Nicky Case has just made a new excellent explorable explanation, in collaboration with Marcel Salathé, a Swiss epidemiologist. It is about (drum roll) ... ... ... COVID-19! 😷 In fact the article nicely explains the spread of the virus and the effects of collective actions on it. With lots of tunable simulations, as usual. It helps me a lot reasoning about all this. In fact I was lazy until now 😴 and did not try to truly understand what is going on. So thanks, Nicky. 👍 https://ncase.me/covid-19/

That thumbnail image is fantastic.

These are fun simulations, but also tragically misleading, because the human organism has a multi-billion year evolved internal defense system that adjusts automatically, and communicates between people in an as yet unknown manner. So there is nothing constant about these values, and Sweden is demonstrating that you can inoculate your society by letting children get exposed and develop defense at a cost of some old men dying. There are brilliant interviews with some very knowledgeable people in Sweden, and they are looking like they have middling results with zero lockdown, which is quite amazing, and totally contradicts all of the computer models. So Sweden 1, Computer 0. A virus, per the book "The body victorious", can create 10^72 copies of itself in 12 hours. That is not a small number. People have no idea how many viruses there are in our environment, how far they can travel, and how completely unavoidable they are once they settle into the population. It is also medical misinformation to constantly portray viruses as only negative, just like bacteria, when we rely on special viruses to survive (bacteriophages used in our intestinal tract to control bacteria). We have really dumb people in the news spreading fear, who don't know squat about virions, and their care and feeding, and what it is to be the assemblage we call a human being. Computer simulations of this kind, which over simplify a very complex process, cause societies to make poor decisions. A similar problem occurs with the global warming models, which are even more flawed. People blindly believe computers now, which a mistake because people wrote the software, and they are fallible.

I’d happily take a world where everyone had read, played with, and understands the basics of @ncase’s explorable explanations, however “wrong” they might be over whatever we have right now.

For me what makes explorable explanations in general, and this one in particular, useful is not the answers but the interactive informations that help grasp some important aspects of what is going on. And to make it possible, of course it has to be simplified. And it is said explicitly several times that the reality is more complicated. Explorable explanations are tools that help people understand things. At least in that case it helps me a lot understand some basic things about COVID-19.

Fundamental flaw in the linked simulations: the 1 in 20 going to ICU is fixed and probably wildly wrong.

You are right, the parameter corresponding to the 1 to 20 going to ICU is very important and it would have been nice to make it explorable as other parameters. Alas, I have just checked and it is buried in an obscure formula in the JavaScript code. This is an argument in favor of explorable explanations with explorable/editable code ! 🙂

For those interested in a discussion with a very knowledgeable modeler of biological systems, here is a prof. from Stanford Medical School talking in an interview, and near the end he goes into some detail about how irresponsible it is for models such as the Neal Ferguson model which caused the UK to dramatically change its position on protocols. Being off by a factor of 100 is considered by some to be worth it because it generates action, but this speaker was dismayed at the attitude that being off by a factor of 100 is not good science. Not many people have knowledge of computer systems and medicine. He is a nobel prize winner, so please don't discount him too quickly. From the Cambridge Biological Systems group, which has generated 26 nobel prizes. Arguably the best lab in the world.

He might have a Nobel prize (in an unrelated category), but (a) has a questionable understanding of R0, (b) considers Sweden and Germany in the same category of “good examples” for “not extreme lockdown”, and (c) assumes herd immunity as a given. That’s enough severe mistakes to discount him for me. Sweden and Germany have taken very different approaches (Germany is pretty much locked down, not sure why he thinks otherwise) and the two countries are almost an order of magnitude apart in deaths per million people, Sweden is killing people faster than the US in relative numbers, but to his credit he also says from an economic perspective younger people are more worth than older people so as long as only older people die it’s ok. (I’m not making this up, watch the video at 15:42.) Also, we don’t know yet for sure if people who contracted the virus will be immune and for how long. For his version of reality to play out he needs to be damn sure that immunity is guaranteed. Taking the publication into account this is political propaganda trying to dramatize the economic effects of the pandemic. I’m sure they’re happy they found a Nobel price laureate who is willing to provide his credentials for some questionable misinterpretation with a political agenda for people who are easily impressed by such things.

