Here is what we’ve gotta do.
I want every strategy we’ve got on Near Earth Object Collision, OK?
Any ideas, any programs, anything you’ve sketched on a pizza box or a cocktail napkin…
– Armageddon (1998 film, when NASA realizes that there are 18 days left before the asteroid hits the Earth)
This Whole Thing
On January 20th someone shared, in a facebook group that I’m a part of, four facts about an emerging viral infection in China: (1) high death rate, (2) high contagion rate, (3) long incubation periods, and (4) the fact that it appeared uncontained. Despite the (at the time) relatively low number of cases, those four facts did not seem to paint a pretty picture of what was about to happen.
This was immediately alarming to a lot of people in my circles, and for good reason. Matthew Barnett, Justin Shovelain, Dony Christie, and Louis Francini sounded alarms as early as mid-January, and the rest of the EA and rationalist cluster followed suit. It makes sense people in this cluster would be concerned early on, as many of them have looked at global catastrophic risk scenarios for years, and were already well aware that the world was unequipped to deal with an infectious disease with all of the above four properties. Pandemic preparedness programs have so far relied on luck. For instance, in his 2015 TED talk “The next outbreak? We’re not ready” Bill Gates uses as an example the 2013 Ebola outbreak: “The problem wasn’t that there was a system that didn’t work well enough. The problem was that we didn’t have a system at all.” Accordingly, that particular outbreak didn’t become a disaster because of sheer luck: the disease only becomes contagious when you are already very sick and it didn’t hit a major urban area, so containing it was possible. But this time around we don’t seem to have the same luck.
Since, I’ve seen many thought leaders I respect succumb to focusing on this topic: Robin Hanson, Eliezer Yudkowsky, Paul Graham, Tyler Cohen, Sarah Constantine, Scott Alexander, Scott Aaronson, Joscha Bach, Ryan Carey, William Eden, Robert Wilbin, etc. etc. Not to mention the way these people are publicly responding to each other and building a parallel narrative on a higher level of complexity than most everybody else****. These and many other well respected intellectuals have been going on and on about the situation for over a month now. An exponentially growing curve in its early stages may not be alarming to most people, but it certainly was to people like this (Ps. 3Blue1Brown, Kurzgezagt, and Mark Rober also recently joined the conversation).
This all adds up to a vibe of countdown to Armageddon: “X days until hospitals are overwhelmed, Y days until a million people die, Z days until a vaccine will be found”. In line with this perceived, if not frighteningly real, urgency, we’ve seen countless facebook groups, subreddits, and forums scouting for novel ideas and projects to help above and beyond what the governments of the world are already doing (e.g. Covid19RiskApp, Give Directly Response, Covid Accelerator [of technology to decelerate the spread [possibly a terrible or brilliant branding]], List of Predictors, and Corona Variolation).
I personally gave a lot of thought to pandemics several years ago (in college I was on the fence between working on pandemic prevention and consciousness research as a career), so my immediate thought when learning about the virus and its properties was “we are screwed, this can’t be contained with how the world is currently set up”. While containment might have been possible at the very beginning with some luck, it very quickly becomes unmanageable. That said, I’d like to explore here ways in which the world could be realistically modified in order to contain, mitigate, and ultimately reverse the spread of novel contagious diseases including this one. After all, the WHO director general said on March 9th: “The rule of the game is: never give up.” So, well, let’s give it some more thought. I hence offer my ‘sketches on a cocktail napkin’ type of ideas in case they find any application:
Let us start by breaking down “social networks” into (1) contact networks, and (2) information networks:
- Contact networks are weighted undirected graphs where each node is a person and each edge encodes the frequency and intensity of the contact between the people it connects.*
- Information networks are weighted directed graphs that encode the amount of information transmission that there is between pairs of people. To a large extent, contact networks are subsets of information networks.**
Contact networks are what matters for modeling infectious disease transmission. Despite the constitutionally granted freedom of assembly, one can posit that if the risks to the public are high enough, it is justified to place some constraints on the nature and properties of contact networks. In a free society that truly grasps the danger of pandemics and is determined to squash them at the very beginning, contact networks might require some degree of top-down control. Perhaps, if we are serious about future pandemic prevention, we could re-conceptualize freedom of assembly as pertaining to information, rather than contact, networks.***
So in what ways could a contact network be pandemic-safe? As an intuition pump for what I’ll be discussing further below, I’d like you to consider what it might be like to live in the original “Halo” Ringworld (and Ringworld too). Assume that unrestricted travel in Halo is limited to land roaming with a maximum speed, and that in order to use a spacecraft or tube across an arc of the circle, you need to be thoroughly tested and quarantined in-between. With these constraints, we would naturally infer that the structure of the contact network of the people in this world would be embedded in the ring itself. Meaning that if an infectious disease originates somewhere on the Ringworld, containing its spread would be as easy as blocking movement on two small fronts around the epicenter of the outbreak. This even allows you to control and ultimately fully suppress diseases with long incubation periods. It is a matter of estimating how long the incubation period is, and quarantining the entire region of “furthest possible transmissibility”.
