The 1977 flu killed about 700k people, and is uncontroversially accepted to have arisen from a failed vaccine trial. I've seen legalistic attempts to define that as something other than a lab accident, but there's no question that the virus spent 1957 to 1976 in a lab freezer.
No prior pandemic is known to have arisen from a genetically engineered virus, but the technology for that has only recently come to exist. No one had ever died in a plane crash before the Wright brothers, but the risk was obviously still there.
The 1977 virus was already circulating in the population, hence why it ended up as a vaccine. Therefore, it was not a lab leak (or any human-driven mechanism; nobody's certain) that first caused humans to be infected with the virus but good, old fashioned, zoonotic transmission.
>> No prior pandemic is known to have arisen from a genetically engineered virus, but the technology for that has only recently come to exist. No one had ever died in a plane crash before the Wright brothers, but the risk was obviously still there.
Unlike plane accidents, diseases and pandemics happened way before labs that could leak them.
We're talking about prior probabilities here, yes? "How often has something happened in the past". Whether the probability of lab leaks increases now that there's a possibility of a lab leak is a conditional probability: "given that a lab leak is now possible, what is its probability?".
And btw that will remain very low because of the infinitesimally low prior probability of a lab leak.
> The 1977 virus was already circulating in the population, hence why it ended up as a vaccine.
The 1957 flu evolved naturally, but it went extinct in the wild. By 1976, it wasn't known to be circulating anywhere. It then reappeared in 1977 with an almost identical genome, without the mutations it should have picked up in 20 years of cryptic spread. This is consistent with storage in a laboratory freezer, and very little else. The exact site of reintroduction isn't known, but it's generally believed to be a vaccine challenge trial in China:
> Although there is no hard evidence available, the introduction of this 1977 H1N1 virus is now thought to be the result of vaccine trials in the Far East involving the challenge of several thousand military recruits with live H1N1 virus (C.M. Chu, personal communication).
So the scientists who undertook that trial (or whatever the activity was) caused that pandemic. If not for their actions, those 700k people wouldn't have died. Of course a different flu virus would have circulated, but 700k was much worse than an average season.
Does that not concern you? The reckless trial caused those influenza deaths, just as surely as the leak at Bhopal caused those poisoning deaths. The long chain of transmission between the site of reintroduction and the site of the deaths doesn't change the causality--if those researchers hadn't reintroduced the virus, then those patients wouldn't have died.
> Unlike plane accidents, diseases and pandemics happened way before labs that could leak them.
That's not a meaningful statement. Cancer happened way before radiological accidents that could induce them, and most cancer is natural; but Edison's X-ray tube nonetheless killed his assistant. When a new technology creates a new way to die, a period will exist in which it hasn't killed anyone yet; but someone can still be first.
The correct guide is therefore the most similar pre-existing technologies, which in this case have already killed 700k people. It's possible that by failing to take any lesson from that, researchers at the WIV have now killed an additional 7M. If that's the case, then I hate to think what comes next.
This was a discussion about "prior probability", according to the terminology that you introduced. I pointed out that if you really want to use "prior probability" then you have to take into account the, well, prior. probability. Of the two events you're considering.
Even if you insisted on the 1977 virus being an instance of a lab escape causing a pandemic, the prior probability of that kind of event would stay zero, maybe with a desultory and lonely 1 after three or four decimal zeroes. So, zero.
Are you using "prior probability" in a figurative sense? Are you saying that we should set our priors according to personal preference rather than observed frequency of real-world events? If so, well I guess you're a true Bayesian then and I salute you, but please make that clear.
>> That's not a meaningful statement.
Yeah, thanks a lot for the respectful debate there. You should read again the bit where I point out the difference between prior and conditional probabilities and if you really want to use the terminology "prior probability" make yourself more familiar with it.
It came from the first article that I linked, by a medical doctor and professor of epidemiology. From that article:
> Over the ensuing years, after its emergence in 1977, the virus went on to infect a significant portion of the worldwide population, killing roughly 700,000 people.
