Latest / 80,000 Hours Podcast / #224 – There's a cheap and low-tech way to save humanity from any engineered disease | Andrew Snyder-Beattie
Transcript
- Andrew Snyder-Beattie: This is an elastomeric respirator. I could wear this six months in a row without needing to change the filters. The shelf life is 20 years. If a government were to purchase a bunch of these, that gives you basically 20 years of protection that you’re protecting your population. The protection factor for something like this is about a factor of 100, so that means it filters out 99% of particles coming in. This right here is propylene glycol. It’s a chemical that kind of disrupts the membranes of pathogens, it also dehydrates them. We have enough of it to cover basically all industrial floor space in the US plus a wide variety of residential space 24/7. It’s in fog machines at like Broadway shows, it’s extremely safe. Hypochlorous acid is also an interesting method of sterilising things. You can make this at home using salt water and electricity. You don’t want to get it in your eyes, but it’s totally safe for skin, you can like put it on your face or as hand sanitiser. Philanthropists could potentially cover their entire countries. You know, if there’s like someone in Norway, it’s like $50 million: no more pandemics in Norway. There’s another type of offence dominance, which is: no matter how much the defender spends, they can get through, and I think nuclear weapons are a good example of that. Biology, I do not think that that is the case. There are tractable things we can do. I think some of them could be surprisingly affordable, at least to buy time for the more expensive countermeasures to come online. Rob Wiblin: Today I’m speaking with Andrew Snyder-Beattie, who runs Open Philanthropy’s biosecurity programme, which has so far dispersed hundreds of millions of dollars and is looking to disperse hundreds of millions more. Andrew has spent many years — I guess at least eight, possibly more — thinking about how to prevent human extinction from the worst biological catastrophes. And his team at Open Philanthropy has come up with a concrete plan that they think can drive down those risks by at least half, possibly a whole bunch more. They call this plan the “four pillars” strategy, but it does have quite a lot of crazy-sounding components to it — crazy-sounding components that we’re going to interrogate today. Thanks so much for coming on the podcast, Andrew. Andrew Snyder-Beattie: Great to be here. Thanks for having me. Rob Wiblin: What are the threats that keep you up at night? That really make you worry? Andrew Snyder-Beattie: One interesting place to start might be thinking about the historical Soviet biological weapons programme. In the ’70s and ’80s, the Soviet Union basically had tens of thousands of scientists, in violation of the Biological Weapons Convention, coming up with the craziest possible weapons that they could come up with. So things like: combining smallpox and Ebola into a chimaera virus, creating strains of plague that were resistant to 16 different kinds of antibiotics, or putting things that would give pathogens an autoimmune reaction which would make it very difficult to counter. All of those were things that the Soviet weapons programme was doing in the ’80s — so you can imagine with 40 years of additional technology, the possible biological weapons of the future are much scarier. Rob Wiblin: If a bunch of those biological weapons had been released, what would it look like? Andrew Snyder-Beattie: You know, in fact, we don’t need to have hypotheticals here. There were a number of accidents in the Soviet Union. In one of them, they were testing a smallpox weapon over the ocean and it hit this fishing boat. Then when the fishing boat got in, they had to quarantine and vaccinate everyone. And it’s very fortunate that it ended up getting contained within that. There were a number of other accidents as well. There was this big Sverdlovsk accident of anthrax. That anthrax plume hit over a city. But all of these were relatively contained — because, again, these were weapons from the ’80s. You can imagine much worse things happening in the future as well if there was something that was contagious and very deadly. Rob Wiblin: Yeah. I guess if you managed to release an Ebola that was much more contagious and actually was able to spread using the respiratory tract rather than only on surfaces. Do you think these things could plausibly cause actual extinction? Could they kill everyone or close to everyone? You kill enough people that eventually civilisation just falls apart and then it’s questionable whether humanity continues? Andrew Snyder-Beattie: Certainly, yeah. I think there are a few things going on here. One is you might think that civilisation is kind of like riding a bicycle: you need to have momentum, and if everything kind of falls apart, it’s very hard to put things back together again. But you might also think that biological weapons could directly kill very large fractions of the population on Earth. There are a number of ways that this could happen. One is that perhaps the pathogen is spreading for a long time before we even know that it’s spreading. You could take HIV as an example. We discovered that in the ’80s, but it had been circulating for well over 50 years before we discovered it. And fortunately it was not some airborne virus — but had it been, maybe we’d be in a much more dire situation. Similarly, you could have pathogens that are not just spreading from humans to other humans; you could have pathogens that are kind of persistent in the environment. There have been a number of species extinctions, like frogs that go extinct due to these various fungal infections that are kind of pervasive. And then finally, and this sounds a bit more sci-fi, but Ryan Greenblatt had mentioned it in the previous podcast: you could imagine that an AI system might be willing to gamble on some very risky strategies, including kind of knocking down humanity with biological weapons in order to rebuild from the ashes. So even if the system can’t get everyone in the first go, maybe there are followup attacks following that. Rob Wiblin: Yeah, this is something that people have speculated about: that one advantage that an AI would have is that it’s not vulnerable to any of these biological catastrophes at all. So if it really is something that destroyed humans or most biological life, then that could potentially put it in a much better position to take over. Or even just threatening to — saying, “I’ve gone rogue; I’ve got biological weapons that could kill you all” puts you in a very strong negotiating position if you’re trying to not get shut down. I guess it’s all a bit speculative, but I guess worth at least considering. Andrew Snyder-Beattie: Yeah, and I think this is an interesting argument for why even some of these response things could at least help with a deterrence by denial strategy to make these biological weapons less appealing for future AI systems. Rob Wiblin: Are these biological weapons programmes still going on? Does Russia still have one now? Andrew Snyder-Beattie: The State Department has publicly stated that they believe that Russia and North Korea both have active, ongoing offensive weapons programmes. Rob Wiblin: OK. But