More so, given the overall circular geometry of the world, after a brief period of quadratic growth of the epidemic (as concentric circles expand around the epicenter) one would expect to see a threshold after which there is merely linear growth in the number of cases as a function of time!
Network Geometry as a Containment Strategy
To a first approximation, the single most important problem to overcome for containment is the exponential growth of the early stages of an outbreak. Of course in some cases an exponential growth is not itself the problem: and R0 = 1.001 leads to exponential growth, but it is still so slow that it can be easily dealt with. Likewise, a sub-exponential growth can still be unruly, as in a polynomial growth with an exponent of 20. But to a first approximation, I would argue that if you can get rid of exponential growth you can manage an outbreak. The example above of a Ringworld shows that exponential growth in contact networks can be slowed all the way down to linear growth at relatively early stages. Similarly, “thin” toroidal planets would also enable easy containment of outbreaks (Anders Sandberg‘s amazing work on the physics of toroidal planets finally pays off! It remains to be seen when his work on stacking high-dimensional polytopes finds real-world applications).
But we don’t have to go all the way to high sci-fi scenarios to encounter sub-exponential growth of infections in human contact networks. You see, the black death happened at a time when the contact network of humanity had a quasi-quadratic structure at the largest of scales. Villages almost certainly had a scale-free structure (e.g. the priest touching everyone once a week and the lone serf perhaps only interacting with two people a week), but once you look at the structure at scales above the village, you would find routes between neighboring villages weaving a planar graph with a 2D Euclidean geometry. The trade routes, though, provided an exception, and in the end they turned out to be key for the spread of the plague. That said, in the absence of cars, trains, or airplanes, the maximum speed of transmission was seriously limited. Historians can tell when different parts of Europe got the plague because it really took a long time to spread; we are talking about years rather than weeks.
So imagine having a contact network structure characteristic of the medieval times, but with an information network structure akin to the ones we currently have. Then controlling the black plague would be a piece of cake! You would simply need to close central trade routes, track down which villages are already infected, and put a perimeter around them.
Ok, so how do we generalize this idea to the modern times in a realistic way? I think we should perhaps think outside the box here. Remember, the core intention here is to make the spread of an infectious disease not behave in an exponential way at the beginning so that we can “segment out” the part of the network affected (i.e. quarantine) because the “surface area” of the region is not very large. Now, most analysis of disease spread on networks focus on analyzing how realistic-like network features affect disease spread. For example, clustering coefficients, the steepness of the slope of power law networks, the distribution of in-betweenness centrality of the nodes, and so on.
In a perhaps high-modernist style approach to network engineering, one can ask how the spread of a disease would change depending on alterations we could make to the network. The simplest real world case is the reasoning behind adding travel restrictions, which aim to block the spread between very large clusters (i.e. countries) and the closing of schools, universities, and large gatherings, which decrease the interconnectivity of each region of the network. A slightly more sophisticated version of this approach would be to come up with a “Pandemic Klout Score” for each person based on the their “network influence” and pay them to quarantine early on during an outbreak.
I actually worked at Klout as an intern in 2010, and my contributions were mostly on the (unfortunately slightly evil because it’s marketing) following problem: “How do you maximize the spread of a commercial campaign by giving free products to people?” Klout had what they called “perks” which was how they made money. They had contracts with other companies to give free products to “influencers” so that they talked about the perks on their social media accounts. To maximize the spread of a commercial campaign meant to distribute perks in such a way that the largest number of people made mentions of the campaign on their networks (including people who didn’t receive the free products). This is how they measured success- at least when I was there- and what the companies paid them for.
