Nassim Nicholas Taleb and Yaneer Bar-Yam Discuss COVID Risk

we're done we're going hi hey how's it going great so we're going to have a regular conversa resume our regular conversation we i think we stopped in march right exactly okay you're a busy man i'm uh you know i got free time okay so do this more often it's fun okay so the the i have two points to discuss with you and then and then we'll take your your comments the first one is that we should insist on scale transformation when it comes to uh diseases okay that a doctor has you know is very good at understanding the risks for one patient but when you're talking about the aggregate the risk for a collective it becomes very different for example if a patient has a certain disease okay that's uh you compute the probability or you may miscompute the probability but still it's it's a problem that's very different from talking about collective because then they cannot compare they make mistakes of comparing that's from kovid from death on the road when that's on the road are not contagious or slightly contagious if you want but not seriously whereas death from covid are highly contagious so if my neighbor dies on the road odds are there's you know my probability of dying on the road has not changed but if my neighbor dies of covid my the probability of my having covert increases so when we deal with and the way we do it is using extreme value theory if if in other words if you disappear from earth for one year to go work on your next textbook no contact okay and come back from whatever planet you're on and and upon arrival they tell you whether a billion people died you know that that billion people did not die of car accidents or all these special risks that these people people died of only problem the most likely causes are wars and pandemics not pretty much okay so uh that's that's how so that's my my first point should i respond to that one and then we'll go to the second one uh let me make them both and then and then okay my second point is evidence-based what people call evidence-based is usually not uh not not scientific it's not evidence evidence is not necessarily evidence when it's an anecdote it's more like you should rename it anecdote based science not evidence-based science because evidence requires some statistical uh you know significance that is often not met by these and and science is not about also predicting science is about understanding the properties of something which can or cannot be met with naive predictions because sometimes these predictions can be just anecdotes so i'm done with my two points let me add a couple of things and then we can go back um so with about the first one the other part of it is dynamics so scale in a pandemic happens over time and so it's not just what happens once and we know this right because the fact that there is that what you look at one person is sick and the fact that it has larger consequences has to do with contagion which happens over time so you have to not only think about the patient himself and you have to think about the community but you have to think about over time because it spreads over time and and thinking over time about what's happening now let's let's compare the number of cases we have of this now versus the number of cases we have of this now uh it doesn't reflect the fact that indeed over time one case can turn into a million cases versus the other case is not going to so it's it's it's both the understanding of individual versus collective but also understanding time and yeah this is okay it's quite interesting what you're saying because in the beginning of skin in the game i said i'm taking the world as it is and the analysis as they are and adding time and dynamics and and and effectively they're the same thing yeah well time yeah is how we identify sorry uh scale and dynamics and effectively that's right scale and dynamics both related to each other of course but i have an example and a kind of statement about it is that we see doctors that are treating a patient or they know about a patient and they send them home to infect their families how can we do that and and the part of this is the psychology part of this is the psychology of the separation of responsibility a doctor has a patient walking into their office they're responsible for what happens in the office and to that patient a patient comes into a hospital they're responsible for what happens in the hospital but send them home it's a community there's no connection between what's going on in the hospital or in the doctor's office and what happens in the community or in the home of the person which is a very strong way to say it but otherwise how can you understand that you would send someone home and the result of that is going to be to infect their families yeah okay this is classical problem of uh expert uh think because people um are are totally incapable of getting the expert problem that that expertise is something that is should be treated a lot more delicately and and it comes back to macro and micro uh micro expertise typically isn't expertise like a plumber has a micro expertise but but plumbers cannot make claims on the general uh plumbing uh anything and the second point we'll make there is you and i so this is what i'm saying now is present in the last paper we wrote together but in the first paper we wrote together on uh precautionary approaches to two things we discussed the carpenter fallacy and the carpenter fallacy can you describe it i mean well the carpenter fallacy is that a carpenter will understand something made what what the uses of something made out of wood yes right if you if you make if you know how to make something out of wood does it mean that you understand you know say in the example like is a roulette wheel right how do you understand the probabilities of roulette wheel if you know how to make something out of wood i.e the roulette wheel it's not the same thing that's obvious exactly and it was meant to