I guess it’s not public exactly what they’re doing. We just have to speculate based on what has come out about what they were doing in the ’80s. Andrew Snyder-Beattie: Yeah. So you can read about the history of the weapons programmes, and occasionally there’ll be publications that scientists come out with. One interesting thing is, in the Soviet programme, a lot of people in the weapons programme didn’t even know that they were part of the weapons programme. There were these concentric circles. So they had this giant institute that was studying plague — and most of those people were just studying plague, thinking that they were kind of reducing the risk of plague. Then there was this inner circle that was using all those research results to figure out how to weaponise it. Rob Wiblin: A lot of people have said it doesn’t ever really make sense to use biological weapons, because almost always, if you release some super plague, it’s going to blow back on you — so what’s the tactical or strategic use of these things? Do you think that is a strong argument to think that North Korea or Russia, possibly they would develop these things, but there’s no way that they would ever release them? Andrew Snyder-Beattie: Yes, I do think this is a strong argument. The strongest counterarguments, I think there are two. One is you might think that it would be useful as a second-strike weapon, so you might have kind of a layered defence. And that was, I think, how the Soviet Union was thinking about it. The second thing is, just empirically speaking, the Soviet programme was investing a lot of money into smallpox, which is a very contagious virus. So even if this was somewhat irrational, the fact is they were in fact doing it. And the other interesting thing, if you read about the history of these weapons programmes, is that they somewhat get divorced from what you would think would be the rational, kind of strategic move. These weapons programmes get bureaucratic interests of their own. They want funding for biological science, they want funding for other things. So there’s this kind of runaway culture where they’re just trying to come up with the nastiest possible things, even if it’s somewhat divorced from the strategic usefulness of such a weapon. Rob Wiblin: If I recall, Gorbachev wanted to shut down Biopreparat, which was the Soviet bioweapons programme, but I think he was told that there would be a coup to remove him if he did so, so powerful was this interest group, and they didn’t want their programme to be shut down. That was their livelihood. Andrew Snyder-Beattie: Right, yeah. I don’t know about that exact fact, but certainly something like maybe up to 1% of the Soviet defence budget was spent on biological weapons. So quite a large, significant undertaking. Exactly. Rob Wiblin: OK, that’s biological weapons. There’s this whole other cluster that’s come on people’s radar in the last year, which is mirror bacteria or mirror life. Can you explain what that is? Andrew Snyder-Beattie: Sure. So many molecules on Earth can exist in one of two forms: a left-handed version and a right-handed version. A common example of this is sugar: glucose can exist in the right-handed version — that’s the version that we eat — as well as a left-handed version that you cannot digest, which is pretty interesting. These two molecules are identical if you put them in a mirror. So it’s similar to your hands. Your hands in some sense are identical, but they are mirror images of one another. There are lots of properties where it’s the exact same and there are lots of properties where they’re different. For example, you can’t put a left-handed glove on a right hand. What’s interesting is that many of the molecules in your body — and in fact all of the big, most important molecules — have this chiral property. So if you imagine a strand of DNA, all the little As, Ts, Cs, and Gs use the right-handed version. And all of the proteins in your body, like the bigger molecules that comprise the bigger machines, all use the left-handed version. So if you’re a scientist in a laboratory, in the same way that you can create the mirror image version of sugar, you can also create the mirror image version of those little As, Ts, Cs, and Gs. And if you put the mirror image version of those little As, Ts, Cs, and Gs, you can create a mirror-image DNA strand that spins in the opposite direction, and it looks like the mirror image of regular DNA. One interesting thing is that this is not just true of human biology; this is true of basically all life on Earth: bacteria, humans, plants, everything. All animals use right-handed DNA, left-handed proteins. So a lot of scientists were thinking, “Wouldn’t it be interesting if we could create the mirror-image version of not just DNA or proteins, but an entire mirror-image version of a bacteria, like a whole mirror image organism?” There were a number of labs that were looking into this as a possible exciting project. The NSF even funded about a $4 million grant to look into this. But there’s a major problem with this: your immune system has been trained on molecules that it recognises. And if you flip that molecule to the mirror-image version, your immune system is not going to be able to detect or break down those molecules. What that means is that if this bacteria were to get into your lungs or get into your bloodstream, there is a decent chance that it would grow on achiral nutrients and it would cause a lethal infection. Now, you might then be asking, “There are plenty of bacteria that cause lethal infections. What makes this so bad?” The reason that this is bad is because it’s not just true of human immune systems; most immune systems on the planet have been trained on a certain chirality. So this would not just potentially infect and kill humans; it would potentially infect and kill many species of animals, possibly even species of plants. Plant immune systems work in a very similar way. What that means is that this could be very persistent in the environment. It could be kind of pervasive. This would be a lot less like a human-to-human pandemic, but it would be something that is persisting in the soil, persisting in the environment. If there’s a tree that’s infected outside of your house and the wind blows in, then that would potentially infect you. So it would be much more akin to living without an immune system. And people that have genetic diseases that have certain receptors broken typically die in childhood. It’s a very nasty disease. This would be like the whole world ending up in that situation. Rob Wiblin: So if this theory is right, then these mirror bacteria would have an enormous competitive advantage against every other organism, because they would potentially be able to evade the immune system of basically all other organisms, and they wouldn’t have any natural competitors in that sense? Andrew Snyder-Beattie: Sort of. They would have a big competitive advantage inside of an animal — where it’s evading the immune system, but other pathogens are not evading the immune system. So it would have a competitive advantage there. I think it’s a lot less clear how big of a competitive advantage it would have, say, in the soil or in the dirt or something like that. It would have some advantages. For example, viruses would not be able