Pod Insights a16z Podcast Pandemics: Early Detection, Networks, Spreaders About this episode with @nachristakis @jorgecondebio
Going from rapid warning to early detection through social network sensors can make all the difference when it comes to contagion/ the spread of disease and pandemics. Can we get public health bio surveillance without sacrificing privacy and agency?
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:01 Sonal? Hi everyone, welcome to the A6 and Z podcast. I'm Sonal. Today, we're continuing our ongoing podcast series covering different aspects of the coronavirus pandemic. As a reminder, you can catch all past and recent content on this at a6andz.com/coronavirus. Today's topic is on tracking networks of spread for predicting pandemics and loads, which has implications for resource allocation, getting back to work, and much more. Our guest is sociologist and physician Nicholas Christakis, who directs the Human Nature Lab, part of the Yale Institute for Network Science. He and his collaborators previously published work in PLOS on the H1N1 pandemic, which informed their newly released app Hunala,
0:45 Sonal? which is an example of a privacy-preserving approach for tracking pandemics to allow not just rapid warning, but early detection. They also published a recent paper in Nature on how "population flow drives spatio-temporal distribution of COVID-19 in China", which, among other things, may comment on the accuracy of the early data released by Chinese authorities, all of which is covered in this episode, and you can find all the links referenced in the show notes. Also, a quick note that we have another episode on not just tracking and predicting, but the privacy aspects of contact tracing, a separate but very related topic, on 16 Minutes. You can find that show in its own feed if you're not already subscribed there
1:30 Sonal? in your favorite podcast app. In this episode, Christakis is interviewed by A6 and Z general partner Jorge Conde on our bio team. The two also share their thoughts on how bad is it, but they begin with where testing and tracking fits and what's been missing, given that pandemics themselves are a lot more predictable than most people realize.
1:52 Speaker 01 I need to emphasize just how predictable this is. It's not just, you know, for example, Bill Gates had this TED Talk in 2015, or Tony Fauci has been publishing papers about this for 20 or 30 years. I could reach over on my bookshelf right here and get a book called "National Strategy for Influenza Pandemic", and it describes everything we're experiencing. There's like a playbook, like what you're supposed to do. What's not predictable is the intensity or the precise timing of its arrival. So, we were kind of lulled into a false sense of security. In 2003, we had the first SARS epidemic, which ultimately only afflicted like 8,500 people worldwide. And it put the fear of God in Taiwan and Hong Kong and Korea and Japan and China, but not us. Because, you know, we had like 30 cases or something.
2:36 Speaker 01 And then in 2009, we had the H1N1, which did afflict more individuals, but it was so mild that most people just had a cold. So again, we didn't take it seriously. Well, now we have a perfect storm of a serious pathogen that's afflicting a large number of people, and we're completely unprepared. And unfortunately, we weren't ready with the kind of testing we needed to really respond to the epidemic. And just to be clear, testing has multiple purposes, and there are multiple different kinds of testing. One purpose is to identify people who need treatment or identify people who need quarantine. But another purpose is an instrumentation function. Like you wouldn't fly a plane without knowing your altitude and your speed and so on. If you're a policymaker and you're responsible for the health of the public, you kind of need to know where is the germ and how much is there?
3:21 Speaker 01 And we were flying blind for weeks. And we still are, honestly, flying blind.
3:26 Speaker 02 If you sort of break down the problem, it's because there are a bunch of different things that have been missing in the past that I think we find ourselves having flown blind here. Number one is, we just need better detection technology. So here, it'd be in the form of testing. Number two, we actually need better mechanisms and frankly, incentives for people to share data more broadly, whether it's gathered at a governmental level or at institutional level, or ultimately at an individual level, so that we can get sort of high-resolution maps of where disease is present and where it's not. And number three, we need the ability to analyze these things in very real time. The single unifying goal of public health is to do this and to do this well.