say that probabilistic knowledge is rarely derived from uh micro expertise that's right okay so in sequences and ruin problems they're entirely and then the interesting thing is the more you go collective the more the problem becomes probabilistic now now you may not agree fully because your approach is different my approach is entirely probabilistic right i know but but in the case of a pandemic it's actually also dynamic so it shows us that dynamics plays a role in the context of large-scale things as well as fine scale things it's just when does it show up exactly yeah so i'm saying that you can approach it with dynamics or i can approach it like my methods i stay in my stick to my probability approach by looking at a slice of time or slices of scale by saying how things can be different because you're not meeting uh some independence criteria right a bunch of independence criteria that require that are required for a lot of large numbers to operate yeah and i turn it a little bit around the other way and what i do is i do the probabilities of time histories so you have dynamics and then you have a probability of a certain dynamics which is a little different way of thinking about it okay so let's talk about the second one and remind me the second one was about what did you say anecdote-based science anecdote science so so there are a couple of things that i have to say about it because one of the key things that we've been talking about is how do you understand things when you don't have a lot of data because like you have a new disease you don't have a lot of data how do you think about it and so that creates a challenge for statistics-based quote evidence-based science when you have to make decisions and so this is about decisions under uncertainty which is something we both care about very much and what what do you do when you don't have information how do you think about what's going on but recently i've really realized that there is a a missing part that you understand very well but somehow you know it's really strange for me that others are not really clear about this which is it's superim it's super important what you assume is your reference for what you're doing it's called we call it the null hypothesis in technical terminology but the point is it it you kind of have a model for what's happening in the world and this model for what's happening in the world is your starting point for thinking about everything that happens and it's very interesting that there are two different models that have been used for thinking about the coronavirus as an example one is in the far east they had sars a little bit in canada also and they use the sars as a model for thinking about the disease whereas in the west the kind of common disease that we know about that's a terrible disease normally is is the flu so everyone is trying to think about it in terms of the flu if you say it's going to be the flu unless you prove to me that it's not the flu it's very different than you say it's sars and you have to prove to me that it's not light source as it turns out i think the sars idea is a big advantage because in fact the two diseases are very closely related in fact the coronavirus is sars kovi coronavirus two because in fact their mechanism how they cause disease is very similar and how they transmit is very similar and all kinds of things like that so one of the problems is this idea of the is the null hypothesis so but one of the things that i've seen over and over again is when i try to understand why people who are quote evidence-based are really wrong about how they're thinking about the problem altogether is they assume a particular reference point and they then say okay now you've got to prove to me there are other problems evidence based uh aside from that one and and it's pretty much a functioning of large numbers like uh for example if uh if you have a large deviation then n of one is sufficient if you have a regular a lot of regular event an n of a million sometimes is not under fat tails so that's what's important absolutely yeah so for example and this translates into prediction because in physics for your variance is very low and distributions are very entailed okay a single prediction right would be sufficient because if it doesn't come true it means you don't understand the process whereas when you talk about fat-tailed mechanisms you need a million predictions you see to figure out to capture what you know so this is why all this naive predictive analytics and stuff like that it breaks down under fat tails so all of these come back to the workings of a lot of large numbers that's my last book i would have showed it if someone stole it from my library i want to say something about that because um there is this concept that i really feel is kind of missing and that's the concept of the space of possibilities so fat tales are that you have a space of possibilities that extends very far away from any reference that you start from exactly the common events the events that are common you have to be aware that there are events that are very different and and more generally you can have systems where you have events over here and events over there and you have two different kinds of events and when we think about diseases it's even more sophisticated which is that we have the flu we have the coronavirus or the sars we have the um uh hepatitis we have hiv we have all kinds of viruses and and when you think only in terms of one model you don't realize that they're all of these different things that can happen so the right thing about diseases is actually not even just to think about you know sars and flu but actually to think about hey you know there are 20 30 40 diseases that we know about and look at all the different things that they can cause and and in particular not just the ones that we have left over after we vaccinated away everything that is really harmful but think about the ones that are even more harmful that we've gotten