to infect it. So phages typically cut down on bacterial populations. Other protists that graze on bacteria also wouldn’t be able to eat and digest it. So it would have some fitness advantages, but it would also have some big disadvantages. For example, it wouldn’t have horizontal gene transfer, so it wouldn’t be able to adapt as quickly to different environments. It would also be relatively limited in the types of nutrients that it could get. It would be probably persisting on achiral nutrients, which are a lot less abundant. So I think this would not be something that would literally take over the whole ecosystem, and all bacteria suddenly turn into mirror bacteria because it outcompetes everything. It would not look like that. I think instead it would look like there’s a tiny trace amount of it, but there might be a tiny trace amount of it kind of everywhere. Rob Wiblin: Yeah. When I first heard this idea, I think that the objections that jumped to mind for me were: we wouldn’t be able to attack it or digest it, but it would be able to attack us and digest us. Shouldn’t there be some sort of symmetry there? Andrew Snyder-Beattie: So it’s not necessarily attacking or digesting all of you. The way it would work is it’s only digesting very small parts of you. In other words, the achiral nutrients in your bloodstream. So these are things like pyruvate, glycerol, glycine — nutrients that don’t have this handedness property. And there’s enough of that in your bloodstream such that it could grow and persist. So it would basically just be growing in your bloodstream and eating a small fraction of you, if you will. And that’s enough to cause potentially big problems like sepsis or blocking up blood vessels and causing stroke and stuff like that. Rob Wiblin: Yeah. Another objection you could have is: why couldn’t we just treat these bacteria the same way that we treat other bacteria, using antibiotics? I guess you’ve kind of flagged why that wouldn’t necessarily work: we wouldn’t just have to treat people; we’d have to basically coat the entire Earth in antibiotics in order to stop it growing through all the plants and animals. Andrew Snyder-Beattie: Right. So the first objection is you’re not going to necessarily be able to save the crops or the ecosystem this way, because that would just require far too many antibiotics. But I think there are two other objections to this. First is that, even if we were to pivot 100% of US antibiotic production — including agricultural antibiotic production — we’d only be able to cover maybe about 10% of the US population. So just the scaling of this would be quite grim. The second thing is that the people that have these immunodeficiencies, they tend to die even if you give them antibiotics. So they need to be on the antibiotics prophylactically. So this isn’t something like where you just treat an infection; it means that all of us would have to be taking antibiotics every single day for the rest of our life. Rob Wiblin: And doing that while watching the natural world probably die out. Andrew Snyder-Beattie: Right. And this puts us in a very precarious situation, because then if the power plant gets cut or the antibiotic production facility goes out, it’s game over. Rob Wiblin: I guess we suspect that if we did create mirror bacteria, there’s a good chance that many of the plants around the world would just start gradually dying off as this bacteria spread and began infecting them. Do we know if that would take months or years or longer? Andrew Snyder-Beattie: Yeah. Well, I should also say that there are still a lot of uncertainties here. It’s very hard to predict exactly how this would interact. Rob Wiblin: Because we don’t even know what species of bacteria it is. Andrew Snyder-Beattie: Yeah. I would say probably more than a 10% chance that if mirror bacteria were released tomorrow, it would be catastrophic. But I don’t necessarily think it would be more than 80%, for example. I think there’s still a lot of uncertainty, but more than 10% chance is still like, this is kind of a doomsday scenario. On the question of speed: interestingly, bacterial pathogens tend to spread quite slowly if they only infect plants. There are these studies where an orchard will be infected with something and it won’t even necessarily get to the other end of the orchard within a year. That being said, animals can spread bacteria very quickly. So if insects are infected with mirror bacteria and then they’re spreading it from plant to plant, that could end up saturating a forest quite quickly. Also, just human travel can move things very quickly. I mean, COVID was basically everywhere in the world quite quickly, so there’s no reason to think that wouldn’t be true of mirror bacteria as well. Rob Wiblin: You might think in order to tackle these high-tech problems that we’re potentially creating — by being able to make mirror bacteria or whatever else, or having an advanced bioweapons programme — you need really advanced technology in order to combat it. I think that actually has been a bit of a mindset that people have been stuck in: looking for really advanced technology to counter this stuff. But you actually think, because we need to scale our countermeasures to everyone or as many people as possible almost instantly, you need to figure out what is the most low-tech thing that you can potentially use — something that you might already have around, or something you can manufacture on a completely mass scale within weeks, ideally, or possibly even less. Andrew Snyder-Beattie: Yeah. Rob Wiblin: I think you’ve brought an example of one of the low-tech things that you think might be able to potentially stop any of these. Andrew Snyder-Beattie: So this is an elastomeric respirator, and I think this would help in a wide range of scenarios, perhaps even including something as bad as a mirror bacteria scenario. I do think probably relatively sophisticated countermeasures are going to be required to get us out of some hole if there’s a big catastrophe. But I think there’s a question of how do we buy time. During COVID it took a long time to get those vaccines up and running, and we need to do work to bring that timeline down. But I think we also really need to know how are we going to protect people for the short run while we’re getting those countermeasures up online? Rob Wiblin: Yeah. So the virtue of something like a mask is that they’re not specific to any particular pathogen. In principle, a sufficiently strong or effective mask can basically stop any bacteria or potentially any virus from reaching you if you wore it properly, consistently. But we had N95 masks during COVID and they didn’t manage to stop the pandemic to any significant degree. And they were reasonably effective, in principle at least, stopping people from catching it. Why would we think that something that’s so low-tech that hasn’t really worked in the past would be able to make a radical difference on stuff that is way more dangerous than COVID? Andrew Snyder-Beattie: So the first thing is that we didn’t have enough N95s. In fact, the Strategic National Stockpile had a big stockpile of N95s, but they were out of date, they were expired, and the elastic band was just broken. [Correction: The primary problem with the SNS was that there weren’t enough of them to