Hi everybody, today I'm here talking to Niajesh\h Afshordi who is an astrophysicist at Perimeter\h\h

Institute in Waterloo, Canada, and you've\h heard of him before on this channel because\h\h

we talked about his idea of measuring black hole\h echoes a while ago. So if you want to know more\h\h

about this please go and check out this video. But\h in the recent years - aah, years -- in the recent\h\h

months Niayash has developed a new\h interest which is pandemic models.\h\h

So, Niayesh, why don't you\h tell us how you got into this.\h\h

Oh i got into this how everybody else got into\h this because we all got locked down when the virus\h\h

started getting everywhere around the\h world and including Canada where i live and\h\h

one thing led to another and just before you know\h it everything was shut down and we were homebound\h\h

for months and we didn't have much else to do uh\h except trying to understand what's going on around\h\h

us uh particularly on social media everybody was\h discussing and trying to kind of understand and\h\h

predict what's happening with this thing uh and\h of course there were many many wrong predictions\h\h

but then it occurred to me that i'm a scientist\h and a scientist we are our job is to try to\h\h

understand things that happen in nature and this\h this is one example of something that's novel and\h\h

is happening in nature uh i don't know if it's as\h novel as black holes or or not but it's something\h\h

that you could try to understand and the great\h thing about the pandemic there aren't that many uh\h\h

but one is that it there's a lot of data and it's\h much easier to get data about it compared to the\h\h

gravitational waves on black holes so so that's\h what i tried to do and i i got together in fact\h\h

interestingly there was a a colleague who\h was visiting me uh he his background was in\h\h

virology but he wanted to switch to gravitational\h waves and work on gravitational wave echoes but\h\h

then it turned out that at the same time that he\h arrived this virus arrived in the world stage and\h\h

um longer story short he never got to work on\h echoes so so we ended up trying to understand\h\h

what's going on around us while every\h everybody was going into lockdown and\h\h

uh and the main thing is trying to understand is\h that so what is the data that we need uh together\h\h

to to get to the bottom of what's happening and\h of course the pandemic data was everywhere you\h\h

could download it but that's not that's not really\h uh that's the effect the question is what's the\h\h

cause i mean we know the cause is the virus but\h the virus has to jump from one person to another\h\h

and uh basically we we started we sat down and\h about virtually sat down and tried to understand\h\h

what are the things that help us predict uh\h how virus can jump from one person to another\h\h

of course the requirement is that you need to have\h a lot of people close to each other and and that\h\h

was that was our starting point basically what\h are the measures of uh population density that\h\h

we have across the world uh and are kind of the\h the spark there was that uh people usually talk\h\h

about population density but that's some sort\h of an average thing on some large region that\h\h

doesn't really matter what matters is uh how many\h people are next to you so this is something that\h\h

few people know about but it's called population\h weighted population density and it turns out that\h\h

that's the most important thing nobody had has\h been talking about and that's the best predictor\h\h

of where the virus is growing and how it's\h growing or how fast it's growing of course\h\h

where this starts it's anybody's guess because it\h depends on where it when it got there but how fast\h\h

it grows it really depends on population weighted\h density and then of course other things happen\h\h

there were lockdowns and there was weather that\h happened over time and then we started kind of\h\h

putting the data together on lockdown it turns\h out google may record cell phone activity in\h\h

various places so you could add that to the\h model and then there is weather that you can\h\h

gather for various places so we kind of started\h putting more and more things into the model\h\h

to basically uh predict how fast transmission is\h happening and you could calibrate it with all the\h\h

data because we had lots of data available both on\h mobility and weather but also under pandemic half\h\h

assets was growing so that's kind of how maybe\h if i if i may briefly for clarity at you you're\h\h

using data for the united states right\h so it's not globally that's right so uh\h\h

the reason we kind of started focusing on the\h united states was we had a homogeneous data set\h\h

that we kind of i mean no is anybody's guess what\h is the quality of data i mean every data has its\h\h

own problems but at least we thought as united\h states is more or less homogeneous country so the\h\h