rid of because a new disease and this is what we're really finding a new disease is going to be much more like the diseases that we've gotten rid of because we were so affected by them that they caused so much harm that we spent decades making sure that they wouldn't be around um all right so now it's your your turn go ahead i think i think i so one of the things that we might talk about a little bit is what we've learned and just to be a little bit clear about it what we've learned is what we were warning about up front in our original paper we said we don't know exactly what's going to happen with this disease but it's going to be worse or it can be much worse let me make a statement i discovered about precaution before i forget that's right a precautionary approach is an approach that has the maximum amount okay uh the minimum amount of exposed uh uh you know regret right so in other words you take a decision and you you you take the one today that later on you regret the least yeah there are problems technical problem does not get into ergodicity where this is required okay and and the difference between our approach and others or at least between you know my approach so the we started thinking uncertainty was huge in the beginning right and reducing it right whereas others doing the stupid hypothesis testing uh what i call uh fortune cookie evidence-based methods they went the other direction right they started assuming that they knew what is going on they assumed and then then then you know it's difficult to lie or this or discovering things so so we acted the way this is why our way of our approach is vastly more robust we got we got nothing to regret right and and here is the thing and you've mentioned there good ergodicity or non-ergodicity when things are not reversible we have non-reversible effects of this disease and you know death is clearly a non-reversible effect but what we've learned is that despite the fact that early on people said hey 80 of these cases are mild and these people are gonna are recovering and so on it's not at all the case what we find is that somewhere around fifty percent twenty percent to eighty percent in there is a range of all cases have very very severe effects including heart damage where the the tissue of the heart that the muscle fibers are being cut by the virus actually and the there the cells are being nucleated the nucleus is is leaving the cell so you cannot we cannot really figure out what's going on until you make uh people who recover from mild cases of covalent marathons that's right but it's not even that they can't even get out of bed and go to their desks they can't do work yeah so we have to count them with it so the whole idea is when we started our paper we said we had the the so the best decision is a decision that's never really wrong okay in that sense uh given uh the structure of uncertainty is by saying okay we're gonna consider that they are morbidities we know nothing about that's right yes we know otherwise and as time progressed effectively they started showing up and scaring others in the meantime more optimistic today than i was on day one in january when most people are more pessimistic today than toronto right yeah right it could have been even worse but what we know about is surely bad enough and and i don't know i'm more of other people are expecting your points on uh so you want to address vaccines and uh here's this one one more i want to say two more things that are going on that we found out just really in the last few weeks so one is the heart damage which as far as the people who are experts is permanent and in fact if you think about it you tear parts of the heart it can tear itself apart as it's beating and what we see is that people several months and we don't only have several months after they had covid then the heart actually becomes a problem so there's evidence that it may be tearing itself apart which is horrendous we don't know how bad it's going to get still for what for about 20 of the population it's uh right now in the heart the best information is about 50 action some damage but maybe not perceptible or not in the meantime it's damaged you know is there and in fact what happened is that someone took people who died for pulmonary reasons not from heart reasons and they looked in their hearts and they saw that this damage was there so that's pretty bad so in other words we may have some reduction in in some increase of mortality from uh morbidities that we're seeing today in disability one of the things to know is that sars which remember should be the reference they've done a 10-year follow-on study on the survivors and many of them can't work 10 years later so so okay so there is a non uh uh uh i mean i mean substantial let's say or worry some rate of morbidities that's right the other thing is the brain effects the hippocampus and um the hippocampus which is the part that enables you to kind of walk around and navigate where you're going that's uh damaged and the other one is what's called the insulate that they know there are several other parts insulin is responsible for emotions and and and so it's a lot of psycho uh you know um psychological problems so depression and and panic and so on and in fact that's what's being reported by people who have what are called what's called coca-cola so it's not a trivial disease no there's a thing called brain fog where people can't think straight and and and and the other thing by the way there are tens of thousands of people in facebook groups um uh talking about this okay so let's say until we get an idea of the rate of mobility okay we have to be cautious okay it seems that there is also an effect on male fertility we don't know again how much that's going to be i i don't understand how did barabasi detect it from early in in march just simply from doing network analysis it's very straightforward actually in in the context of knowing what the risks are and that is because you look at where in the