The “basic approach” would be simply to distribute the perks to people with the highest Klout scores, with the additional constraint that those people were influential on the relevant topic (e.g. if you had a popular Twitter account about “beauty and personal care” you might be a prime candidate to get a free “anti-aging sunscreen stick”, or whatever) . But since you can’t actually, you know, entice Justin Bieber (the person with the highest Klout score for several years) with a free Virgin America flight and expect him to either care or talk about it on his Twitter feed, the problem ends up being substantially more complex than just “give people with high Klout the free products”. I am under an NDA about the specific algorithms and research I conducted there. But I mention this because the problem of pandemic prevention could in some sense be thought of as the inverse of the problem Klout was trying to solve. Namely, how you use the node features of the network in order to minimize the spread of a contagious disease. The low-hanging fruit idea here can be to simply allot money to pay people with high Pandemic Klout Scores to stay home or cut their human touch in half whenever an outbreak arises. I would expect this to be significantly better at reducing the reproductive rate of a contagious disease than choosing people at random (or even just based on how many people they interact with on a daily basis).
That said, given the risks and costs involved with pandemics, especially in the long term in light of bioterrorism, we should not close off the possibility of making drastic changes to humanity’s contact network for the sake of our collective wellbeing. That is, merely asking some people to stay home may not be enough. We should contemplate what it would really take to be able to fully contain any future pandemic.
In terms of large-scale network geometry rather than just dealing with one node at a time, perhaps the key point to make is that we should really not fetishize and romanticize the “six degrees of separation” that results from the small world-like structure of the modern human contact network. Yes, “it’s a small world after all“, but you forgot to mention “and that’s what will get us all killed in the end.” Let’s not allow misguided network idealism to murder grandma. We need to make the contact network a large world, and save the small world exclusively for the information network.
Intuitively, it is precisely the small world-like property of our contact network that allows us to: meet many new people on a regular basis, collaborate with people around the world, be able to attend large gatherings, raves, and festivals, and travel care-free across the planet. Meaning, most people might think that changing the contact network structure to make it pandemic-proof would come at the cost of sacrificing what makes society so interesting and worth living in. I would disagree. I think that such a line of thinking is just the result of a failure of the imagination. We can, I posit, have contact networks that allow you to do all of that and yet be pandemic-proof. I will argue that with intelligent top-down network engineering you can in fact achieve this. Here is my case:
The main concept that one needs to understand for my argument is that the options for large-scale network structure go far beyond the textbook examples of small worlds, scale-free, random, planar graphs, etc. In fact, one can create all kinds of fascinating hybrid networks where the properties vary by region and scale. The examples I am about to show you play with the notion of scale-dependent geometry. Meaning that the network properties depend on the number of interconnected nodes that you are considering. In particular, I’ll break down networks in terms of their micro (1 to 1,000 nodes), meso (1,000 to 1,000,000 nodes), and macro (1,000,000 to 1,000,000,000 or more nodes) structure:
The first example is one where the structure of the network leads to quadratic spread on the micro level, linear spread on the meso level, and exponential spread on the macro level. We achieve this by having the nodes arranged along a rectangular grid at the micro level. As one zooms out, the grid hits a limit on two fronts so that the advancement of an infection disease will start growing linearly as it only has two directions to grow in (for the sake of symmetry you can glue the two fronts to make a tube, for a meso network structure akin to that of a toroidal planet). Finally, at the largest scale this network looks like a binary tree, where the growth can reach an exponential rate.
The same scheme will apply to all of the following networks. That is, the letters indicate the ordering of the types of growth for the micro, meso, and macro scale. What I will instead focus on is explaining the advantages of these structures. In this case- the case of QLE- the primary advantage is that the spread can be entirely contained by cutting connections around the epicenter. And the best part is that even if you hit the exponential scale (i.e. you start spreading from “one arm to another”) you will still have long periods of linear growth as each “arm” will grow linearly, so cutting it will remain an option at any point. The “surface area of the spread” will remain tiny relative to the size of the network.
A very nice property of this network is that you can have “villages” of up to 1,000 people where everyone can interact with and touch each other. Within each of these villages you have super efficient in-person information transmission and contact hedonism without restrictions. Then each of these villages would be connected to two neighboring villages, perhaps not unlike how kids in grade school often make friends with other kids in the grades immediately above and below (and only rarely with grades that are further apart). The spread of disease would very quickly engulf each village, but thankfully that would be it. After that you would have a very slow village-by-village take-over that could be stopped by ‘cutting’ the contact channels between two pairs of villages (or four if you started at an intersection of the macro structural grid). More so, you could conceive of a “conveyor belt” approach where every month half of the village moves in one direction while the other one stays put. This way over the course of years you would still be able to get to know tens of thousands of people, party like crazy in raves touching everybody, and be able to retain long-term friendships by coordinating with them to either move or stay. And you could do all of this while living in a pandemic-proof world!