kind of problems that appear in the data would be\h similar in different places and we have recently\h\h

started working on other countries but it's a\h bit it's more of an uphill struggle because uh\h\h

understanding the quality of various data sets\h is is harder right so that's that's how we have\h\h

of course that so we looked at the kind of the\h smallest scale in the united states that we could\h\h

which is the county level so there are around 1300\h counties in the united states in other words 50\h\h

states each it's just it has many counties in it\h and there are 500 counties that have a significant\h\h

enough epidemic that we could actually fit the\h model to and the main thing that we did that\h\h

nobody else has done is that we actually use the\h same parameters to fit all the epidemics that are\h\h

happening everywhere so all 500 epidemics\h were fit by fitted by the same parameters\h\h

and the differences are really different\h between the conditions of the counties they\h\h

have different population densities different\h weathers different mobility restrictions\h\h

and these are all the things that are recorded\h so we didn't have to assume something for it so\h\h

these are all the data that's available publicly\h and basically all we have to do is define the\h\h

right parameters and they're basically about 10\h parameters that we have to fit to we have to find\h\h

the best values that would fit this 500 epidemics\h and there are basically roughly 7 000 data points\h\h

that we could fit with these 10 parameters and\h so that's that's kind of where where we kind of\h\h

where we are right now and we started with less\h data but that's that's where we are right now so\h\h

maybe that's a good point to add that Niayesh has\h written a paper about this already in July right\h\h

so a month ago and that's on the archive and\h i will add a link in the information below the\h\h

video and he also has a website uh i can't\h recall the url but i'll i'll leave that too\h\h

maybe you can tell us a little\h bit about what's on the website\h\h

yeah so the website is nafshordi.com/covid so\h that's uh easy and um right so so the main idea\h\h

and this is uh i really hope that if you had\h a little bit more help help we could kind of\h\h

extend it to elsewhere in the world but we so far\h we have only expertise that all the data we need\h\h

inside the united states uh so basically based\h on this model that we fitted to all the data\h\h

and and i should say the model is is in fact is is\h a epidemiological model so a model is a model of\h\h

people interacting with each other uh and getting\h the virus and then incubating the virus and then\h\h

transmitting to other people so it is not some\h like fitting function even though there has three\h\h

parameters that we have to fit to the data but a\h model is basically a model of people in tracking\h\h

with each other getting infected and then at some\h point becoming infectious themselves so this is\h\h

that that's that's the kind of model that we're\h dealing with and and the nice thing about it is\h\h

that basically if you if you can constrain the\h models then you can predict where things are going\h\h

and then based on the data you could say\h okay so if if this county goes to lockdown\h\h

say 50 of people have stopped going to work\h what's going to happen or if people in this\h\h

county everybody starts wearing masks or nobody\h uh wears masks what's going to happen so that's\h\h

that's the power of the model because the model is\h a physical model of the world and it it quantifies\h\h

dependence on various physical properties so if\h you go to the website and in particular if you\h\h

go to our dashboard it tells you each county what\h happened in the past and it tells you if people\h\h

go to lockdown or lift the down at various\h levels uh what's going to happen into the\h\h

future and it also uses actual historical weather\h or whether in the past and whether into the future\h\h

because that turns out to be a big factor\h it turns out and if if weather is very cold\h\h

the virus spreads more quickly but but\h surprisingly if it's also very very warm\h\h

it spreads quickly and that's another thing\h that is discovered and so so we have all of\h\h

these different so wait it spreads quickly when\h it's very cold and very warm but not in the middle\h\h

not in the middle so uh so i uh i don't know if\h this is the right explanation but i think that my\h\h

explanation for this is that basically there is\h an optimal weather for people to spend outdoors\h\h

if it gets too hot everybody goes indoors in in\h air conditioning at least in the united states\h\h

uh and then if it gets very uh cold the\h same happens everybody goes indoors so i\h\h

think it's probably really a proxy of how\h much time people are spending doors right\h\h

but that's that's an effect that's there and it's\h it's very significant in the data uh as is many\h\h