I'm not aware of any controversy over that number. It seems like people don't quite believe it, as if it's impossible that researchers would kill 700k people with no consequence, so it must be wrong or fake somehow. They're just as dead as from any other cause though, and the failed challenge trial (or whatever it was) caused the deaths.
And no one cares, and I don't really understand why. It may partially be the delayed discovery, since research-related origin was denied at the time. (Quite familiarly, the WHO wrote in 1978 that "laboratory contamination can be excluded because the laboratories concerned either had never kept H1N1 virus or had not worked with it for a long time".) It may also be the long distance between the initial cause and the death, in the same way as for pollution, climate change, etc.
> Even if you insisted on the 1977 virus being an instance of a lab escape causing a pandemic, the prior probability of that kind of event would stay zero, maybe with a desultory and lonely 1 after three or four decimal zeroes. So, zero.
The flu vaccine was invented about 80 years ago, so a pandemic arising from it couldn't have occurred before that. We obviously haven't had 10,000 to 100,000 pandemics since then.
The definition is loose, but counting introductions of pathogens into humans with a combination of novelty and mortality sufficient to get scientific attention over that period, I get maybe a couple dozen. If we count only pandemics with deaths comparable to the 1977 flu's 700k then it's fewer; but I think it makes sense to include SARS-1, MERS, etc.
If I had to guess knowing only that such a novel pathogen had just been introduced, I'd thus bet with p ~ 5% that it was a research accident. If I had to bet knowing only that a novel pathogen had been introduced sometime in the past million years, then my estimate would of course be lower; but I don't see how that's relevant here. For SARS-CoV-2 I believe additional evidence significantly increases the probability, as discussed in other comments.
>> It came from the first article that I linked, by a medical doctor and professor of epidemiology. From that article:
>>> Over the ensuing years, after its emergence in 1977, the virus went on to infect a significant portion of the worldwide population, killing roughly 700,000 people.
That estimate is not referenced in the paper you link. I looked for a different source estimating the number of deaths. I found this article [1] that reports an earlier estimate of a mortality rate of less than 5/100,000. "Mortality rate" is taken over the total of a population (not infections). With a mortality rate of less than 5/100,000, for the virus to kill 700,000 people, as in the claim in the article you link, it would have to circulate in a population of 14,000,000,000 (see rule of three).
That's not possible.
>> If I had to guess knowing only that such a novel pathogen had just been introduced, I'd thus bet with p ~ 5% that it was a research accident.
Does that mean you believe every recent pandemic has a ~5% chance to be a "research accident"? SARS, MERS, HIV, Ebola, mpox, Zika, etc, all had a ~5% chance to be research accidents or is Covid-19 special?
What happens to this 5% number as we see more pandemics? Does it go up or down? Can you say?
>> In 1977, an H1N1 influenza virus appeared and circled the globe. Colloquially referred to as the “Russian flu,” as the USSR was the first to report the outbreak to the World Health Organization (WHO), the 1977 strain was actually isolated in Tientsin, Liaoning, and Jilin, China, almost simultaneously in May of that year (1). It was atypically mild for a new epidemic strain; the influenza mortality rate (IMR) of the 1977 flu was calculated to be <5 out of 100,000, less than typical seasonal influenza infections (IMR of 6/100,000 people) (2). In addition, the 1977 strain appeared to affect only those 26 years of age and younger (3). These odd characteristics turned out to have a simple scientific explanation: the virus was not novel. The 1977 strain was virtually identical to an H1N1 influenza strain that was prevalent in the 1950s but had since dropped out of circulation (4).
> That estimate is not referenced in the paper you link.
I linked two papers, and that quote appears in the first. Perhaps you were looking at the second? Here's the link again, with the text that I quoted highlighted:
Influenza statistics are noisy, since most patients are never tested. We could compare the methodologies of those estimates, and try to decide which is better; but I don't think that's a productive use of our time, unless you think 700k deaths are worthy of concern but 200k deaths (or whatever the right number would be) are not.