This one is perhaps the least viable because it relies on most persons only having contact with two people. That said, the spread would start very, very slowly, and so it might be ideal for the worst possible pandemics. At the macro level the network looks like high-dimensional cardboard boxes, where each “cardboard side” is glued at the edge with one or multiple other sides.
A “continuous” version of LQE could use hyperbolic geometry at the macro level, such as what you get when you sneak a pentagon here and there in an otherwise rectangular grid so that locally you have a square spread, which slowly turns into an exponential spread as you begin swallowing pentagons. (Or a few heptagons in a grid of hexagons).
This one is pretty similar to ELQ, and you can do pretty much the same things I mentioned about ELQ. The main difference is that this structure is safer at the macro level but riskier at the meso level. So if you expect diseases to be really really contagious, then this structure might prevent “the end of the world” but it might be somewhat susceptible to “pretty bad scenarios”, while ELQ works the other way around.
I find this network very interesting because to build it I had to come up with the idea of connecting lots of cycles of different lengths with each other by having them share nodes. You can also easily construct a network like this by starting with a scale-free network and replacing the edges with long chains of nodes.
This would perhaps be the steel-manned version of the toroidal or ring world we discussed in the introduction. Here the infections would spread first slowly at a quadratic rate, then quickly accelerate once you reach the edges of the planar graph you start in, and finally there is a massive linear bottleneck at the macro scale. It’s like Ringworld, but where you interact with people in an interlaced braided mesh embedded inside the Ringworld rather than only in its meager inner surface.
Because each of these examples contain a “linear bottleneck” at some scale, an outbreak of a disease would be easy to contain at some scale. Which network is ideal for which kind of disease will depend on things like its incubation period and its contagion probability. But any of these examples is vastly safer pandemic-wise than our current contact network.
Is Biology Doing This Already?
One thing this exercise has made me wonder is if perhaps our bodies are already using this kind of strategy. I mean, looking at QLE reminds me of the structure of blood vessels in the kidney and liver. It would make sense that evolution would identify great micro, meso, and macro network structures in order to give each organ appropriate contact networks at the scale that matters to conduct its function, while creating network bottlenecks at other scales for protection against pathogens and the spread of cancer. In contrast, the immune system would have every reason to maximize spread at the largest scale while having compartmentalized spread at the micro scale (example: Topological Small-World Organization of the Fibroblastic Reticular Cell Network Determines Lymph Node Functionality). Finding the sub-exponential chokepoints in the human body would, I posit, give us a new angle for understanding it more deeply.
Creating a Global Human Organism
If this analysis pans out, we could perhaps think of the challenge being presented to us by SARS-CoV-2 and future pandemics as a wake up call to “scale up the network-protective measures our bodies are taking to combat disease while maintaining functionality” all the way up to the structure of all of human society. Indeed, wouldn’t it be amazing if we coordinated to be a harmonious large-scale global organism?
Now, I am not saying we should simply adopt one of these network structures. They are just proofs of concept to show it is possible to have humanly-desirable properties that come with highly interconnected networks along with a linear (or at least sub-exponential) bottleneck at some scale. The bottleneck does not even need to be visible or detectable from the point of view of each individual!
Even if we cannot construct an ideal world from scratch, we could still try to bootstrap it from within our current world. To do so we have a number of options. I will mention two and then dive into them in greater depth. The first is the strategy of “network modification” and it consists of developing gradient descent algorithms that point us to the modification of the network that would maximize a scale-specific sub-exponential bottleneck. Of course this could lead to local minima, but we don’t care about achieving the best configuration, just the closest one that is “good enough”. The second approach is that of “network nucleation” to bootstrap a pandemic-protected contact network by connecting with other people who can prove that they do not have the disease. They could all get to know each other, and then submit a list of “people they would like to hang out with on a regular basis”. An algorithm would then optimize the network so that each person can hang out with as many others as possible while making sure the overall geometry of the network is desirable for disease contention. If lucky, we could even bootstrap this system all the way up to the entire planet, starting from a mixture of people who’ve demonstrably been quarantined for a long time and people who have already recovered from the disease. And since, of course, people would eventually get sick of hanging out with a restricted list of friends, they could periodically re-submit another list and the algorithm would take into account this dynamic so that the geometry can be stable over time.