I agree, this isn't the best link to post, to back the end-lockdown argument, but that doesn't mean the arguments against lockdown aren't very strong. But maybe we've strayed from the original intent of this thread - to discuss explorable explanations. 😄

Well, if we want to come back on trail of the original thread (i.e. explorable explanations as tool to help people grasp complex topics) and yet want to discuss the view/outcome of the original article from Nicky Case and Marcel Salathè, we can find a very interesting follow up (at least for me). I understand from your comments, @Edward de Jong / Beads Project and @Duncan Cragg, that the simulations in the article miss some aspects of the situation that can be viewed as important to fully understand what is going on. So, if you are incline to do so, to move forward, can you tell what you consider as useful information/simulation from the article as it is? what is wrong? and what is missing? Ideally it would be nice to see what can be done to give the "reader" all the controls (parameters and ultimately the code) to actually play with this and to feel/understand more things than before.

One of the problems with simulations is that they don't expose their assumptions. If you watch this video by a very august Stanford Prof, you will see how the key Ferguson paper made some bad assumptions, leading to an error of over 100x actual results, and how specifically the data doesn't show exponential growth, which is embedded into most of the covid models. He also points out that the Korean data shows that being indoors makes you 18 times more likely to get the virus. The exploratory models don't include an indoor/outdoor switch. Traveling inside the NYC subway for an hour each way every day is way different risk than riding on horseback in Montana. The amount of recycled air you are breathing is a massive factor. The other thing is that stats for the virus are very dependent on the age of the person. Unlike the flu, this virus doesn't seem to bother young kids much. A significant percentage of acute cases are for men over 80. The models could have some immunity transference factor that lets you adjust how much immunity is somehow transferred. Then you could see that a high transference factor would suppress the growth rate. The other thing people don't seem to grasp is that viruses are part of our everyday life, and that social distance stuff is pretty nonsensical. Here is a Nature report showing that Flu virions are found in every liter of air you care to sample in a school. Thus this whole event focusing on the virus is a misdirection instead of looking at measuring the robustness of our immune systems. Here is the article on the flu virus: https://www.nature.com/articles/s41598-020-58588-1 As you can see from that article you get bombard by hundreds, thousands of virions, and maybe they get going in your body, and in an environment like a school it is just minutes before you are going to encounter them, and therefore since the air is saturated with them, you have to have a defense. People have been playing around with linear vs. exponential growth curves since Thomas Malthus, and invariably exponential equations always result in some catastrophic endpoint. The Club of Rome simulation underlying the "The Population Bomb" by Paul Erlich is another alarmist work based on a completely inaccurate computer model.

How might a simulation be made that does expose its assumptions?

If every numeric parameter is exposed so that no parameter is embedded inside the code, that would help. Invariably there are formulas with exponents or multiplicative factors that are super influential to the results. There should be no hidden constants or training data that forces it to match historical results but is basically a cheat. In the climate models this is a huge problem; they have matrices of training data that guarantee it looks really accurate when in reality they are using polynomial curve fitting to match the data without a real model. The other sleazy trick the climate people use is to "adjust" the data. Recently a british lab threw out hundreds of years of temperature data because it was too erratic (and makes their models harder to fit). If you don't measure temperature at the same time of day it is fairly hard to fix it. You can hide a lot of bad things if your program is big enough, and the climate people have huge datasets, which prevents outsiders from checking their work in any reasonable amount of time. There has been so much data fudging it is embarrassing. The temperature of the earth is a strange thing to quote; at any point in time there is a 100 deg range of temp on the planet. Virus simulations are way easier. If you have population density factors, along with "packing factors" in terms of how much people are squeezed together, and "air recycling" factors that would help. Also the number of days over 100 deg F, plus days above 80% humidity (above 80% humidity the virus' stick outside loses ability to stick, like a post-it note). Just in New York state alone you an see the huge difference between NYC and upstate NY which is very rural. The rural environments have healthier people, more outdoorsy, etc. You could build a simulation and tune the parameters so it matches the cases way better than the simplistic, wildly inaccurate model that Neil Ferguson used to change UK's whole policy. Groups who generate inaccurate models should be punished, and the teams with good models should be rewarded. There needs to be some accountability in computer modeling, esp. if the population is going to obey them with blind faith.