> Does that mean you believe every recent pandemic has a ~5% chance to be a "research accident"? SARS, MERS, HIV, Ebola, mpox, Zika, etc, all had a ~5% chance to be research accidents or is Covid-19 special?
If I knew nothing else about the pandemic, I'd bet with p ~ 5% that it was research-origin--that's my prior, knowing only that it just emerged. For real, I of course know more. For e.g. MERS, the proximal animal host was found, and genomic evidence shows repeated spillover; so that additional evidence makes me effectively certain it was natural (p ~ 0%). For the 1977 flu, I'm very confident it was unnatural (p ~ 99%). Those average to something close to 5%.
If we get strong confirmation that SARS-CoV-2 (or any other new pandemic) arose unnaturally, then my prior would go up. If we get many more natural pandemics without any unnatural ones, then my prior would go down.
I would also incorporate other information. For example, if many more scientists undertake high-risk research on enhanced potential pandemic pathogens, then my prior would go up. If such projects are restricted and become less common, then my prior would go down. I think that's fairly normal reasoning?
>> I linked two papers, and that quote appears in the first. Perhaps you were looking at the second?
Yes, that's the paper I looked at. "Not referenced" means that the 700k number given in the paper has no reference. The author seems to have pulled it out of thin air. Here I'm copying the text from your link with the highlight:
>> By comparison, the 1977 "Russian flu" virus produced a milder illness in most people (Gregg, Hinman, and Craven 1978). Over the ensuing years, after its emergence in 1977, the virus went on to infect a significant portion of the worldwide population, killing roughly 700,000 people. But this particular pandemic isn't important only because of its impact on disease and death: it also reveals uncomfortable truths about the complex relationship between viruses and humans. The 1977 pandemic is a cautionary tale, one that—hopefully—we can learn from.
Note there's no citation next to the 700k number. So where does it come from? Nowhere. The author just made it up. Perhaps that's the reason for the lack of concern at the 700k deaths: they didn't happen.
And where did your 200k number come from? Can you show your work?
>> If we get strong confirmation that SARS-CoV-2 (or any other new pandemic) arose unnaturally, then my prior would go up.
That's not a prior, it's a posterior: it's the probability of an event given the evidence. We're interested in a prior probability when we don't have enough evidence to calculate a posterior. Like with Covid-19.
Once more I suggest to try and get the terminology right so you don't confuse yourself and everyone else. It will also help you to find better ways to reason under uncertainty.
So the author didn't just make it up. Peer review isn't magic, but it does tend to catch such a blatant level of fraud; so it's odd that was your first assumption. He should have included the reference, but it's not uncommon to get sloppy when a result is widely reported.
I haven't yet traced this number back to the epidemiological methodology. As noted, I fully accept that it might wrong, since influenza numbers tend to be calculated from excess mortality and not directly counted. Until we've reviewed that methodology, I agree that it's reasonable to question the exact number. I believe that uncertainty applies to most epidemiological estimates, including but not limited to the 1977 flu.
But I've explicitly acknowledged the uncertainty, so I'm not sure why you're continuing to focus on this. There is no significant question that the 1977 flu was reintroduced into humans due to a research accident, and no question that it went on to infect a large fraction of the world. There's no question that many people died, even if the exact number is uncertain. Are you proposing that it's few enough that there's no reason to care? If so, what's your cutoff?
The 200k was an arbitrary placeholder. I thought the phrase "or whatever the right number would be" following it would make that clear, but maybe not.
> That's not a prior, it's a posterior: it's the probability of an event given the evidence. We're interested in a prior probability when we don't have enough evidence to calculate a posterior. Like with Covid-19.
I mean that if SARS-CoV-2 were proven to have arisen unnaturally, then my prior that a different future pandemic was unnatural would go up. That again seems very normal to me, since humans predict the future using the past. As I noted, I would adjust my prior based on factors other than a simple count of past research-origin pandemics too; but the count is obviously relevant information. Your updated prior of p ~ 0.0001 to 0.00001 seems to also be derived from a past count, just reaching into prehistory to inflate the denominator.