My prediction is that the current strategies that are being used to reduce the spread of disease would show up as a tiny subset of the set of possible effective strategies, many of which are currently invisible- and in some sense inconceivable- to us. This is because, in part (as far as I know) nobody is thinking in terms of scale-specific network geometry. Also, little is known about the actual empirical structure of the human contact network. In this sense, removing super-spreaders or closing schools may be re-conceptualized as pointing in this direction, and yet perhaps may not even make the Top 10 list of best cost-effective strategies. This is because just removing high-degree nodes in a scale-free network won’t automatically prevent exponential growth; since exponential growth is the killer, making strategies directly targeted at it will probably be vastly more effective. Let’s investigate these strategies in more detail:
Option 1: Network Modifications
The first thing we should do is find what actual contact networks look like, so that we can identify the smallest possible modifications to them in order to create sub-exponential bottlenecks on some scale. I have not found a good study on this, since there really aren’t public datasets of “who is physically hanging out with whom”. Though, if you were to combine, perhaps, the datasets of USA’s NSA, UK’s GCHQ, Russia’s KGB, China’s MSS, cellphone location information, census responses, and commercial surveillance camera data you might be able to get a very decent version of it. In fact, there is reason to believe Israel is already in the process of constructing this dataset.
In the absence of contact network data, we can nonetheless learn from other social and information networks. In particular, the best research I’ve read about the macro-structure of complex networks comes from the lab of Jure Leskovec (I recommend watching his CS224W lectures from past years, which are all available online):
We study over 100 large real-world social and information networks. Our results suggest a significantly more refined picture of community structure in large networks than has been appreciated previously. In particular, we observe tight communities that are barely connected to the rest of the network at very small size scales; and communities of larger size scales gradually “blend into” the expander-like core of the network and thus become less “community-like.” This behavior is not explained, even at a qualitative level, by any of the commonly-used network generation models.
– Lescovec et al. 2008, “Community Structure in Large Networks: Natural Cluster Sizes and the Absence of Large Well-Defined Clusters“
As you see, large-scale analysis of real-world networks indicate that they are not adequately described by the classic textbook structures that are most well known. Rather, there seems to be a kind of “galactic shape” at the more macro scale, where there is a highly connected giant core of overlapping communities surrounded by loosely connected superstructures (nicknamed ‘whiskers’):
Given this structure (and assuming it generalizes to contact networks), one could divide the problem into two rough components: (1) how to you deal with ‘whiskers’?, and (2) what do you do about the ‘galactic core’? I do not have answers here, but I do think that having more people who are good at math and computer science think about this would be very good. For what is worth, I have the hunch that in particular the following two network analysis techniques will be useful to tackle this problem:
- Spectral Graph Theory: This is a set of techniques that can help us ‘see diffusion bottlenecks in graphs’ at a glance. For instance, these techniques reveal the presence of network “chokepoints” that create insulation in heat flow. Clearly heat flow does not behave in the same way as the spread of disease, but the similarity makes it worth highlighting.
- Discrete Differential Geometry: An emerging field that blends differential geometry with network analysis and has shown amazing applications for graphics which can help us ‘see the curvature and dimensionality of a network around each of its nodes’ at a glance. Note: As much as I love hyperbolic spaces, I must admit that from the point of view of early pandemic prevention living in a contact network with hyperbolic geometry is a terrible idea.
Flatten the Network!
One additional interesting approach for Option 1 would be to apply topological clustering techniques to the contact network so that we can identify the hubs with the least desirable network geometry and try to “flatten them”. And policy-wise, I might imagine that in the long-run we could improve the flattening of the contact network by encouraging people to use things like the Bumble app for dating, where you find people physically near you with whom you could form a healthy relationship.
Option 2: Network Nucleation
Green and Red Countries
Joscha Bach predicts that in a couple months there will be “green and red” countries, meaning that the outbreak will be completely under control in some countries, and completely out of control in others. I’d also add “grey” to refer to “unreliable statistics”, as many countries might just choose to not monitor the situation. You can imagine what the travel restrictions may be between green, red, and grey countries, as green countries would not find it worthwhile (or at least not politically viable) to accept the risk of reigniting the spread. Grey countries may end up also avoiding red countries while not being allowed to enter green countries.
Speculatively, this would perhaps lead to a worldwide Sakoku phenomenon, but where rather than just Japan, we would have all of the countries of each color becoming economic and cultural blocks.
