I am the co founder of and researcher at the quantitative long term strategy organization Convergence (see here for our growing list of publications). Over the last decade I have worked with MIRI, CFAR, EA Global, and Founders Fund, and done work in EA strategy, fundraising, networking, teaching, cognitive enhancement, and AI safety research. I have a MS degree in computer science and BS degrees in computer science, mathematics, and physics.
JustinShovelain
Karma: 244
Moloch and the Pareto optimal frontier
Thanks for writing the post! I think we need a lot more strategy research, cause prioritization being one of the most important types, and that is why we founded Convergence Analysis (theory of change and strategy, our site, and our publications). Within our focus of x-risk reduction we do cause prioritization, describe how to do strategy research, and have been working to fill the EA information hazard policy gap. We are mostly focused on strategy research as a whole which lays the groundwork for cause prioritization. Here are some of our articles:
Heuristics for cause prioritization and assessing interventions
Components of strategy research (one of which is cause prioritization and it is dependent upon the others)
We’re small and relatively new group and we’d like to see more people and groups do this type of research and that this field get more support and grow. There is a vast amount to do and immense opportunity in doing good with this type of research.
The bet is on.
Following Sean here I’ll also describe my motivation for taking the bet.
After Sean suggested the bet, I felt as if I had to take him up on it for group epistemic benefit; my hand was forced. Firstly, I wanted to get people to take the nCOV seriously and to think thoroughly about it (for the present case and for modelling possible future pandemics) - from an inside view model perspective the numbers I was getting are quite worrisome. I felt that if I didn’t take him up on the bet people wouldn’t take the issue as seriously, nor take explicitly modeling things themselves as seriously either. I was trying to socially counter what sometimes feels like a learned helplessness people have with respect to analyzing things or solving problems. Also, the EA community is especially clear thinking and I think a place like the EA forum is a good medium for problem solving around things like nCOV.
Secondly, I generally think that holding people in some sense accountable for their belief statements is a good thing (up to some caveats); it improves the collective epistemic process. In general I prefer exchanging detailed models in discussion rather than vague intuitions mediated by a bet but exchanging intuitions is useful. I also generally would rather make bets about things that are less grim and wouldn’t have suggested this bet myself, but I do think that it is important that we do make predictions about things that matter and some of those things are rather grim. In grim bets though we should definitely pay attention to how something might appear to parts of the community and make more clear what the intent and motivation behind the bet is.
Third, I wished to bring more attention and support to the issue in the hope that it causes people to take sensible personal precautions and that perhaps some of them can influence how things progress. I do not entirely know who reads this and some of them may have influence, expertise, or cleverness they can contribute.
Updating the Fermi calculation somewhat:
P(it goes world scale pandemic) = 1⁄3, no updates (the metaculus estimate reference in another comment counteracted my better first principles estimation)
P(a particular person gets it | it goes world scale pandemic) = 1⁄2, updating based on the reproduction number of the virus
P(a particular person dies from it | a particular person gets it) = 0.09, updating based on a guess of 1⁄2 probability rare equipment is needed and a random guess of 1⁄2 probability fatality without it. 1/2*1/30 + 1/2*((Probability of pneumonia: (1/3+1/4 )*1/2)*(Probability of fatality given pnemonia and rare equipment is needed: 1⁄2)
=> P(death of a randomly selected person from it) = ~1/67
I’m not entirely sure what to think of the numbers; I cannot deny the logic but it’s pretty grim and I hope I’m missing some critical details, my intuitions are wrong, or unknown unknowns make things more favorable.
Hopefully future updates and information resolves some of the uncertainties here and makes the numbers less grim. One large uncertainty is how the virus will evolve in time.
Update from Convergence Analysis
In July, we published the following research posts:
Improving the future by influencing actors’ benevolence, intelligence, and power: This post outlines a framework for coming up with, and assesses the expected value of, actions to improve the long-term future, and discusses nine implications of this framework. --We were excited to see this framework already drawn on in two new forum posts by other authors (1,2).