cover essential workers, not that they were expired or nonfunctional. —ASB] So you couldn’t even wear them. If you compare that to something like an elastomeric respirator, an elastomeric respirator has something like a shelf life of 20 years. So it survives a lot longer. The other thing is that COVID was in this kind of dangerous sweet spot of being dangerous enough to be deadly and kill people, but perhaps not dangerous enough that people would take very draconian measures and, you know, do whatever it takes. I think there’s some irony, which is if COVID had been substantially more lethal, perhaps we would have actually been able to contain it, because there would have been more motivation to do so. You saw this with SARS, which was another coronavirus that was successfully contained in places like South Korea. Rob Wiblin: I guess that was about 10x as deadly, and that prompted a more significant response. People were actually willing to wear masks more often. Andrew Snyder-Beattie: Yeah. Rob Wiblin: What do you think is the chance that within our lifetimes there is a biological catastrophe such that we basically want everyone who’s leaving their house to be wearing one of those masks or better? Andrew Snyder-Beattie: So the Center for Global Development has the number at basically a 50/50 chance within the next 25 years of a pandemic that would be as bad as COVID or worse. Then there’s the separate question of what’s the probability of a catastrophe where it would actually threaten human survival? I think that number is also quite high. I would say that’s something like 1% to 3% — which sounds like a low number, but what that means is I think there’s a higher probability that I die due to a biological catastrophe than I die, say, in a car accident. Rob Wiblin: Well, that everyone dies in a biological catastrophe than that you die in a car accident. Andrew Snyder-Beattie: Correct. Yes. Rob Wiblin: Why have you chosen to work on biological catastrophes in particular over all of the other risks? I mean, I guess we tend to talk more about AI on this show. It’s particularly salient at the moment. But there’s other things you could have gone and worked on as well. Andrew Snyder-Beattie: Yeah. This is interesting. I was working at this existential risk institute way back in Oxford — with you, actually. Way back. And basically at the time, everyone was focused on AI, and there were very small numbers of people focused on biosecurity. The thing that attracted me to biosecurity is, although there was a big biosecurity community focused in general on biosecurity, there was a very small number of people thinking about catastrophic risk and these worst-case outcomes. So the short answer would be that I think it is very tractable. I think it is extremely neglected. I think there are basically fewer than 100 people working full-time on strategies that would encompass the worst-case scenarios. And I don’t think the importance is, say, more than an order of magnitude less important than something like AI risk. Rob Wiblin: Yeah. Why do you think it’s not an order of magnitude less? Andrew Snyder-Beattie: I think basically it depends on how much risk you put on AI. If you put a 30% chance of extinction from AI, then a 1% to 3% chance of extinction from bio is still within the realm of… The tractability and the neglectedness arguments can still outweigh that. Rob Wiblin: I thought you were going to make an argument that the risks from AI and risks from bio are connected. Andrew Snyder-Beattie: Oh, they are. Yeah. Rob Wiblin: I suppose we’ve mentioned one reason already why the two are intertwined, which is that making a biological weapon is one way that an AI might threaten humanity in order to get its way. I guess people think that there’s a high risk that especially open-weighted AI models might be extremely helpful for bad actors who want to run their own biological weapons programme. Basically they can get much better expert advice and assistance, and something that previously would have required something like the Soviet programme with thousands of workers maybe could be done with just dozens at some point in the future. So that’s quite troubling. Andrew Snyder-Beattie: Exactly. That is quite troubling, yeah. Rob Wiblin: Is there anything we can do to stop that from happening? Or is it just a foregone conclusion that sooner or later these open-weighted models are going to be jailbroken and they’re going to be extremely helpful at doing this sort of work? Andrew Snyder-Beattie: I don’t think it’s a totally foregone conclusion. You can imagine different things, like better DNA synthesis screening. So even if a model is giving you terrifying instructions for how to make a biological weapon, ultimately you still have to get physical things in order to conduct that attack. And it’s not trivial to get those physical things, so perhaps there could be various checkpoints to prevent terrorism. Rob Wiblin: Yeah. To what extent do you think it helps us that, even if you’re getting good advice from a computer kind of telling you what to do, it’s just like difficult to run the experiments, it’s difficult to do the work in a lab? I guess biology is very fiddly, kind of famously so. Andrew Snyder-Beattie: Yeah, absolutely. Rob Wiblin: You mentioned in your notes that you want to do a crash programme to put in place your four pillar strategy — which we’re going to talk about later — in about two and a half years. Is that sort of set by this AI deadline, where you feel that the threat is getting worse because of AI advances? Andrew Snyder-Beattie: Possibly. In some sense it’s because I think we could do it in two and a half years, so why not set a very ambitious deadline? But yes, with AI progress being what it is, the possibility that it could help create biological weapons faster, I think even if there’s only a 10% chance of that, that’s something that we should be really sprinting towards and working to close that risk. Rob Wiblin: So I guess ideally, we would stop these pandemics or these biological catastrophes from existing — Andrew Snyder-Beattie: That’s Plan A, yeah. Rob Wiblin: Prevention is ideally better than the cure. Andrew Snyder-Beattie: That’s right. Rob Wiblin: What are you all funding on the prevention side, and what do you feel most excited about? Andrew Snyder-Beattie: A lot of great things. Funding work to think about better DNA screening mechanisms, and funding work to think about how policymakers should be getting better DNA synthesis screening into regulation. We’re thinking a lot about AI models and the ways that you could put guardrails on those that would prevent them from divulging ways of creating biological weapons. So just a lot of very common-sense things that basically everyone agrees would be good. Rob Wiblin: Are you spending most of your budget on prevention? Andrew Snyder-Beattie: Yes. Rob Wiblin: Do you think that’s going to remain the case? Andrew Snyder-Beattie: I think it depends. Maybe some of the strategies that we’ll talk about could absorb a lot more funding, and I think could involve a broader coalition of philanthropists. So that remains to be seen, but for the time being, most of our work is just on prevention. Rob Wiblin: The mirror bacteria and mirror life stands out as a pretty unique risk in its profile and