What I’ll describe below is a kind of generalization of this possibility. Namely, that the blocks don’t need to be country-based.
A very interesting question to ask is “what possible partitions of humanity could create sets of people for whom a green/red/grey dynamic would successfully create clusters of wholly virus-free people?” The existence of at least some greens opens up the possibility of:
Reversing The Pandemic
I address you tonight, not as the president of the United States, not as the leader of a country, but as a citizen of humanity. We are faced with the very gravest of challenges. The Bible calls this day Armageddon. The end of all things. And yes, for the first time in the history of the planet, a species has the technology to prevent its own extinction. All of you praying with us need to know that everything that can be done to prevent this disaster is being called into service. The human thirst for excellence, knowledge, every step up the ladder of science, every adventurous reach into space, all of our combined technologies and imaginations, even the wars that we’ve fought, have provided us the tools to wage this terrible battle. Through all the chaos that is our history, through all of the wrongs and the discord, through all of the pain and suffering, through all of our times, there is one thing that has nourished our souls and elevated our species above its origins, and that is our courage. Dreams of an entire planet are focused tonight on those 14 brave souls traveling into the heavens. May we all citizens of the world over see these events through, Godspeed, and good luck to you.
– Armageddon (1998 film, when the president of the US announces the plans to avert an asteroid that would destroy the earth) [See also: what if they don’t come back?]
Nucleating Whole Virus-Free Communities
The simplest way to create a virus-free community would be to think of verifiable self- quarantining as an investment. If you can prove you’ve been physically disconnected from everyone for 30 days, you would be let into a club for people near you who have done the same already. This could become a large set of people, especially if it turns out that cash handouts are insufficient for millions of people who might end up needing to work in a month or two and defy any kind of large-scale quarantine. Those who can afford (and prove!) that they’ve been diligently quarantining would be allowed in. For a stricter “inner set” there might be stricter criteria where you would need to submit an unfakeable biosample to prove you are not infected (which would be tricky but not impossible given pre-existing DNA databases like 23andMe). Then the algorithm would group you with a subset that you can realistically physically meet, and then allow you to make friends with them. Finally, as you submit a list of people you do want to hang out with long-term, the algorithm would run an optimization process to make as many of the people happy and return the curated list of people you could hang out with so that the network as a whole has convenient scale-dependent sub-exponential chokepoints. I know this sounds like a lot. And it is. But again, pandemics can be really bad. And we have the technology, so why not try?
In a way this idea is the complementary problem to “keeping the virus out of the general population”. In the latter you start out in a fully virus-free situation and try to keep it that way, while the former starts out in a highly contaminated population and tries to “spread health” from the standpoint of a verifiably healthy core. That is, how you create pockets of health in a virus-saturated general population and grow them as much as possible.
Another approach in this vein I can think of is to seed a location with an excess of people who already have immunity and cannot transmit. The people there who haven’t gotten the disease would in a sense be lucky to find themselves around people who won’t transmit it, and thus be blessed with spontaneous herd immunity. That said, the key sacrifice here would be the potential damage elsewhere, where herd immunity would be reached later due to the removed group of immune people. This and the previous approach incur the cost of having to associate with new people, and the relocation challenges would be a logistical nightmare. But perhaps worth doing.
Finally, another approach to this problem would be to use an app with a personality test that is hard to fake, so that only healthy people who score in the top 2% of both introversion and conscientiousness could join the club. It would tell you where to go live with other people who meet the same criteria, and to get a comprehensive test of all major transmissible diseases and treat those you have before relocation. Given the temperament selected for, everyone who becomes part of the community would be extremely diligent about not physically meeting people outside the group and follow the contact network prescriptions dictated by the algorithm. If this sounds like hell to you, well, perhaps it is not for you. But at least this way there would be some pockets of fully healthy people, and that would have a lot of value. (Cf. Rat-free Alberta).
What are your options for modifying a network in order to remove (or at least tame) exponential growth? The one’s I’ve considered are:
- Remove nodes with a high “Pandemic Klout Score”
- Creating sub-exponential chokepoints:
- Option 1: Gradient descent methods:
- You make piece-meal modifications to the contact network one connection at a time in order to improve the prospects of the entire network.
- Each person would receive a set of options for mild modifications to their contacts so that whichever they chose would lead to an improvement of the network geometry.