Moral circles: Degrees, dimensions, visuals: This post overviews the classic concept of moral circles, discusses two important complexities that conception overlooks or fails to make explicit, and how they can be represented visually. --This post also led to several researchers getting in contact with Michael Aird (its author) and then being connected with each other.
Crucial questions for longtermists: This introduces a collection of “crucial questions for longtermists”: important questions about the best strategies for improving the long-term future. This collection is intended to serve as an aide to thought and communication, a kind of research agenda, and a kind of structured reading list.--In August, we followed up with a post focusing specifically on Crucial questions about optimal timing of work and donations
Additionally, our Researcher/Writer Michael Aird published:
Thanks for the article. One thing I’m wondering about that has implications for the large scale pandemic case is how much equipment for “mechanical ventilation and sometimes ECMO (pumping blood through an artificial lung for oxygenation)” does society have and what are the consequences of not having access to such equipment? Would such people die? In that case the fatality rate would grow massively to something like 25 to 32%.
Whether there is enough equipment would depend upon how many get sick at once, can more than one person use the same equipment in an interleaved fashion, how long each sick person needs the equipment, are their good alternatives to the equipment, and how quickly additional equipment could be built or improvised.
So the case I’d be worried about here would be a very quick spread where you need rare expensive equipment to keep the fatality rate down where it is currently.
Safety regulators: A tool for mitigating technological risk
The ‘sprinting between oases’ strategy
Nice post!
Here are a couple additional posts that I think are worth checking out by Gwern:
https://www.lesswrong.com/posts/ktr39MFWpTqmzuKxQ/notes-on-psychopathy
https://www.lesswrong.com/posts/Ft2Cm9tWtcLNFLrMw/notes-on-the-psychology-of-power
The exponential growth curve and incubation period also have implications about “bugging out” strategies where you get food and water, isolate, and wait for it to be over. Let’s estimate again:
Assuming as in the above comment we are 1⁄3 of the exponential climb (in reported numbers) towards the total world population and it took a month, in two more months (the end of March) we would expect it to reach saturation. If the infectious incubation period is 2 weeks (and people are essentially uniformly infectious during that time) then you’d move the two month date forward by two weeks (the middle of March). Assuming you don’t want to take many risks here you might have a week buffer in front (the end of the first week of March). Finally, after symptoms arise people may be infectious for a couple weeks (I believe this is correct, anyone have better data?). So the sum total amount of time for the isolation strategy is about 5 weeks (and may start as early as the end of the first week of March or earlier depending on transportation and supply disruptions).
Governments by detecting cases early or restricting travel, and citizens by isolating and using better hygiene, could change these numbers and dates.
(note: for future biorisks that may be more severe this reasoning is also useful)
It’s based on a few facts and swirling them around in my intuition to choose a single simple number.
Long invisible contagious incubation period (seems somewhat indicated but maybe is wrong) and high degree of contagiousness (the Ro factor) implies it is hard to contain and should spread in the network (and look something like probability spreading in a Markov chain with transition probabilities roughly following transportation probabilities).
The exponential growth implies that we are only a few doublings away from world scale pandemic (also note we’re probably better at stopping things when their at small scale). In the exponential sense, 4,000 is half way between 1 and 8 million and about a third of the way to world population.
I wonder what sort of Fermi calculation we should apply to this? My quick (quite possibly wrong) numbers are:
P(it goes world scale pandemic) = 1⁄3, if I believe the exponential spreading math (hard to get my human intuition behind) and the long, symptom less, contagious incubation period
P(a particular person gets it | it goes world scale pandemic) = 1⁄3, estimating from similar events
P(a particular person dies from it | a particular person gets it) = 1⁄30, and this may be age or preexisting condition agnostic and could, speculatively, increase if vital equipment is too scarce (see other comment)
=> P(death of a randomly selected person from it) = ~1/300
What are your thoughts?
By total mortality rate do you mean total number of people eventually or do you mean percentage?
If the former I agree.
If you mean the later… I see it as a toss up between the selection effect of the more severely affected being the ones we know have it (and so decreasing the true mortality rate relative to the published numbers) and time for the disease to fully progress (and so increasing the true mortality rate relative to the published numbers).