how much damage it could potentially do. What’s being done to make sure that no one ever makes that damn thing? Andrew Snyder-Beattie: Yeah. Well, I think a lot of it is driven by a number of scientists who have really kind of been on the forefront of this — including most of the scientists that used to think that they wanted to build mirror bacteria because it was interesting to them. And I think after considering the risks and talking with other scientists, they’ve really been leading the way in kind of setting a norm and setting a taboo that this is not something that we should be doing: we should not be building mirror bacteria. I think there’s now an open question about, where do you want to draw the red lines? If we all agree that we don’t want to be making mirror bacteria, what about mirror proteins, or really complicated mirror proteins? Most of those are going to be totally fine to make, but maybe a mirror ribosome, for example, or an entire mirror proteome might be crossing the line. So I think there’s still going to be a discussion among the scientists, and we’re supporting those discussions. Rob Wiblin: I guess the challenge there is that while a mirror ribosome might not be dangerous in itself, if we develop the science and the technology to really easily do that, then it’s reducing the breakout time that it would take for some rogue actor or some crazy reckless person in the future to actually jump from what is safe and permitted to actually making a full mirror bacteria. Andrew Snyder-Beattie: Yeah, exactly. So the question is, how far do you want to set that threshold back from what a terrorist could potentially do? Right now that would be a very tall order. But one interesting thing is that there have been some terrorist organisations with relatively large biological weapons budgets. Aum Shinrikyo was this doomsday cult. Their chemical and biological weapons programme had something like $3 million per year allocated to it. It would cost a lot more than that to make a mirror bacteria. Some scientists estimate that it would be like $500 million, so you’re still two orders of magnitude away. It would also require some very, very talented scientists, which typically terrorist organisations don’t have. But yes, I think that that’s a good reason to keep the threshold high. Rob Wiblin: So my understanding is that it’s close to unanimity among the scientists in this area that they don’t — Andrew Snyder-Beattie: Almost, yeah. Rob Wiblin: How much do you think that has reduced the risk that there could be some sort of mirror bacteria released over the next 50 years? Andrew Snyder-Beattie: Quite a bit. Maybe a factor of two, factor of four, something like that. For context, I had maybe a 1% to 2% chance of extinction due to mirror bacteria previously. So maybe that’s down to 0.25% — which is still a terrifyingly large number, and we need to drive that risk down further. Rob Wiblin: Given that there’s no one who really wants to do it, how would we end up with mirror bacteria? Andrew Snyder-Beattie: I think it would be that we don’t set the threshold far enough, so we go right up to the brink. And even if all the scientists agree that we don’t want to make mirror bacteria, perhaps all the components are there, all of the precursor materials. Perhaps AI systems make it so that it’s quite easy to take all of the things that are right against the threshold, and talk you through how to build that, how to test it. So the barrier to entry just keeps getting eroded steadily in the future decades, until the point where it becomes quite accessible. That would be the nightmare scenario. Rob Wiblin: I guess North Korea would be another possibility, that they might develop it for deterrence purposes? Andrew Snyder-Beattie: Maybe. It’s quite a wild thing. You’d be developing a weapon that would be killing the political leadership of your own country. There are other arguments against it as being a good doomsday weapon: it would be very hard to test it and know with reliability that it would work. The advantage of a nuclear weapon is you can test it. You can run it through a bunch of tests; you know exactly what you’re going to get. With something like a biological weapon, it’s actually quite hard to make a credible deterrent, because it’s very difficult to know exactly how it would work in the environment. Rob Wiblin: I think a lot of people have this idea in their heads that we can’t really go extinct from biological threats because they’re somewhat self-limiting — because a disease that kills people really quickly or kills almost everyone who gets infected with it has a really difficult time spreading. I suppose you’ve discussed one way that that can fail to be the case, which is if the pathogen is spreading through the environment, through all of the plants, through the soil, so you could catch it that way. Are there any other ways that we could end up going extinct, despite this factor that more lethality can often reduce spread? Andrew Snyder-Beattie: Yeah, if the lethality is delayed. HIV again is a good example here, where the lethal symptoms don’t typically appear until eight to 10 years later. That’s far outside of the window in which the virus is getting spread. So that could be another way that this could be very bad. Rob Wiblin: OK, so what are the four different pillars of your defensive strategy here? Can you basically name them all, and explain what each one is at a high level? Andrew Snyder-Beattie: The first thing I should say is that the four pillars plan is basically a plan to buy time. You want to basically intervene at the earliest possible stage and make sure that society and civilisation can keep running while we buy time for medical countermeasures — which I think is eventually the way that we would need to get out of a really bad scenario. The four pillars, in order, would be: The first is personal protective equipment: things like really good elastomeric respirators, masks to keep people protected from respiratory pathogens, possibly other types of PPE. The second pillar is basically protected buildings: eventually you’re going to need to take off your PPE, and you need to do that in a clean, safe environment. The third pillar is detection: you want to make sure that nothing is spreading without you knowing about it, so having really good detection systems is very important. And then finally, medical countermeasures. I think there’s a question about where the offence/defence balance eventually bottoms out with medical countermeasures, and hopefully it bottoms out with the defender winning. If that’s the case, then I think medical countermeasures would be very promising. Rob Wiblin: OK, so pillar one is personal protective equipment to protect individuals. Pillar two is biohardening of environments like homes and offices to prevent anything from getting in that might infect people. Pillar three is detecting the pathogen as early as possible, and being able to observe if it’s spreading where you thought you had prevented it from doing so — and the earlier we know, the sooner we can put in place these other measures and the less it’s killed people in the meantime. And the fourth one is trying to escape from this situation where everyone is in their biohardened homes, where ultimately we want some more permanent