4:08 Speaker 02 One of the holy grails, I think, for public health has been to create some comprehensive sort of biosurveillance. Essentially, a weather map of where there is health and disease across the country or across the world. And specifically within the US, there are several efforts to sort of get at this. The CDC has this sort of flu-tracking map. I worked in a previous life on a product against a virus called RSV, respiratory sensitive virus that affects primarily premature infants, elderly, and other immunocompromised people. And one of the big challenges we had every single year, we had a therapeutic that would be given to premature infants that were particularly at risk, was finding out where and when premature
4:51 Speaker 02 infants would be at elevated risk and therefore would be good candidates for prophylaxis treatment. And what we found time and time again was that the resolution of the weather map, if you will, was so low and so latent. It took, we were usually getting reports in weeks and weeks later. So there really wasn't sort of a way to get a good forecast of where to be with the appropriate prophylactic treatment. And if we look over sort of the scope of the public health apparatus around the world, and certainly in the United States, we're trying to come at this from two extremes, I would say. On the one hand, we want to figure out, can we get accurate forecasts? And then the other end of the spectrum is, how do we get to ground truth as to where the virus is?
5:35 Speaker 01 Well, on that issue, we have this app called Hunala that is like ways for respiratory disease or ways for influenza-like illness. So each person signs up, they contribute a little information, and then you can see what is your risk of getting respiratory infections around the bend, like upcoming.
5:55 Speaker 01 Kind of crowdsourced, non-governmental, privacy-respecting way of getting information about your risk of coronavirus. The science behind it is stuff that we published for the H1N1 pandemic back in 2009. And the gist is that if something is spreading in the network, it should reach central people, people with many friends, whose friends in turn have many friends, sooner in the course of the epidemic than it reaches peripheral people, people who have few or no friends, people who are hermits. That means that certain people in the middle of the network are going to be more likely to get anything that's spreading earlier in the course of the epidemic than random people. As you probably know, about 10 years ago, Larry Brilliance's team proposed this notion
6:38 Speaker 01 of Google Flu Trends, which was a kind of a big data idea, let's use what people are searching for today in order to know where is the epidemic today. But if you could identify central people in a network, they would act as canaries in a coal mine, because they're going to get the epidemic early. If you can see the epidemic spiking in those individuals, you can know where the epidemic will be in two weeks, for example.
7:04 Speaker 02 I like the Waze analogy, because Waze has, you know, sort of two modes. One is, it gives you objective truth of where you are on the map and which direction you're heading, and then it gives you like subjective truth in like who saw a cop and who ran over a pothole and all of those things. And you want to ideally marry those two things.
7:21 Speaker 01 I mean, yeah, there are two perspectives on trying to understand what's happening when it comes to the pandemic. One is the individual perspective, like I want to know what's happening around me, what's happening in my community, what's happening among my friends or the people I'm interacting with. And then, of course, there's the collective perspective where public health authorities or hospital officials or politicians want to know what's happening. Like, can we understand where is the germ? We have a dashboard function, which would be as if the police or the state authorities interested in highway maintenance could monitor all these reports and see dozens of people are saying there's a pothole here, or this is where accidents occur. The same thing happens where hospital directors or public health authorities, if there's enough
8:04 Speaker 01 adoption, that's the crucial thing of our app, it could also serve that function, like what's happening right now in this district versus that district? What's happening in my catchment area in terms of, you know, symptom reports, for instance? So it's like if you can log on to Waze and see today is a traffic-y day in New York City, but what about on my street, the part of the city that I am in? What can you tell me about that? After you complete your symptom check-in for the day, you get immediate feedback on what is your risk of infection, given where you live, and what is your risk of infection, given where you are in the social network? For example, your friends' friends' friends, two weeks ago, had the flu, and so now it's
8:49 Speaker 01 winding its way towards you. When they first sign up, you also tell us a little bit about who your friends are. And we use that data, it's all with respect to privacy, we don't track that information, but we use that information to ping those people, and then the app uses machine learning to process large amounts of information that you've told us about yourself, and that the whole network is generating, to give you a personalized assessment of your risk of getting a respiratory disease, based on where you are.
9:18 Speaker 02 What is the assumption around network coverage, in terms of adoption, that you would need for this to be informative?
9:23 Speaker 01 Yeah, so it is exactly like Waze, the more people that use it, the better it is. The risk we run, of course, is that adoption will be low in any given location. If we have about 100 people in a county, we think that it will begin to perform much better than just naive feedback to the users. It's like trying to estimate how traffic is New York City, 1, 2, 3, 4, 5 people telling you they're looking out their window and seeing the traffic isn't good. By the time you have 100 people in the city, saying, "Oh my God, traffic's at a standstill outside my window," then you can begin to make a generalized estimate, and the more people you get, of course, the more specific you can get.