This thread started when you asserted that "every single pandemic and disease that has ever infected humans did not leak from a lab". Exact numbers for the death toll and prior aside, do you at least agree that's false?
Table 1 in the other paper you cite also has no reference to the number. That's because the number is made up, hearsay, someone's back-of-the-napkin calculation etc.
I'm "continuing to focus on this" because it's obvious the numbers you quoted, and on which you based your entire argument about the 1977 flu, are completely made up and I want you to understand that.
>> That again seems very normal to me, since humans predict the future using the past.
Humans predict the future? That sounds like magic. Are you sure that's true? Are we talking about prior probabilities or palmistry?
Note that it doesn't make sense for a prior to be "your prior" or "my prior". I think that's a turn of phrase that's common in EA circles, where people really have no clue about probabilities or statistics. My advise is to not copy their mannerisms and try to learn about probabilities from a good source, like a textbook or an online course. There used to be a good course on Udcity presented by Sebastian Thrun, based on his course at Stanford IIRC. If you can find it, have a go at it, it's really good.
>> This thread started when you asserted that "every single pandemic and disease that has ever infected humans did not leak from a lab". Exact numbers for the death toll and prior aside, do you at least agree that's false?
Not at all. Besides which the thread started when you used the term "prior probability" without understanding what it means.
Table 1 actually does have a reference in the text, but it doesn't seem to be helpful. I think the major point of unclear methodology is the period over which those deaths are summed, since the lineage has continued to circulate since reintroduction:
> The human H1N1 lineage caused pandemic and endemic influenza from 1918 to 1956, then disappeared entirely around 1957 only to reappear in relatively low-level pandemic form in 1977. It has continued to circulate endemically in humans up to the present time (2009).
Michaelis's Table 1 says "1977-1979", but 700k seems too high for that; so the table seems misleading or wrong to me. Perhaps they're actually counting deaths from descendant lineages up to present--if a terrorist reintroduced smallpox, would you not blame them for every death until it was re-eradicated, even if that took many years? That might be what Burke meant by "over the ensuing years", though in that case it should have increased since 2009. That seems more plausible to me, if anything low.
It seems most likely to me that no one has calculated very rigorously, and the reviewers did back-of-the-napkin calculations similar to those I've just done and decided it was close enough, within the usual uncertainty (rigorous estimates in countries with good statistics are often ranges over 5:1 or more). I agree their failure to explain their methodology is bad, and I won't provide numbers in future without this context.
The calculation you provided earlier seems to be only for the period immediately around 1977, not including descendant lineages in subsequent years. I'd guess that's the reason for the apparent inconsistency, since almost all the death comes later.
So after all this, do you at least agree that (a) many people died, (b) the research accident caused those deaths, and (c) that's bad? I absolutely agree it's (almost surely) not 700.000k; but can you agree it's not zero?
> Humans predict the future? That sounds like magic. Are you sure that's true? Are we talking about prior probabilities or palmistry?
I assume you don't think weather forecasters are magicians? Or almost any human activity involves aspects of prediction--I cross the street because I predict that the barking pit bull might attack me, or I choose an item from the restaurant menu that I predict I'll enjoy eating.
So I don't see what's abnormal here. I don't think this dorm-room epistemology serves much purpose beyond distracting from the practical and serious topic at hand.
> Note that it doesn't make sense for a prior to be "your prior" or "my prior".
I think the possessive is just an acknowledgement that the right answer is disputable, and reasonable people may end up with different numbers. I'd tend to agree there's an EA-adjacent tendency towards misleading quantification, but the concept of a prior here seems fine to me.
> Not at all. Besides which the thread started when you used the term "prior probability" without understanding what it means.
So are you saying the 1977 flu doesn't infect humans, or that it didn't leak from a lab? I'm not sure what you mean here. I also think you've confused me with user
rich_sasha, since I didn't mention prior probability until you did.
Yes, I confused you with rich_sasha, that's right. My aplogies.
As you say 700k people over two years is an impossibly high number so we can disregard this number and any other number that someone makes up just because they want to make a point. That's not the way to make a point with numbers.