medical fix that can actually just get rid of the pathogen or protect people forever. Andrew Snyder-Beattie: Exactly. Rob Wiblin: We’re going to dive into the plan in a second, but one of the reasons you’ve come on the show is that you’re trying to hire for a whole lot of roles, including some pretty senior roles, to make this plan happen. We’re going to talk more about that at the very end of the episode, but I thought it would be useful to just flag what some of those positions are now, so that people can be thinking in their heads, “Maybe I’d be suitable for running one of these four pillars of the programme.” What are the roles that you’re trying to recruit for at the moment? Andrew Snyder-Beattie: Yeah, recruiting for a lot of roles. At Open Philanthropy itself, we’re going to be on the hunt for grantmakers: people who could deploy tens of millions of dollars to reduce these biological risks, who can learn about a new field, get information, talk to people. I think it’s a common misconception that grantmaking is about sitting in a chair and reading grant applications coming in. We make very few of our grants that way. The vast majority of our grantmaking is you have to spend the first 5% of a project doing it yourself to understand what it is; you have to get to know people in the community to understand who’s doing good work here; and then you want to be headhunting the top people to do the things that you need to have done in the world. On that note, we’re also looking for many roles outside of Open Philanthropy. For example, on the PPE project that we’ll talk about, we need someone to lead a really good nonprofit — who’s going to basically run that, run the manufacturing, run the distribution, think through all the things that need to be done there. It’s a very senior role. On some of the other things that we’re looking for: people to basically just own very large components of the problem. We’re looking for more researchers, and we’re also looking for even people who are just pivoting their career and don’t know quite what they want to do. We have a scholarship and fellowship programme for people to transition into biosecurity. So we’re going to have an online form — there’ll be a link to that — and I would encourage you to fill that out if you’re interested in transitioning into biosecurity. Rob Wiblin: I think up until now, most people have thought of biology as kind of the archetypal case where offence is stronger than defence, and it’s going to be potentially just extremely difficult to protect us from these threats. That could lead to a degree of fatalism about how maybe the situation is just hopeless, or all we can do really is try to prevent people from creating these things in the first place — but if they do, then we’re kind of screwed. Where do you think that the balance lies in offence and defence in bio? Andrew Snyder-Beattie: I think there are a lot of different things that you could mean when you talk about an offence/defence balance. One thing you could talk about is the cost of the attack and the cost of the defence. There is an example where I think the attacker has a huge advantage. One very concrete example of this is that after 9/11, the United States bought well over 300 million doses of smallpox vaccine to basically cover the entire US population. That cost well over $1 billion. Then there’s the question of how much would it cost to create smallpox? One number here is there was a postdoc that synthesised horsepox, a very similar virus, for about $100,000. And you know, even if you added an order of magnitude onto that for evading synthesis-screening mechanisms and acquiring the expertise, you’re still looking at an offence/defence ratio of 1,000 to 1. If you take the $100,000 number, that’s like 10,000 to 1. So it’s very skewed, and it’s hard to think about any other area of national security where there’s a 10,000 to 1 cost ratio. So that’s one thing that’s quite scary. And I think that’s what we mean when we say biology is potentially offence dominant. There’s a separate question though. Take any given person or any given city: could you protect that city or the majority of the people in that city if you really had to, and you were willing to spend that money? Because you could imagine there’s another type of offence dominance, which is: no matter how much the defender spends, they [can] get through. I think nuclear weapons are a good example of that. Maybe we have missile defence now, but before that, basically there’s almost nothing the defender could do. In biology, I do not think that is the case. I do think there are tractable things we can do — and in fact, I think some of them could be surprisingly affordable, at least to buy time for the more expensive countermeasures to come online. Another way you could think of offence/defence balance is through kind of a silly thought experiment: imagine there’s a person in a box, and both the attacker and the defender get to release some sort of biological or chemical agent onto that person in the box. The question is: can the defender successfully keep that person alive? Here I think the answer is probably not. I think the attacker has the big advantage. But I don’t think that’s a realistic thought experiment, because people are spread out, and it’s quite hard to get physical things delivered to people — and I think that’s fundamentally the thing that the defender has the advantage on. So you could imagine an alternative thought experiment, where there are like 10 people in 10 different boxes that are connected, and the attacker gets to infect the first person, but the defender is trying to protect as many of the other people as possible. Maybe trying to protect the first person, but maybe that’s hopeless. But what the defender can do is just build a wall and prevent the disease from spreading. I think that’s an area in which the defender has a real fundamental advantage. Rob Wiblin: Yeah. Is that basically the core weakness that you have? If you’re a bacteria or a virus trying to kill everyone, basically you just have no way to penetrate through walls or through physical barriers. And I guess you’re also vulnerable because you’re so small. That creates the advantage that people can’t see you – you can sneak into people’s lungs without being noticed — but on the other hand, you’re so small you’re completely vulnerable to heat, to UV, to chemical attacks that would just disintegrate you. You basically just aren’t large enough to defend yourself. Andrew Snyder-Beattie: Basically, yes. The other thing I would add to that is just straight-up filtration. And maybe you would say that maybe in the future we could have nanotechnology, and these little microscopic robots that could burrow through your filters and burrow through your walls. But that also has a lot of other constraints, like just the simple amount of energy that each of those things would need to be holding. It’s also not clear that you couldn’t use similar countermeasures — like you could have your nanobot pesticide that kills it in a similar way. So I think these are a lot of ways that the defender basically is fundamentally protected. Rob Wiblin: I suppose the other way in which they have a disadvantage against humans is that we’re kind of intelligent, and we can use science and technology to