10:03 Speaker 02 And so to abuse the Waze analogy for a second, how do you think about the potential that this particular virus has a very, very high rate of asymptomatic people? That either aren't asymptomatic for a long time or remain asymptomatic over the course of the disease and therefore the equivalent of there are invisible cars on the road causing traffic.
10:24 Speaker 01 The analogy would be the cars that are not speeding or cars that have not been pulled over or don't have any accidents. So you're right, we think about 50% of people are asymptomatic from this condition from coronavirus, the best estimates now. Such individuals would still benefit, of course. We don't forecast if you're infected, how seriously ill might you get. We're just letting you know what's happening around you and your symptoms reports wouldn't contribute information in that regard, but there are other ways that you contribute. This is overusing the Waze analogy, but for example, the fact that you were able to transit from point A to point B down a segment of highway with nothing happening to you still tells us something. One trivial example of that, let's say Sonal
11:06 Speaker 01 is your friend and you are my friend and Sonal gets respiratory disease and she has symptoms, she gives it to you and you do not have symptoms, but you give it to me. Through you, even though you have no symptoms, the app will be able to take advantage of the information that Sonal has symptoms. So by connecting us through a geodesic of two, you know, she is my friend's friend, even though you don't have any symptoms, your presence in the system is contributing information or can help me be situationally aware of what's happening around me.
11:39 Speaker 02 Because if the network gets dense enough, even if you have a high rate of asymptomatics, you could actually impute infected, non-symptomatic individuals.
11:48 Speaker 01 Yes, that's exactly right. But stepping back, the distinction between advanced warning and rapid detection is important. There is a difference between early warning and rapid detection. Most technology nowadays that tries to monitor the course of an epidemic is like the CDC sits in Atlanta, it has a distributed arrangement with testing facilities around the country, patients go and get tested for influenza-like illness, ILI. Those test reports are sent back to Atlanta and then two or three or four weeks from now, the CDC knows where the epidemic is today. With the Waze analogy is you're driving down the highway and there's no traffic jam where you are right now. There's a traffic jam coming up in the future. You can avoid that traffic jam.
12:31 Speaker 01 Then that goes back to the three different timeframes for like the CDC know in three weeks what was happening today. You can use big data techniques about what's happening today to forecast to know what's happening now. For example, the Google Flu Trends idea, which, you know, has come under criticism. And also this paper we have that looks at mobility data. You can look at the flows of people out of Wuhan and you see that, oh, if many people from Wuhan go to these prefecture versus few people from Wuhan go to that prefecture, this first one is going to have many more cases versus lastly, being able to exploit some understanding about spreading processes on graphs in order to be able to forecast
13:13 Speaker 01 using this canaries in a coal mine idea, what's going to happen in the future. And the deployment of resources is not a trivial thing. If you're a public health official and you're trying to see where do I send my PPE, like the ventilators example, could have known two or three weeks early that nope, California's going to be okay, but New York's going to be hit. We can reposition our equipment. Or if you're a manufacturer and you need to forecast demand, then knowing who's going to want these products in Texas versus Illinois is helpful. What we're doing is a kind of decentralized collective effort. Some people might confuse it with a facility that is being offered by these big tech companies. It's like Waze versus having police satellites above the city, you know, tracking where all
13:54 Speaker 01 the cars versus each person contributes some information voluntarily and anonymously. There are a bunch of apps out there now already, which are sort of citizen science apps, like tell us what's happening with you so we can monitor the public. And that's great and people can do that if they want, but with us, they get something of advantage to themselves, namely an assessment of their own risk. It's not just a pro bono publico. It's like the difference between dial the number and tell us where the pothole is so that the municipality can go out and fill the pothole versus inform other people on ways that there's a pothole so we all benefit from knowing where the pothole is. I'm very tech friendly, but I've been very worried at the use of these technologies in the hands of authoritarian governments. And in some ways, I am more comfortable having Google and Apple know about me as a private
14:39 Speaker 01 company than I would be about the United States government knowing. So I think we have to keep our eyes on the ball is how can we use these technologies? Yes, in a way that's helpful to public health, yes, but without not only not violating our civil liberties today, but without empowering the abandonment of civil liberties in the future.