other effects and so there is the effect of of\h course shutdowns when people are stuck going out\h\h

then the virus doesn't spread so that's obvious\h the one that is most surprising to us but also\h\h

most significant you could argue is\h that places that has had a lot of\h\h

mortality those are the places that a virus stops\h spreading and that's independent of all the other\h\h

effects but it's an effect that that's uh that's\h expected and that's related to the immunity\h\h

and the more people get infected the less\h susceptible people uh remain in the society\h\h

so that would slow down the spread of the virus\h now we're not sure that's what's happening but\h\h

that's certainly consistent uh with the with\h the model and with the expectations from uh from\h\h

epidemiology uh and yeah so that's that's another\h effect that we find that in basically places where\h\h

somewhere between point one percent two point two\h percent of the population uh have uh passed away\h\h

from covid uh then those are the places where\h um basically the epidemic stops the spreading\h\h

it doesn't mean that it stops it's just that it\h doesn't exponentially grow anymore and that's\h\h

another thing that comes out of our analysis okay\h so you're not actually using any demographic data\h\h

is that right uh we are using democracy we use\h age distribution um yeah so we have uh that's\h\h

that's another dependence that uh uh in fact the\h the number of older people or the density of uh\h\h

older uh citizens is is is more important for the\h uh for the prediction than the younger people so\h\h

uh older populations have a faster epidemic growth\h right so yeah so we have we use that as well and\h\h

a host of other information we also use google\h trends for face masks so basically that would\h\h

be uh that's our proxy or guess for how many\h people are wearing face masks is basically\h\h

if they're searching for face max and google\h then probably they're looking for buying them\h\h

that's another there is some correlation\h with that as well as it turns out\h\h

so so what would you say is the biggest difference\h between your model and let's call it you know the\h\h

standard model the compartmental models um with\h you know the usual models that they are using\h\h

with the susceptible infected uh recovered\h division yeah uh i think uh our model is uh\h\h

it's not terribly different\h i mean you could say that\h\h

our model is kind of a slightly more complicated\h version of that our epidemiological model\h\h

so that's really not that different uh so i think\h if if you have like two or three compartments you\h\h

could kind of reproduce what our model\h does in terms of uh its uh its dynamics\h\h

uh but i think the main power that\h we have is we actually identify\h\h

how the parameters in the model depends on local\h conditions of the society so historically what\h\h

people do do and i mean i've looked through these\h papers over the years and of course for this\h\h

pandemic they either assume that they are constant\h these parameters are constant and then deal with\h\h

the mastoid models or they assume these various\h stages of lockdown and reopening and then you have\h\h

to kind of keep adding various stages to fit the\h data and then you do this separately for different\h\h

cities or counties or metropolitan regions and\h of course at some level i mean if you have enough\h\h

parameters and enough stages of reopening and\h shutting down you can fit anything you want right\h\h

um what we do is that we actually identify um\h the drivers that you could actually objectively\h\h

measure in various places based on various\h datasets that we have available from google\h\h

to weather to uh to the epidemic history and\h ah and says you could use can you use the same\h\h

parameters everywhere because it's the same\h virus and these are same people so basically\h\h

can you use the universal model so the same way\h that we could use Kepler's law laws to predict\h\h

uh planetary dynamics for all the planets can we\h use basically the same universal laws to predict\h\h

epidemics everywhere in in the united states and\h hopefully across the world and that's kind of i\h\h

think that's the real power which i haven't seen\h done every elsewhere uh because so at other places\h\h

you just try to defeat different laws in different\h places and we try to fit the same law everywhere

so you you already said you've been working\h with a virologist um together um and so you're\h\h

a physicist uh you have some other people\h people on your team yeah it's um so in fact\h\h

so that my first collaborator is ben holder\h he he is a physicist by training but he's\h\h

been working on um modeling virus growth in\h human bodies for the past 10 years so it's\h\h

he wasn't like doing epidemiology but i mean\h there's some of the formalism is similar um so\h\h

he he was he's a physicist by training\h but he has a lot of work on on virology\h\h