think up specific countermeasures that can target them in particular. Whereas a bacteria can’t be doing science in order to figure out how to outwit us. It does have evolution as an option to try to move in a more dangerous direction or to evade our countermeasures, but that probably would be slower, basically. Andrew Snyder-Beattie: Yeah. And evolution is not going to be optimised for killing people. So if there’s some pathogen that’s evolving, probably it’s going to be evolving in a direction that’s less lethal. In fact, this is interesting. During the Soviet weapons programme, they were generating these gigantic vats of anthrax. What they would find is that the anthrax would evolve to get very good at growing in these giant containers and less good at killing people — which is like exactly what you would expect from evolutionary pressure. So I think there would be a similar dynamic if there’s something spreading through the environment. Rob Wiblin: All right, pillar one of the plan is personal protective equipment. Do you want to bring out the mask that you’ve brought again? Andrew Snyder-Beattie: Yeah. So this is an elastomeric respirator. Elastomeric respirators are really excellent for a variety of reasons. The first thing is that I could wear this six months in a row without needing to change the filters. So this is not like an N95 where it’s disposable, and you end up needing to change it every single day. I could wear this six months, and it protects me for the full six months. Another thing is that the shelf life is 20 years. So if a government were to purchase a bunch of these, that gives you basically 20 years of protection where you’re protecting your population. The other thing is that the quality of protection that you get is higher on average than for an N95. The protection factor for something like this is about a factor of 100. That means it filters out 99% of particles coming in. And interestingly, that also applies to particles coming out. So sometimes with an elastomeric respirator, you’ll see it has a valve. This one does not, so this also filters the breath coming out. What that means is that if both of us are wearing one of these, that would be a factor of 10,000: 100 reduction for me and then 100 reduction for you. Rob Wiblin: Is that enough to stop even extremely contagious diseases? Andrew Snyder-Beattie: We think it is. In fact, you can actually put an upper bound on the amount of aerosols that a human generates. And our understanding right now is that basically nothing gets through a factor of 10,000. That’s basically an upper bound. So even measles, which is the most contagious virus, would not be getting through a factor of 10,000. That’s assuming it fits properly — and the nice thing about this is that you don’t need training to fit it. This will fit 90% of faces, just first go basically. It just has a much wider margin of error. Whereas with the N95s, if you’re a healthcare worker, you typically need to get the fit testing and whatnot. I don’t know if you had this experience, but if you’re wearing an N95 and your glasses are fogging up, then that means it is not fitted properly. Rob Wiblin: Yeah, there’s definitely air getting in on the sides. This thing, if you look on the inside, it’s got this part where it has a much bigger buffer, where it can fit different face shapes and it’s still going to have a seal. Andrew Snyder-Beattie: Yeah, exactly. So if I wear that, it’s very sealed. Rob Wiblin: I think the audio people are not going to love this. You can do like a fit test, so you push on those things there — Andrew Snyder-Beattie: Yeah. And you can tell that it seals. Rob Wiblin: You can tell that there’s no air that can get in around the side. Andrew Snyder-Beattie: Yeah, exactly. Rob Wiblin: For the people who don’t have video, you should go check out the video. I guess it looks… You mentioned Darth Vader earlier. Fortunately, it’s a nice blue colour. So it doesn’t look too weird. It’s a sort of latex silicone-y thing. It’s got two things on the side, a little bit like a fallout mask. And it’s got these elastic bands that go around the head. And you think that these things would last 20 years? Andrew Snyder-Beattie: That’s what the current testing suggests. I think more tests could be good. The interesting thing about this specific model is that this is called the EM Pro, and this actually was the result of a grant from BARDA, the government agency in the US, which was funding a lot of great research on mask innovation. And they funded the research that went into this. It’s really cool for a variety of reasons. This particular mask is made of one piece of medical-grade silicone. So that means that you don’t need any labour to assemble pieces. Other masks, you need to put together different pieces. But this, you manufacture it in a silicone injection mould. You can imagine a waffle press type thing. Rob Wiblin: Just like inject the hot silicone in there? And I guess as it cools off… Andrew Snyder-Beattie: It cools off and then you pop it out, yeah. Then basically this is the thing that comes out of it. [Correction: injection molding begins with liquid silicone rubber; it isn’t melted and cold in the mold —ASB] Rob Wiblin: So it has no weak points or no moving parts that can break. Andrew Snyder-Beattie: And you can pop this in the dishwasher to sterilise it. It’s one piece. You’d have to pull the filters out. Rob Wiblin: And the filters are replaceable if they get damaged? Andrew Snyder-Beattie: The filters are replaceable, but you don’t need to replace them that often. People are maybe used to thinking about the electrostatic filters. Those are some of the ones that are disposable, where they rely on this electric charge and they lose the charge over time. These are just mechanical filters, so you could be using these for six months. Rob Wiblin: I guess eventually they get clogged up, but it takes ages for that to happen. Andrew Snyder-Beattie: Yeah, yeah. Rob Wiblin: I heard on grapevine that these were good, so I bought some. I think I paid $50 per one. I guess you hope to get the price down to $5? Andrew Snyder-Beattie: That would be ideal. We’re reasonably confident we can get the cost down to $10 per mask. Rob Wiblin: What are the different ways to shave off the cost? Andrew Snyder-Beattie: One thing you can do is replace the medical-grade silicone with just food-grade silicone. Medical-grade silicone you typically need if you’re going to have something inside of you, but if it’s just on your face, probably food-grade silicone is fine. That also means that the speed of manufacturing might be faster. The logo: you don’t need the logo, you could get rid of that. There are other various innovations. Like right now there are two filters; maybe you could replace that with one central filter. A lot of different things like that to get the cost down. Rob Wiblin: Because also if you’re just making 100 or 1,000 times as many, presumably you’ll find all kinds of ways to make it cheaper. Andrew Snyder-Beattie: Economies of scale, yeah. Rob Wiblin: Is it hard to get the food-grade silicone? Andrew Snyder-Beattie: Nah, it’s pretty cheap, and you can get a lot of it in the US. So maybe one interesting implication is that if we can actually get the cost of this down to $10 or maybe even $5, this becomes