14:58 Speaker 02 On the work that you've done. With the Wuhan data, given that last point you made, is something akin to that even possible in the United States?
15:06 Speaker 01 Yeah, so the Wuhan data project, that was a project where we did it in partnership with some Chinese colleagues and colleagues in Hong Kong, and they had a partnership with one of the primary telcos in China, and we got data on population movements out of Wuhan. So between January 1st and January 24th, we had data on 12 million transits through Wuhan, spreading out to 296 other prefectures throughout the rest of China. So we know that, you know, 100,000 people went from Wuhan to this prefecture, and 10,000 from Wuhan to that prefecture, and none from Wuhan to this other prefecture. And then we had follow-up data from the Chinese CDC on COVID cases through February the 19th,
15:47 Speaker 01 and basically, we were able to perfectly predict the location, timing, and intensity of the coronavirus epidemic in each of the remaining 296 prefectures of China, based on population outflow. Incidentally, that allowed us also to validate the accuracy of the Chinese data, subject to a couple of provisos. For example, we cannot validate the size of the number of cases in Wuhan, which is probably much higher than the reported number, partly because at the beginning, they couldn't even know who was sick, and maybe they fudged the data a little bit too, in Wuhan. But certainly, for the remaining prefectures, it would have been very difficult to create the pattern in the data that we saw deliberately.
16:29 Speaker 01 So, yes, only some nations would be able to commandeer such data from the private sector and then use it. But it doesn't need to be phone data, it could be car data, for example, or train ticketing data, or car tolling data, or all kinds of other data that tracks movement. And once again, it could be voluntary. People could report they're taking a trip from point A to point B. When enough people start reporting they're taking a trip from point A to point B, you can actually use that collectively generated information outside the government to make predictions about where the intensity of the epidemic is likely to be. So, yes, you're right that you do need some source of data, if you're going to take advantage of these big data techniques.
17:14 Speaker 01 It doesn't have to be phone data, it doesn't have to be in the hands of the government, but of course, you do need data to actually make it work.
17:21 Speaker 02 In the context of the work you've done, whether it's with Wuhan, with the stuff you're doing with the Hunala app, where does what I would call ground truth fit into this? In other words, this idea that it is a knowable fact as to whether or not someone has a virus. That's an objective truth. It's, you know, at some level.
17:40 Speaker 01 It's an objective truth. But even that, if you use the Heisenberg uncertainty principle, you know, not all tests are perfect.
17:46 Speaker 02 Fair.
17:47 Speaker 01 There is an objective reality out there. That's 100% true. I agree with that.
17:50 Speaker 02 Exactly. So there's an objective reality out there as to where their virus is present or not present in an individual. And the best approximation we have to that objective reality is a test, right? And the tests have all their imperfections, including not only how the test performs, but whether or not they are in fact available. Where does testing fit on the one side? And on the other side, where does testing fit at the aggregate population level with the kinds of things you did with Wuhan?
18:16 Speaker 01 Okay. Well, the first question, we do allow people to tell us if they've had a COVID test and if the test was positive or negative. But stepping back, even the CDC's tracking of so-called ILI, influenza-like illness, or other outbreaks, for example, when the WHO tracks outbreaks around the world, there's a rush to define what counts as a case, even in the absence of a test. So I can say, if you have a fever of above or 1.4 and you have a cough and you were in China, then I'm going to say that you had COVID, even if we don't have a test on you, right? Or for example, if we're trying to get people to self-report whether they had the flu, if I asked you, you had something last week, was it a cold or was it the flu?