uh so we had we then got some people from Wolfram\h because they were interested in data and covet\h\h

and i knew how to use Mathematica so\h somehow that that matched um and then\h\h

more recently Steve Weinstein has joined us he's\h a philosopher of science and i think that's i\h\h

think that's a good uh quick possibility because\h he knows how scientists can talk to each other\h\h

or don't uh as as you do probably uh and and\h i i'm hoping we can kind of find a good fusion\h\h

in this collaboration he has a physics background\h doesn't he i actually don't know i mean i know\h\h

he's in philosophy department uh but i mean\h he's certainly very much interested in physics\h\h

yeah yeah yeah so um so so your your group is\h very physics heavy um how has the reception of\h\h

your model been by the epidemiology community\h yeah i think it's they have for the most part\h\h

ignored us uh i have talked to uh a few of them\h i for the most part is i haven't heard back\h\h

um there is a small fraction of a\h small number of epidemiologists who\h\h

who kind of take this question i've heard\h immunity or studied this question i've heard\h\h

immunity in more detail in particular there\h is a there is a a mathematical modeler in um\h\h

Scotland, Gabriela Gomez and she's become famous\h because she has she has shown that in in this in\h\h

some models and other peoples have shown this as\h well the herd immunity threshold could be much\h\h

lower than what classically people know about so\h classically if you look at a kind of website and\h\h

textbook they say that if for example for Covid\h that the production number is three or four\h\h

somewhere between 60 to 80 percent\h of the society needs to be infected\h\h

to reach herd immunity and so herd immunity\h basically means that if this many people get\h\h

infected 60 or 80 percent then beyond that point\h there are not enough people left in the society\h\h

uh so that covid would exponentially grow so it\h would it would decay from that point on it doesn't\h\h

mean that people stop getting infected it just\h means that there will be fewer and fewer people\h\h

who get infected um now it turns out that these\h simple estimates there assume a homogeneous\h\h

society by that it i mean that everybody is likes\h everybody is like everybody else uh every person\h\h

has the same probability of getting infected\h and same probability of infecting others once uh\h\h

infected and of course in real societies that's\h not the case and in particular for covid we\h\h

know that's definitely not the case and it was\h shown that in fact this threshold is much lower\h\h

if you have basically super spreading events\h or a lot of super spreading driving the\h\h

covet infections so there are already published\h papers that say that this number could be as\h\h

low as 40 percent and there are other works\h that are more controversial for reasons that\h\h

i don't understand but they say that if you just\h fit the data you get more like 10 to 20 so this is\h\h

work of Gabriella Gomez and her collaborators that\h says that you only need 10 to 20 of the society\h\h

to be infected uh or community to be infected\h although this number does vary from place to place\h\h

uh so such uh i talked to her and i think\h it looks like what we're finding very much\h\h

consistent with what uh she's finding uh of\h course the problem is that we don't we never\h\h

really know how many people in a large community\h get infected because you cannot test everybody\h\h

but if you go by a number that we're\h finding based on just mortality\h\h

so we find that places that between 0.1 to\h 0.2 percent of the population die of covet\h\h

you reach heard immunity uh and if you say\h infection fatality ratio is about one percent that\h\h

means that in these places ten to twenty percent\h of people have gotten infected uh and so so i\h\h

think uh our results very much agrees with her\h and she's uh i mean she's in agreement with that\h\h

but the rest of the epidemiology community just\h come completely ignored us uh i would imagine\h\h

the same thing happened if epidemiologists write\h a paper on big bang or black holes and i would\h\h

probably ignore them so i don't uh i don't blame\h them too much to be honest but uh here here we are

yeah um well i would argue that the\h situation is a little bit different\h\h

because i mean as a physicist\h and i mean you've worked on\h\h

astrophysics with a lot of data analysis and\h this is basically a piece of data analysis so\h\h

i think it's something that deserves to be\h taken seriously well i couldn't agree more

um yeah okay so i think i think you've pretty\h much answered uh all my questions um i think\h\h

that's a good point to thank you for the\h conversation and thanks everybody for watching\h\h

thank you very much sabine