- Option 2: Network nucleation:
- You create a criteria for what constitutes “infection-free” such as:
- Self-enforced quarantine on one extreme, and
- Provable DNA-matched tests on the other extreme.
- Allow people who qualify to meet each other.
- Everyone submits a list of people they’d like to hang out with.
- The algorithm would optimize the connections to make everyone happy and at the same time maximize the sub-exponential chokepoints of the network (such as by making it a planar graph with a high clustering coefficient, etc.).
- You create a criteria for what constitutes “infection-free” such as:
- Option 1: Gradient descent methods:
Now, perhaps if all of this sounds insane and like too much trouble, there is always the option of, er, becoming comfortable with no human touch…
A Religion of Abstinence of Human Touch
I know how hard it is, what is being demanded of us.
Especially in times of needs such as these, we like to be close to one another.
We understand care and affection in terms of human closeness and human touch.
But at the moment the exact opposite is the case, and everybody really must understand that.
At the moment, the only real way of showing you care is keeping your distance.
– German Chancellor Angela Merkel, at a Nationwide TV Address (March 18 2020)
Have you ever noticed that it is possible to reproduce without any human touch? Artificial insemination conducted with robotic arms is not a far-fetched prospect. A further question is: can we do away with human touch entirely for all functions of life?
You don’t need to be anywhere to be everywhere.
– John C. Lilly
You may say: wouldn’t a community of touch-free individuals somehow lack the most basic of human qualities, i.e. interpersonal intimacy? I reckon that you would be wrong on more than one account. First of all, insofar as touch-based intimacy is based on endorphin and oxytocin release in conjunction with nervous system entrainment under the hood, there is no reason why one couldn’t engineer a brain-stimulation technology ecosystem so that people receive the same kind of physically, psychologically, and spiritually rewarding feelings of connection by merely acknowledging each other’s presence or synchronizing with each other’s brainwaves. Perhaps even you could achieve this despite doing away with technology, as the power of deep metta meditation would suggest. Perhaps we could all cultivate a loving temperament that embraces all of the universe of sentient beings. Here, the commitment to each other’s physical wellbeing is possible without sacrificing the emotional richness of communion; in principle they could be simultaneously satisfied. Alas, the evolutionary roots of human touch are deep, and trying to mess with them with humans as they currently are is far fetched. But just wait until a virus with 0.98 fatality rate and R0 = 6 is discovered and see what people are willing to do to survive.
This concludes my presentation of the cocktail napkin ideas I’ve considered so far to deal with pandemics. But I still have a couple more things to say about this topic, so I’ll take advantage of the soap box I’m standing on and add:
Now That The World Is Paying Attention
I’d like to draw your attention to the following highly relevant goals that the current crisis highlights:
1) We ought to recognize the existence of extreme suffering so that we focus our efforts on its prevention (asphyxiation is an example of extreme suffering, which is how people are dying of COVID-19).
2) Investigating what makes MDMA and 5-MeO-DMT so special and useful for treating PTSD (as people recover from the disease it will become apparent many experience PTSD associated with the episode – this will need to be addressed on a massive scale).
3) Get factory farms banned (for real, they are the breeding grounds of future pandemics – and they of course also cause the bulk of easily preventable suffering, so there is that too. Every animal product you put on your plate is a probabilistic pandemic on its way. Sorry!).
A Few Final Thoughts
The Framing Effect
Recall the “Framing Effect” – the cognitive bias where we prefer an option when the problem is framed in a certain way, and a different option when it’s framed differently even though the corresponding options in each framing are of equal expected value.
I worry a lot of the people in my friend network, and in fact worldwide, might be falling prey to the framing effect for the coronavirus situation:
Here is how the “containment vs. mitigation” problem is being “framed” right now (assume 5 million people will die worldwide if nothing is done, but you can choose to invest your resources on ‘containment’ or ‘mitigation’):
Option A: 10% chance 0 people die (i.e. successful containment), and 90% chance 5 million people die.
Option B: 100% chance 4 million people die.
Clearly option A is more ‘heroic’. Alas, it is the one that leads to more expected deaths.
Now consider the alternate framing that might make you feel differently about the options:
Option A: 10% chance of saving 5 million people (i.e. successful containment) and 90% of saving nobody.
Option B: 100% chance of saving 1 million people (i.e. mitigation prevents many deaths).
In both cases option B is much better by a huge margin. In fact by an expected number of 500,000 people saved. Yet when framed in the first way option A seems a lot more attractive. Why? And should we try to get rid of this bias?