body the receptor the ace2 receptor is expressed and it's expressed in the lungs we know that in the intestines but also in the heart and in the in the testes and in the brain and there are several other places where it's expressed but these are the places where you would look for harm because that's where the receptor is all right now that one point you had about vaccines i guess we've got a few minutes three four minutes so the vaccines point is they're now talking about using a vaccine before it's been extensively tested and there's a good reason to think about doing that because you know after all this is a terrible disease and maybe we can stop it by using the vaccine and if you didn't have any other way to stop it then you know maybe that would be a good risk now we have other ways to stop it that except for a vaccine that might be better and the reason why is that you test the vaccine progressively in larger numbers of people because just like feeding people you know a new kind of plant that you don't know what's going to happen vaccines can have bad side effects including possibly the effect of actually making the disease worse rather than better or having a reaction some kind of immune reaction it may happen immediately or it may happen a little bit later we don't know so the idea is to test it in larger and larger groups now they're suggesting to start a vaccine that hasn't been tested extensively and i can understand that it would be better to get rid of the disease but so if we want to do that why don't we use it on the smallest population that is going to have the biggest effect on the transmission of the disease and so many people in fact probably most people at this point are taking a lot of precautions or not you know to doing risky things so why don't we take the people who are particularly doing risky things like not wearing masks and we can vaccinate them and then we'll really stop the disease because it's really you can't get them to wear masks i guess it's going to be tough to convince them so vaccination might be a good idea and then they're taking the risk of the vaccine and that would be a you know reasonable thing to do instead of having everyone be vaccinated because the other people are not taking the risks in principle they don't really need it and shouldn't be taking that risk does that make sense um again it's if you if what's feasible all right you can't even wear masks uh you know in the south and i'm sure the vaccines for them are like uh they'd rather take have poison cyanides and vaccines but again i mean if you can but so let me let me ask you the people that you're vaccinating the people who are already taking precautions that doesn't help no it doesn't help and and assuming the precautions are sufficient but also vaccination yeah i mean you can also uh given the risks of this you can also vaccinate people who want to travel or want to do things that that that expose them to the virus and and they can you know calculate the risk and tell themselves okay what it looks like the risk of the vaccine also it gives the history of vaccines because i assume that you can use um the his the history of vaccines that went wrong right and you we know the history of vaccines that went wrong we know how they went wrong and but one thing at the collective the individual like vaccines may be bad at the collective level something tells me that vaccines have much much lower risk than the disease again you can't do you know to compare it's a question of what is the alternative as well so for one person they maybe break even but for the collective it would be huge payoff because a vaccine effectively demultiplies whereas the other one multiplies yeah and then the question is local problem that is not contagious unless the vaccine causes you to do contagious things or go on a rampage or become an economist and do harmful things to society or something like that you see i think that you're right and i think that one of the things that's missing and i really agree that this is missing is a real assessment of risk versus benefit at the societal level foreign i think i see it's a it's the it's a no-brainer vaccines are better and the individual point in time right you wait for the test to be completed that's the question right if you waited two months or three months or five months or whatever it is until the next level test is completed or you do it before and that requires the kind of analysis that is as we said is not part of the standard training yeah exactly when you say you do cost benefit analysis for one not for society now another final question about testing aside from i mean you have your pet idea about scan tests but um it's not a good idea i think it was very powerful but yes they see the results here in cellular automata doing simulations and what happens if you have a bed a bed test and you define you can define a bad test as something that has a very high uh false negative rate right and that which is a high false positive rate is not a big deal as a big deal and then you can compute what it does and effectively it does wonders even the bad tests works uh because plus if you make have a correlation between uh false negative and how infectious a person is right but it depends so the details here matter right so if you use the test and you say to people which some people are saying hey you can take this test and then you can go out and party um that's not gonna work because the the transmission is still going to be larger than one and then you're going to have zoom it's going to go up just you know go off a little bit slower you need to know what rate what rate and what what both false negative rate and and what rate you need to bring your your r0 below uh one right so the point is that let's say you use this test in addition to things that are already doing a fairly good job like r is already one and then you add another testing to it and then you reduce r and now you have a decline and you can get to zero