So in summary.. expose all numbers as params, and avoid opaque datasets. I'd say the formulas are just as important. But when you expose too much, it becomes less accessible, and thus less useful as a teaching/learning tool. You don't need an encyclopedia entry to contain every possible detail in order for it to usefully convey critical info. So we need some other solution than just "make sure everyone is fully stats-literate and give them all the data and formulas". That doesn't scale.

Mr Beads: do you have a reference for the throwing away temperature data? I presume you follow Tony Heller? (Apologies for derailing once again)

Bringing it half-way back to the original discussion: it seems that simulations get more scrutiny than written media — why is that? On the one hand I can see that giving the reader/user more authority to come to their own conclusions manifests as sort of a loss of control on the author/developer side. In written media we seem to have accepted lower standards and let such things pass, often if at least some references have been given (but we don’t look at them to verify) or given the article comes from a “reputable” source (which means different things to different people, depending on their filter bubble).

@Duncan Cragg here is the NY times article on the actual temperature data being replaced by "homogenized" data. https://archive.nytimes.com/www.nytimes.com/gwire/2009/10/14/14greenwire-scientists-return-fire-at-climate-skeptics-in-31175.html The source code to the UK model that caused billions of dollars in damage has been posted, and i saw a reddit thread of people who were plowing through the 50,000 lines of C++ (although written basically in C). There a key 5000 line long function that is the core of the model. Almost completely uncommented, and using plenty of single letter variable names, written over 10 years, this is not the kind of quality software you want to trust. The code for important govt. models should be open to scrutiny. There are computer models that are accurate. For example, in Forestry, the FORESEE* model predicts forest growth*.* The reason they are accurate is that the people running the models take their measurements over a long period of time (20+ years), and adjust their parameters so that the growth factors are accurate. And you try new sample areas and compare how the predictions worked out. The Foresee shows that if you take a forest and divide it into 100 equal area regions, and clear cut 2 of them, in 50 years you will have cycled through the forest, and this practice generates ultimately 20-30% more tree cover without any damage to wildlife. Optimization can be improved with computer models, but you have to take the time and effort to verify they are accurate.

(sorry busy times, so I can't come here as often as I want even if this thread interests me a lot)

Here's my public comment on this topic, which I think is very much in the spirit of a better "future of coding": https://blog.khinsen.net/posts/2020/05/18/an-open-letter-to-software-engineers-criticizing-neil-ferguson-s-epidemics-simulation-code/

one month ago some people claimed Sweden was showing everyone else is wrong. Do we still believe that?

That's the daily average from what I can see; I think total deaths are comparable to the UK if I recall correctly?

@gman the graph that you are showing is highly misleading. 1) it will be some time, probably not until 2021, until the dust settles and we see the actual changes in the total death rates so we can determine how much extra death covid cost. Keep in mind there are no agreed upon standards for reporting cause of death. Did someone die with covid? or did they die of covid? Or did they die of 10 things, and covid was one of them? So these charts are useless. There is a talk recently from a top statistician in UK, and so far it looks like 80% of people in the UK appear to be immune to the disease for unknown reasons. They are cranking the numbers going backwards from effect to cause, and discovering that Vitamin D, spicy foods, all sorts of things seem to offer significant protections. Even in the USA totally different regimens have been followed; South Dakota vs. California for example. Belarus' plan from their president was to gargle with vodka and take saunas. So in 2022 we will be able to look back dispassionately at the numbers and see what worked. People want Sweden to fail in Norther Europe because they bucked the crowd. See this talk by Karl Friston for some very interesting mathematical talk The problem with using poorly designed mathematical models which don't even have a parameter for susceptibility, is that you can generate such alarming results that it frightens people. The truth is that the virus is so small and untrackable that the underlying mechanisms cannot be directly observed, and thus working backwards from results to causal factors becomes the best way. And you can't do that while it is happening, this is all rear-view mirror processes. I predict they will be arguing about which way was better for years, everyone trying to justify the decisions they made.