>> So after all this, do you at least agree that (a) many people died, (b) the research accident caused those deaths, and (c) that's bad?
Not at all. Note that my initial point, quoting from my initial comment, was that "every single pandemic and disease that has ever infected humans did not leak from a lab and instead evolved in nature, usually in animals".
The 1977 flu did not start in a lab but instead jumped from animals to humans, just like any other disease that has ever bothered us. This happened in the 1950's. In the 1970's an earlier strain of the virus re-emerged but we don't know where it came from and it did not cause a pandemic.
The Covid-19 virus instead is said to have first come from a lab, and some people even think it was a human-made virus, possibly as a result of gain-of-function research. Nothing to do with the 1977 flu.
Have you looked for the Udacity course I recommended, on probabilities? It is really good and it will help you a lot.
> As you say 700k people over two years is an impossibly high number so we can disregard this number
I agree it wasn't over two years. I never said it was over two years, though, and neither did Burke. Michaelis might imply that, and I criticized them for that.
Do you not think it's plausible that all the deaths tracing back to that reintroduction sum to 700k? It's a big chunk of global flu mortality over almost fifty years now. If you think 700k is too high, what's a better estimate? I asked before if you could agree that it was greater than zero, and you didn't answer. If you're still unwilling to, can you at least explain why not?
Perhaps it's misleading to attribute the deaths to the "1977 flu" when our count isn't limited to that one year. I don't think it's that unusual, though. We often say ~7M people died from COVID-19, even though very few of those were in 2019. In any case, I believe the period in question should be clear now.
> The 1977 flu did not start in a lab but instead jumped from animals to humans, just like any other disease that has ever bothered us.
You've changed your terminology here, from "leaked" to "started". There's no question that the 1977 flu, SARS-CoV-2, or any other virus traces back to nature--the technology doesn't yet exist to create a functioning viral genome de novo. All laboratory viruses are derived from natural viruses, whether in simple ways (like freezing for twenty years) or complex ones (like the genetic engineering proposed in DEFUSE).
All viruses trace back to natural evolution. The question is whether their path to humans involved a trip through a lab, and for the 1977 flu the answer seems to be yes. Do you disagree? Even the paper you linked yourself--written by a longtime proponent of high-risk virological research--opens with "The 1977-1978 influenza epidemic was probably not a natural event".
> and it did not cause a pandemic.
Can you explain why you don't think it was a pandemic? Even if you're (bizarrely) unwilling to admit that anyone died, pandemics are defined in terms of cases or infections, not deaths. There's no question that the 1977 flu spread unusually quickly to infect a large fraction of the world. It's widely referred to as a pandemic. What do you know that all these professors don't?
> Have you looked for the Udacity course I recommended, on probabilities? It is really good and it will help you a lot.
It sounds like a good course and I might take a look. My own professors did give me passing grades, though? I don't think introductory statistics is the major point of disagreement here.
is incorrect, since it falsely implies the 700k was just from one season. The 1977-78 season did show excess mortality in some estimates, though it's not clear whether that was from the H1N1 or the co-circulating H3N2 (which was probably a minority of the infections, but a majority of the infections in older people at highest risk of death, due to their prior immunity from 1957 and earlier). Mortality in the <19 age group reached its all-time high, around 2.4x average; that's a tiny share of the total (just 234 deaths), but perhaps an interesting gauge of the H1N1 spread.
Unless I'm missing something, I do believe the 700k has travelled remarkably far with remarkably weak provenance, and I appreciate the push to investigate that. Perhaps there's just no one with any incentive to correct it--for authors opposing high-risk virological research it's impressively large enough, while authors supporting such research don't wish to draw attention to this mortality at all. I don't think the exact number changes much practically, but I'd still rather not perpetuate that carelessness.
https://muse.jhu.edu/article/936217
No prior pandemic is known to have arisen from a genetically engineered virus, but the technology for that has only recently come to exist. No one had ever died in a plane crash before the Wright brothers, but the risk was obviously still there.