19:01 Speaker 01 There are non-test-based, non-biomedical test-based questions you can ask people to tell the difference between conditions. You know, in medicine, we call this taking a history. And in fact, there's a joke when I was in medical school that we were taught, you know, that 80% of the diagnostic information comes from just talking to the patient. And 15% comes from examining the patient and only 5% comes from testing the patient. By the time you get to the test stage, typically you're just confirming what you think the patient has. Even in your example of the respiratory syncytial virus, a case that we were talking about earlier, even in that case, many doctors will know that this patient likely has RSV and will do the test to confirm their suspicion. And sometimes a test will refute what the doctor
19:44 Speaker 01 thought. But nevertheless, a lot of the variance is going to be explained just by the history and the physical. So it's possible to, certainly at the aggregate level, but even at the individual level, to use patient reports to get some sense of what's happening. Now, we won't necessarily be able to tell the difference between you had influenza virus caused your respiratory problems last week versus a mild case of coronavirus caused it last week. But in some sense, we don't really need to know that. From a public health point of view, I would argue that is not as crucial.
20:16 Speaker 02 It is an interesting point, that last point you make, which is, at some level, the question you're trying to answer isn't COVID, yes, no, is, am I suffering from a respiratory disease, yes, no?
20:27 Speaker 01 It's a little bit like the shift that's taking place right now around the country on just quantifying excess deaths from any cause.
20:34 Speaker 02 Yes.
20:34 Speaker 01 Because a lot of people that die of coronavirus and we never diagnosed them, or they had a heart attack, or we thought they died of the flu, but actually it was coronavirus. And you can look at these excess deaths and look at the impact of the epidemic in our population in some sense. So I would argue that that's another metric. And incidentally, just speaking for myself, the only numbers that I've been following the last two or three months are deaths. I've not been following number of tests. I've not been following percent positive. I've not been following the case count. And even deaths can be either an undercount or an overcount for various reasons. But I regard that as a harder endpoint, even if it's a lagging indicator of where we are.
21:11 Speaker 02 Given that you're looking at deaths, and obviously you're spending a lot of time thinking about this problem and looking at data, what's your personal take on how bad is this bug? You know, what is the CFR? What is the IFR? Have you thought through that?
21:24 Speaker 01 Yes.
21:24 Speaker 02 And do you have an opinion?
21:25 Speaker 01 I do have an opinion. I think that case fatality rate is going to be between 0.3 and 1%. So it'll be at least three times as deadly as the flu on average and maybe 10 times as deadly. There's a lot we can discuss about that, like, of course, the older the population, the higher the case fatality rate. There's some intriguing evidence about different viral strains being more or less deadly. A possibility I had excluded at the, even two months ago, I didn't think that was very likely, but there's some worrisome evidence that there may be different strains, some of which are much worse. And this is one theory that may partially explain why New York was harder hit than California. I do not think by any means that's the only reason, but it may partially contribute because the New York
22:10 Speaker 01 was seeded by the Italian strain and California and Seattle by the Chinese strain. But the CFR is the probability of death, given that you came to medical attention. And that's, some people don't seek medical care, some people can't get medical care. So that definition's a little fuzzy. Then there's something called the SCFR, the symptomatic case fatality rate. That's the probability you die, given that you ever have symptoms, which is a bit better in the sense that you might have symptoms and never seek medical care, but we're going to count you then, conditional on symptoms, do you die? And then even better from my perspective, if it can be known, if we have testing, for example, is the IFR, which is the infection fatality rate, the fraction of people who are infected that ultimately die. Now here, the numbers range from 40 to 60%, but it's about half in my view. The
22:55 Speaker 01 people who get this disease for very weird reasons never get symptoms. That means the IFR is about half the CFR. So I think the IFR is going to be between 0.15 and 0.5, let's say, which is a bad disease, by the way, just to be very clear. These numbers seem small, but I think it's a serious pathogen, regardless of what fraction of people it kills. What do you think from your desk?
23:18 Speaker 02 Yeah, I would say I think this is likely to become endemic. So that means, you know, that this becomes effectively seasonal. And, you know, I think a big important question is, how do we get from here to there? And there being, you know, sort of the next rebound when it comes. And any effort to open up the economy will be met with resurgence of disease and emergence of new hotspots. I don't think there will be a silver bullet, but from a surveillance standpoint, it looks like we have multiple weapons that we can deploy. We're taking together, we will hopefully get some level of ability to fly with instrumentation, to use your analogy. Without that, we will continue to fly blind. One of the important aspects is whether or not we can effectively run a process for
24:03 Speaker 02 testing, tracing, tracking, and ultimately isolating individuals that are affected. We are way under where we need to be. In fact, there's an op-ed that was recently written by a group of scientists, Sri Kusuri at Okten Bio, Jason Kelly at Ginkgo BioWorks, Fang Zhang at the Broad Institute, and Jay Shender at the University of Washington. These are leading scientists and genomicists and biologists who are essentially making the arguments, it's almost a call to arms, to say, look, we have a lot of the technology in place, we just have to put the pieces together so that we can do this at the scale at which it's necessary. It's almost like a call for mobilization. So on the testing side, there are several efforts to do that.