and that's great that's super important and that solves the problem and then what happened is the compounding effect we've discussed before the the mask situation if i were a mask and you were a mask all right uh the the benefit is uh uh squared right like my mask is effective uh 50 years 50 total 75 it's the same thing with testing plus mascara for example case studies canon in a game next week i'm going to the to the eastern mediterranean to the republic of phoenicia right which is the republic newly formed republic and i'm i'm gonna board the plane in new york everybody must be pcr negative within the last 72 hours okay if you have that plus a mask you see a compound so you need the joint distribution right so that sorry there's a particular those who are not who are probably contagious okay and have a probably infected and have a false negative there are probably going to be a few on the plane maybe not by the way those will not be likely to be infectious there's one problem with what you're saying which is that there's a study that shows that in a plane where they were all using masks and there was only one or two whatever people who were infected there was someone else who was infected now it may be low probability but there was someone who's infected so the key that they think is these these plane rides within the united states inside the united states do not require a negative pcr yeah the the the additional piece that i would add is it is risky to use the bathroom so we have to figure out a way to make going to the bathroom in an airplane much safer it seems to me that the person did not have it i mean the way i use it i use the bathroom i have two surgical masks plus all kind of apparatus on top plus sometimes i'm 95 if i can handle and it seems to me that if you uh and don't stay it's also it's proportional to time yeah don't spend a lot of time in a bathroom also don't sit on the seat okay yeah well it's you know careful the point is i have one more piece for you i have one more piece for you before let me finish my sentence if i tell you that these uh infections on planes uh happened uh on a population that was not pcr negative or assumed to be pcr negative yeah but in this case they knew that there was a very low probability of someone being infectious being infected there was they they found out so the point is low probability in new york today you have about one in uh 200 people one in 150 people of those tested test positive and to get tested it's almost impossible unless you uh you know you you can't walk or something unless you're traveling okay so so it seems to me that the the the the odds of having someone infected in new york is reasonably low today yeah well the problem is that they still have 600 cases per day and the fact that they have 600 cases per day means that they actually you can't open up restaurants in a normal way you can't open up the schools in a normal way or even maybe that would be another if you want debate and open up the schools right but i have one more piece to add which is there is a there is a challenge that we're facing that these uh relatively low um quality or low accuracy tests are super helpful for and that is if you get to zero if you eliminate the disease people are worried that you're going to have a new outbreak because someone is going to come in and do you know do a new outbreak so the key parameter there is if you have someone that comes into the community and is contagious how many cases will there be before you can stop it and and the answer is that depends upon the visibility it depends on people being careful all the time so the trick is that if you use these tests that you have saliva tests you can do on yourself and you have a lot of people doing them then you can detect when there is a small outbreak very early and stop it super fast so the likelihood that you'll have a larger outbreak is everything if the test costs one dollar then we can open schools and test students every day no no either you got it going down or you have it going up if you have it going up you have more cases and more cases if you have it going down then you get to zero and then you can open up the school you don't think that testing the act of systematic testing would would allow you to very quickly identify who in the population right but if you're quickly identifying who in the population has it then by definition you can stop it so there's no there's none of this stuff of we have the disease and we could do testing and do it either this is a next multiplicative process right either it's going to grow or it's going to shrink if it's going to shrink you get rid of it and you're done or if it's going to grow you have more of it so there's none of this this idea that you can somehow keep testing and testing and maintain it in the population it doesn't really make sense to me because it's multiplicative okay so we'll that would be the topic hopefully on our next debate uh try to work out the math here and uh thanks uh yen here so we should do these more often no absolutely absolutely i tell you it's it's been a terrible time you know thinking about all of the suffering and and loss of life but it's always a pleasure when we get together which makes uh you know doing something with people and by the way that's sort of general right we care about each other and and that's what makes it uh worthwhile yeah we can't we care about society we care we're not alone and that's what people can't get and this virus shows us how dependent we are on risks taken by others great thanks incidentally my my background is nicer than yours it is definitely i really love it i'm glad that i'm looking at it you're looking at books or something or cooking uh instruments or something okay all right take care thanks everyone bye have a good day bye bye

Channel: New England Complex Systems Institute

Views: 2,330

Rating: 4.875 out of 5

Keywords:

Id: 1oGXm102ZCE

Length: 33min 55sec (2035 seconds)

Published: Fri Sep 04 2020