It’s true that this is pretty abstract (as abstract as fundamental epistemology posts), but because of that I’d expect it to be a relevant perspective for most strategies one might build, whether for AI safety, global governance, poverty reduction, or climate change. It’s lacking the examples and explicit connections though that make this salient. In a future post that I’ve got queued on AI safety strategy I already have a link to this one, and in general abstract articles like this provide a nice base to build from toward specifics. I’ll definitely think about, and possibly experiment with, putting the more abstract and conceptual posts on LessWrong.
Nice list!
Adding to it a little:
Avoid being sick with two things at once or being sick with something else immediately before.
When it comes to supplements the evidence and effect sizes are not that strong. Referencing examine.com and what I generally remember, I roughly think that the best immune system strengthening supplements would be zinc and echinacea with maybe mild effects from other things like vitamin C, vitamin D, and whey protein. There may be a couple additional herbs that could do something but it’s unclear they are safe to take for a long duration. What you’d aim for is decreasing the severity of viral pnemonia induced by something like influenza.
It’s possible that some existing antivirals will be helpful but currently this is unknown.
Nice article Michael. Improvements to EA cause prioritization frameworks can be quite beneficial and I’d like to see more articles like this.
One thing I focus on when trying to make ITC more practical is ways to reduce its complexity even further. I do this by looking for which factors intuitively seem to have wider ranges in practice. Impact can vary by factors of millions or trillions, from harmful to helpful, from negative billions to positive billions. Tractability can vary by factors of millions, from negative millionths to positive digits. The Crowdedness component generally implies diminishing or increasing marginal returns only vary by factors of thousands, from negative tens to positive thousands.
In summary the ranges are intuitively roughly:
Importance (util/%progress): (-10^9, 10^9)
Tractability (%progress/$): (-10^-6, 1)
Crowdedness adjustment factor ($/$in): (-10, 10^3)
Let’s assume interventions have randomly associated with them samples from probability distributions over these ranges. Roughly speaking then we should care about these factors based on the degree to which they help us clearly see which intervention is better than another.
The extent to which these let us distinguish between the value interventions is based on our uncertainty per factor for each intervention and how the value depends on each factor. Because the value is equal to Importance*Tractability*CrowdednessAdjustmentFactor each factor is treated the same (there is abstract symmetry). Thus we only need to consider how big each factor range is in terms of our typical intervention factor uncertainty. This then tells us how useful each factor is at distinguishing interventions based on importance.
Pulling numbers out the the intuitive hat for the typical intervention uncertainty I get:
Importance (util/%progress uncertainty unit): 10
Tractability (%progress/$ uncertainty unit): 10^-6
Crowdedness adjustment factor ($/$in uncertainty unit): 1
Dividing the ranges into these units lets us measure the distinguishing power of each factor:
Importance normalized range (distinguishing units): 10^8
Tractability normalized range (distinguishing units): 10^6
Crowdedness adjustment factor normalized range (distinguishing units): 10^3
As a rule of thumb then it looks like focusing on Importance is better than Tractability is better than Crowdedness. This lends itself to a sequence of improving heuristics for comparing the value of interventions then:
Importance only
Importance and Tractability
The full ITC framework
(The above analysis is only approximately correct and will depend on details like the precise probability distribution over interventions you’re comparing and your uncertainty distributions over interventions for each factor.
The ITC framework can be further extended in several ways like: making precise curves interventions on the factors of ITC, extending the detail of the analysis of resources to other possible bottlenecks like time and people, incorporating the ideas of comparative advantage and market places, …. I hope someone does this!)
(PS I’m thinking of making this into a short post and enjoy writing collaborations so if someone is interested send me an EA forum message.)
Sure, I’ll take the modification to option (i). Thanks Sean.
How confident are you that it affects mainly older people or those with preexisting health conditions? Are the stats solid now? I vaguely recall that SARS and MERS (possibly the relevant reference class), were age agnostic.
Hmm… I will take you up on a bet at those odds and with those resolution criteria. Let’s make it 50 GBP of mine vs 250 GBP of yours. Agreed?
I hope you win the bet!
(note: I generally think it is good for the group epistemic process for people to take bets on their beliefs but am not entirely certain about that.)