24:47 Speaker 02 No test is perfect, but that gives you the ability to say, okay, this infected individual could either seek treatment or should be isolated, but as we know, before that individual knew that they were affected, they could have infected other individuals. And this is where the tracing and tracking aspect is very important. It's this idea of, if I know someone is now infected, can I essentially go back in time and figure out who else they may have infected? The tracking and tracing side, people essentially contribute their information in terms of being symptomatic. There is a Kinsa that, for example, that has an internet-connected thermometer where
25:29 Speaker 02 they can create a heat map, so you can sort of track at some level the presence or absence of disease, or at least a proxy for the presence or absence of disease, by looking at above or below normal levels of temperatures.
25:42 Speaker 02 So there are several efforts to do this. The folks at Google and Apple have developed technology where potentially you could use Bluetooth to warn you if you've been around someone that's been determined to be infected. But these are all very backwards looking. So it's almost like watching the weather forecast a week after you've experienced that weather. And so really what we need is largely through this testing, tracking and tracing approach. And then you use the information in terms of who they have been in contact with to figure out who are the most likely nodes in their immediate network so that we can get those individuals tested. And once you do that in aggregate, that's what gives you sort of the living weather
26:23 Speaker 02 map that's very actionable from a policy standpoint, whether it's, you know, at the governmental level, but also figuring out, are my employees safe? Can I open my business? Can I go back to work? Can I expose my children to go back to school?
26:39 Speaker 01 Yeah, I mean, I agree 100%. I think I'd add two things to what you're saying. There's no way that we can resume a normal economy unless we unfortunately are willing to tolerate some risk. And for example, this applies most directly to the reopening of schools and businesses. So I think we're going to need some kind of universal liability waiver, some kind of change in laws, ideally coupled with some kind of no-fault insurance so that, you know, we can compensate people who fall ill or die. Second, my lab does a lot of work on social contagion and we've been experimentally working, we've been doing field experiments with support from the Gates Foundation, with support from the NOMIS Foundation in Switzerland, the Robert Wood Johnson Foundation, with the Tata conglomerate.
27:21 Speaker 01 We've been doing experiments in India and in Honduras where we try to create artificial tipping points in social behavior. So how do we change vaccine behavior? How do we change breastfeeding behavior? How do we change latrine behavior? Public health behaviors in these settings? Well, the same thing is relevant, of course, in our country. How are we going to change mask-wearing behavior? You know, how do we get people to keep six feet distance, for example? How do you get people to wash their hands? There are all these actions that individuals can take that reduce collective risk. And so one of the things we've been doing is applying some of our ideas and techniques and mathematical tricks that we have developed for changing population behavior at scale in these developing world settings to how can we apply them in the United States?
28:05 Speaker 02 You talk about social contagion. I'd be curious to get your take on two fronts. Number one, obviously, when it comes to something like a coronavirus, the contagion here is physical, right? You have, you know, the it's the proximity to the individual where the virus can jump from person to person. I would imagine that sort of a central node in the network infects more and more nodes because they have higher connectivity. How does the concept account for the possibility, if such a thing in fact exists, of a super spreader?
28:39 Speaker 01 So we know there are super spreading events with all epidemics. Typhoid Mary, for instance, was a super spreader. The question is, what explains it? Couple of things when thinking about super spreaders important to distinguish. First is in any distribution, there are going to be outliers. So, you know, we have a thousand people. So let's say the effective reproductive rate of the pathogen is, let's say, two. And it either could be that every single person has the same reproductive rate. Everyone in expectation infects two people. Or we could have some range. Some people infect nobody. They zero. Some people infect one person, some two. And one or two guys infect 100 or 300 people. They could be super spreaders, but maybe they're just super spreaders by chance. There's nothing about them or that they did or their environment that made them a super spreader.