Of course in the real world you don’t have to choose between A and B entirely. You can try to do both containment and mitigation. But you *do* need to choose how to allocate resources, and I believe this framing issue does actually come up in our current situation.
I do want to say that, as Robin Hanson suggests, if we are doing the containment strategy we need buy-in from the population. Some personally costly and dramatic public display of commitment from many people would be useful. I am personally very happy to commit in public to hard-core quarantine if it’s ethically necessary.
Social Withdrawal and Behavioral Enrichment
Social distancing is painful because we are all opioid addicts, namely, addicts to the endogenous opioids released when socializing. With a quarantine in place, we can anticipate that people who are on the threshold of being depressed might cross that threshold as an effect of reduced in-person socializing. Likewise, we can anticipate collective health decline at a statistical level due to reduced exercise, sunlight exposure, and sensory diversity (cf. white torture).*****
Possible solutions? Besides being very bullish on at-home exercise routines and HEPA filters, I would also point out the following. I think that we should not be afraid of comparing ourselves with other animals. Bear with me. Humans, not unlike domestic dogs and cats, benefit from being exposed to a wide variety of novel sensory inputs. If you enjoy scents, for example, it would be advisable to order a set of essential oils or perfume samples in order to trick your brain into thinking you are exploring a larger area than you are. Apparently, for example, big cats in captivity are more engaged and less depressed when you spray Calvin Klein perfumes on their territory. Alternatively, if scent is not something you care about, think of perhaps increasing the repertoire of visual art, dance, food, touch, and music you are exposed to on a daily basis. This, I suggest, will help you keep depression away (for a while longer).
Caption: Just a little bit of behavioral enrichment for you! 🙂
Finally (self-promotion ahead), if you have time on your hands, and you’ve been meaning to dive deeper into Qualia Computing, this might be your chance. I’d suggest you start out with the following three resources:
- Top 10 Qualia Computing Articles
- Glossary of Qualia Research Institute Terms
- Every Qualia Computing Article Ever
And if you are really hard core, feel free to reach out to the Qualia Research Institute to help with volunteer work. Also we are going to be doing virtual internship cycles in April, May, and June, so you can stay home and safe and still collaborate with us. But shh! It’s a secret! (Wait, how come it’s a secret but you now know about it? Well, because you’ve scrolled all the way here, that’s some commitment!).
* A more accurate representation might require the use of directed edges to encode asymmetrical contact relationships. For example: the cleaning crew of a hotel might be more exposed to the guests than the guests are exposed to the crew. Also, when two people who have very different habits of hygiene meet, the cleaner person is more likely to get the short end of the stick transmission-wise.
** It is worth pointing out for information networks the “degree of interaction” between nodes is extremely skewed. You may have a thousand friends on Facebook, but the number of people you are likely to interact on a daily basis will be a tiny subset of them, perhaps on the order of 0 to 20. And among the people you interact with, you are likely interacting much more word-count-wise with some than with the others. Indeed, if you plot the number of words exchanged in private messages between people in an information network, the distribution follows a long-tail.
*** In the long-run, this may also have to apply to information networks. Whether information networks will need also some level of top-down control will be a difficult question to answer that requires a complex cost-benefit analysis beyond the scope of this article. The most important variables being (a) what the benefits of fully-free communication are, and (b) the density and severity of memetic hazards in idea-space, in conjunction with the nature of intellectual selection pressures in future societies. If it turns out that people above a certain level of education and intelligence in a future with far more advanced science and engineering are extremely likely to encounter what Nick Bostrom calls “black balls”, there might be no way around developing tight controls on information networks for the safety of everyone. It this happens, we could also use many of the strategies outlined in this article for contact networks. After all, viruses are related to contact networks in the same way as meme hazards are related to information networks.
**** Of course, in some ways this is more about collective emotional processing than about object-level problem solving.
***** It is worth noting that the better air quality might buffer a bit against these negatives.
The network thoughts are interesting, but despite the theoretical arguments that they need not be incompatible with free collaboration and communication, in practice they would have to be. Enforcement would need incredibly intrusive surveillance and control capacities, and there is no chance that, once those capacities existed, the state(s) or other actors which had them would limit their use to their stated purpose of disease impedance.
Also, factory farming did not have anything to do with this disease and while it does build up antibiotic resistance, the last several serious pandemics and threatened pandemics have all been viruses, not bacteria, so there is not actually any basis for your claim that they are incubating the next pandemic threat.