviously, if it turns out there's some really strange effect where you modify the virus the way you thought was harmless and irrelevant, but it turns out to make it much more harmful and then a lot more people die and you go, "Well, I guess I didn't really understand this at all." Right?

Robin: So, for example, I know that one way in experiments in the past that people have used to infect people with say the flu, is just play cards. You have somebody with the flu and a bunch of other people without the flu and they play cards together. You know what, people catch the flu. That is a mechanism of deliberate infection. So it's just showing you, there is a lot of very straightforward mechanisms, but you could pick some clever mechanism that you thought was clever and that you were wrong. You mix it with lemon juice, you don't think lemon juice matters, but somehow it does.

Ben: But, okay. Robin, I hear you're saying before you do the big things, you should do small scale tests, but in the world where it is, the plan B and deliberate infection is the primary solution mechanism, I'm curious what your mainline expectation is for the rough percentage of, I don't know, the US population or something you expect should be deliberately infected?

Robin: Well, I mean, since like the natural R0 of this thing is what, like four or something, then you can calculate from that the percentage of the population you need to be infected in order to get herd immunity, which is like 80% or something, 70%-80%. So that's the number of people you need to be infected one way or the other, and once it works you might as well use it in the sense that like with smallpox, there was no, "Infect some people with the smallpox thing variolation and other people in other way." They were trying to get it to enough people so that basically they stopped it.

Zvi: So I will point out that I don't think it's as simple as infecting 80%. I think the different people matter a lot more or less,and et cetera, et cetera. And you can do very different things. But yeah, we are almost out of time, so I won't–

Robin: Or on the order of half the population.

Zvi: Yeah, I think we did get on the order of half, yeah. Minus whoever's already infected.

Robin: Right. And so the longer you wait, the less opportunity you'll have to infect the other people who could have had a much smaller reaction to it because they could've had a much smaller version.

Ben: All right. Elizabeth, was there any follow up you had on that or is–

Elizabeth: I'm good, thanks.

Ben: Cool. Thanks. I'm going to take you out of the room. 

Ending

Ben: All right. I think that's a wrap.

Robin: I'd like to thank my co-debater for being reasonable and intelligent and thoughtful. And I'm pretty sure if I debate anybody else on this topic in the next month, it will be a much more hostile engagement.

Zvi: And I will echo, yes.

Ben: Yeah, no, thanks. This has been mad. I think I asked you both to do this on Monday or something insane–

Zvi: Something like that.

Ben: –so thanks a lot for getting available quickly. Appreciate it. A bunch of people, like a hundred people, have been watching throughout most of this, so that's ... Yeah, this is quite exciting. So next, there is an online meetup that you can join. We are going to try the open-source sort of VR platform, Mozilla Hubs. You're welcome to find it on LessWrong. There's a link to a bunch of rooms you can join. I assign some probability to this being awesome and some probability to this not working out at all. And in about 20 to 25 minutes, we'll have a good sense of that and we will put up a link to a Discord server instead if it seems to not be working.

Ben: But I am going to be heading into the Mozilla Hubs chat rooms and I expect I will chat with a bunch of you guys there. Is there anything else I wanted you to say before this thing wrapped up? The one thing I forgot to say at the beginning, which was another thing that's inspiring wanting debates for a while, was there was an amazing ... One of, I think, Oli’s favorite Wikipedia pages is the Wikipedia page for The Scientific Revolution. And at the end of one of the sentences there, there's a sort of a lead-up sentence and it has no citations and no links and no further explanation and it just says, "And there were many philosophical battles." And that's been going around our head for a while and why I wanted to start having a bunch of debates, and this was just quite fun. So thanks both for joining, and have a good time quarantining your homes for over the next couple of months. And I'll see you again.

Zvi: That is the plan.

Ben: Yeah. All right. Thanks a lot. Farewell.

3 comments

Comments sorted by top scores.

comment by DanielFilan · 2020-04-08T03:59:26.560Z · LW(p) · GW(p)

Thoughts re: my question and the responses [LW · GW]:

  • The leverage argument for working on COVID-19 rather than existential risk (x-risk) seems weak by itself. My guess is that working on COVID-19 has about 7 orders of magnitude more tractability than x-risk (2 for knowing which risk you're working on, 2 for understanding it, and 3 for less prior effort), but that the scale of x-risk is more than 10 orders of magnitude higher than the direct disease burden of COVID-19.
  • I'm unsure if world-ending pandemics look like this one, but it's a good point.
  • The civilisational threat point seems pretty legitimate, but I'm unsure how to weigh it.
  • The harms done by factory farming of non-human animals seem comparable to the direct disease burden of COVID-19 to me.
  • OpenPhil may be funding some forecasting work with the Good Judgement Project, which is what I was referring to, but as far as I'm concerned, Metaculus (which I'm involved in) is doing better forecasting (and might also be funded by OpenPhil?). See this dashboard of predictions.
  • This situation seems like it reveals a lot of information about governance, but if all one wants to do is learn from the situation, it seems better to document it a bit now and wait later. However, if one wants to contribute to a probable effort to improve governance of pandemics at the tail end of this outbreak, that would require careful analysis and action now.
  • If I mostly wanted to raise the status of the effective altruism movement, I wouldn't push a policy proposal as unpopular as variolation - but it might become more popular as people become better at marketing it?

Overall, I now think that it's worth the full time of a small contingent of effective altruists to focus on policy responses to COVID-19 as well as broadly understanding it, and that we are probably assigning too little focus to this (although within the portfolio of attention, I wish more were directed towards neglected high-leverage interventions). Most of the arguments presented I find convincing, except for the one that I said was convincing during the call.

Replies from: ChristianKl
comment by ChristianKl · 2020-04-08T14:27:55.483Z · LW(p) · GW(p)
If I mostly wanted to raise the status of the effective altruism movement, I wouldn't push a policy proposal as unpopular as variolation - but it might become more popular as people become better at marketing it?

Doing variolation without doing animal trials first is going to be unpopular as it's an untested and potentially dangerous procedure.

The popular way to do it would be to call for animal trials of variolation.

Replies from: jimmy
comment by jimmy · 2020-04-14T18:34:55.690Z · LW(p) · GW(p)

It depends on what you mean by "unpopular". If you mean that someone is going to ignore the lives that would be saved and accuse you of being uncaring, then that's certainly true and you would need to be ready to deal with that.

On the other hand, if you mean that everyone would actually be against this idea then I think you're wrong. I've been floating the idea every time I end up in a discussion about this virus, and while my conversations can't be taken as completely representative, it's worth noting that not once have I had anyone say it's a bad idea. The most negative I've gotten was "that's interesting", and most of the other people I've talked to have said that they would do it right now and that so would a lot of their friends.

In a situation where the risks for healthy young people are low and eventual infection is likely anyway, "If people are going to get sick anyway, let them do it on their terms so that it can be as safe as possible" is not a hard argument to win, and the people who need to be convinced are likely far more sympathetic to such ideas than you think.

We just need to create the common knowledge that such ideas are thinkable and doesn't have to be a politically losing stance. A lot of "common knowledge" stances have turned out to be wrong and to flip overnight, and you'd be offering people a chance to be ahead of the curve and the first to jump on the winning team that saved the day. It'd have to be done deliberately and carefully, but if you do it right people will take it.