29:23 Speaker 01 There's just a natural distribution and someone has to be at the top and someone has to be at the bottom. But sometimes there is a more interesting biological or social explanation for super spreading. So for example, in many of the cases of coronavirus, we've seen super spreading events that have typically occurred in very compact situations where a sick person was in a contained environment with many other people. If that same person had gone to some other situation, they might not have caused a big super spreading event. But it was something about the environment, maybe the air conditioning or the density of people in the room or certain other kinds of things that might prompt that individual in that setting to be a super spreader. Or it could be something about the person or their behavior. For example, you might have more sensitive
30:08 Speaker 01 Nerves in your throat and your nose so that when you get infected with a pathogen, you cough and sneeze a lot. I get the same pathogen, I'm otherwise identical to you, but I don't cough and sneeze a lot. So you spread to many people and I don't. And it's just something about you, your body, that is causing this event. And sometimes it's even the pathogen, like the strain of the virus you get might actually be more transmissible than the strain I got, for example. So it's the pathogen, the host, the environment, or chance are the four things that might possibly explain super-spreading events.
30:39 Speaker 02 Can you detect super-spreaders in a network? And I think the answer is probably yes. Can you model that and forecast that in any sort of predictable way?
30:48 Speaker 01 I'm reasonably sure that you can forecast the role of super-spreaders and the existence of super-spreaders in an observed epidemic. What I don't know is if you can forecast who will be a super-spreeder. I suppose there are, as we've been discussing, certain things about the individual that you could maybe tell, or certain things about the environment that you could tell. You can know basic things, like, for example, people have many connections. Like, I don't think it's a coincidence that, early on, all these very prominent people, actors and politicians, got the pathogen. People said, "Oh, it's because they're in the public eye," or "They're rich and they had access to tests and that's how we knew it." Yes, those things are true, but also, those people are shaking a lot of hands.
31:33 Speaker 01 They have high degree, they're connected to many people, and they also are acting as a forecast of who's gonna get the disease.
31:40 Speaker 02 So, Nicholas, if we're relying on users to essentially self-report how they're feeling, essentially self-report who their contacts are, there is a level of subjectivity there. Isn't there a risk that, at some level, this approach would be vulnerable to a bad actor that could somehow meaningfully skew the data?
32:02 Speaker 01 Yeah, so there's always that risk of someone, there's always the risk of bad actors. You know, there was that very funny artist, I think he was in Germany or somewhere in Europe, where he got like 100 cellphones and put them on a little wagon.
32:16 Speaker 02 Yeah, to create traffic jams.
32:17 Speaker 01 Yes. And drag them through the city and create a traffic jams on Google Maps. And so, you know, I suppose you could have people like that. And of course, there were always people like Typhoid Mary, you know, was actually a cook and refused quarantine and kept infecting people until she was finally the state basically incarcerated her. So yes, there could be people who act in a way, but they are, as I discuss, actually in my book Blueprint. I'm quite an optimist about our species. We are a species that's capable of love and friendship and cooperation and teaching. And of course, we have evil and violence and selfishness and hatred. And, and we have had those since time immemorial. But I believe the good qualities outweigh the bad qualities.
33:01 Speaker 01 And in fact, I show that provide a lot of compelling evidence in Blueprint. And so, yes, you're right that there could be bad actors, but I think they will be swamped by the good actors. But I do want to go back to this business about subjective versus objective because I see a problem where both the far left and the far right, for their own reasons, wish to deny the existence of an objective reality. And it's enraging to a scientist like me. So, you know, you see people who will say, oh, well, it's just going to disappear, you know, or it's nothing. Or why do we need to listen? But denial of the existence of an objective reality doesn't make that objective reality go away. It's still there.
33:39 Speaker 02 So Nicholas, thank you for joining the A16Z Podcast. And I, like you, share incredible hope that while we will be living with this virus for a long time, we will find a way to live with this virus going forward.
33:52 Speaker 01 No, you're welcome. I mean, I've had a really wide ranging and interesting conversation. Thank you for having me.
Previous Podcasting and the Future of Audio Next Don't Call it a Brain in a Dish! Page updated September 30, 2026 · transcript and summary are machine-generated