one of the most cost-effective ways of preventing respiratory transmission. If the shelf life of this is 20 years for $10, that means basically 50 cents per person per year of protection, which is just outrageously cost effective. So this is substantially more cost effective than an N95, and basically would be blocking most respiratory pathogens, or basically all of them. So if you’re a government, I think it makes a lot of sense to just stockpile enough to cover your entire population. Right now we spend about $10 billion a year on missile defence. Stockpiling one of these for every single person in the US would be two orders of magnitude cheaper than that. It depends on what you think the risk is of a nuclear missile launch versus a pandemic, but this is extremely cost effective. I think any rational national security person should think that a population should have one of these for every person. Rob Wiblin: You said that this is the result of a BARDA grant? Andrew Snyder-Beattie: That’s right. [Biomedical Advanced Research and Development Authority.] Rob Wiblin: It’s like DARPA or IARPA, but for biological stuff? Andrew Snyder-Beattie: Kind of. DARPA and IARPA fund earlier-stage research that’s somewhat new. BARDA will fund more intermediate-stage stuff. So they’ll actually fund companies to develop things. They’re responsible for a lot of the really good vaccine development and stuff like that. They fund a lot of really important stuff. Rob Wiblin: So they had a prize or a competition for people to come up with the best new mask — I guess this was in response to COVID — and this is one of the winners. [Correction: EMPro was not actually one of the winners of this challenge, but instead received a separate BARDA grant.] It’s got a lot going for it: the advantage here is it’s much cheaper per day of use, much cheaper per year of coverage in storage. I guess it requires even less space, because you don’t have to stockpile like an N95 for one person every single day that they’re going to be using it. So I guess it’s like a tenfold or hundredfold reduction in cost relative to the N95s. I guess the N95s were maybe still worth stockpiling if that was all that we had, but at one-hundredth the price. Andrew Snyder-Beattie: Yeah, this is just the common-sense thing to do. I could talk a bit about the disadvantages. I don’t think it dominates on every dimension. One thing is that healthcare workers don’t like wearing these, because they make you look a bit more scary, and it’s harder to talk through them compared to the cloth of the N95. So that’s one disadvantage. If you wear this for long periods of time, the condensation will get kind of gross. So I’m not saying these are perfect in every way, but I think in a life-and-death situation, it’s pretty obvious that this is the thing that I want. Rob Wiblin: I guess depending on how bad the pandemic is, you could potentially take it off in the bathroom or something. Andrew Snyder-Beattie: You could. The other interesting implication of the $10 cost per mask is that this starts to get within the budget range of, say, a group of philanthropists that wanted to do this even if the governments were not making a rational decision and buying a lot of these. That’s something that we’re in the early stages of thinking through. But you could totally imagine a philanthropic effort where if you were just trying to cover, say, the people that had to go outside during a catastrophe, a group of philanthropists could come together to stockpile, say, 30 million of these — that would be $300 million, maybe $150 million if the cost is cheap enough — and hire a bunch of really good people to do the shipping and logistics. And the next time there’s a pandemic, if you need to work in a power plant or you need to work in a water treatment facility, you wake up one morning and there’s one of these on your doorstep for every single essential worker. This is something that private philanthropists could just do to protect people. Rob Wiblin: Yeah. I guess in the US and UK, most houses get a delivery every day anyway, so it’s pretty straightforward that you could deliver it to almost every property within a few days. Certainly if everyone was clamouring for them, then I don’t think it would take long to distribute them. Andrew Snyder-Beattie: Yeah, this is not some hard technical problem that we need to solve. We just need to make this and we need to give it to all the people that need it. And we can map that out, and it’s a very tractable problem. Rob Wiblin: So if we imagine it costs $10, and I guess globally you might have a billion essential workers or something. So it costs like $10 billion globally, and that would cover you for 20 years. So it’s more like $500 million per year. That’s actually a pretty small fraction of global philanthropy, really — like a negligible fraction. And in the US it’s even a lot more, like very obviously within the budget. Andrew Snyder-Beattie: Right. I mean, philanthropists could potentially cover their entire countries. If there’s someone in Norway: $50 million and no more pandemics in Norway. It’s pretty outrageously cost effective. Rob Wiblin: What sort of staff do you need in order to make that vision a reality? Andrew Snyder-Beattie: We need people to basically run this nonprofit. We need people to think through the manufacturing of this — so manufacturing experts, product design people who’ve successfully gotten products across the finish line, people with silicon injection moulding experience, people who’ve worked in the global health space and could think about ways of integrating this into global health systems. Yeah, we need a tonne of different roles. We’re basically establishing a new nonprofit to explore this idea. It’s still very early stages, but I think there are a lot of great things that we could do to explore this idea. Rob Wiblin: Is it possible to mess up wearing it that easily? I guess it seems like it has more of a margin for error. Andrew Snyder-Beattie: Well, your beard is actually gonna be a bit of a problem. Rob Wiblin: I’d rather die than shave off my beard. Andrew Snyder-Beattie: It’s much harder to screw it up. But I mean, during COVID you’d see people with the mask like under their nose, so it’s definitely possible to screw up wearing it, but maybe the room for error is much harder with this. Rob Wiblin: I guess I’m thinking about that because during COVID people were so lacklustre in their effort. But I suppose we’re imagining here we’re trying to protect against pandemics that kill almost everyone who gets the disease. The motivation is going to be pretty high. Andrew Snyder-Beattie: I think that’s right. Rob Wiblin: To what extent do we need to separate in our minds the kind of classic respiratory virus that only spreads between people, and something crazy like mirror bacteria that would just be everywhere, potentially ambiently in the environment, when considering whether this would be sufficiently effective to protect people? Not just most of the time, but we really need these essential workers to be surviving until we can come up with medical countermeasures to end it, so you need them to consistently be wearing the mask properly every single day. Andrew Snyder-Beattie: Yeah, absolutely. Either way, you’re going to need respiratory protection. So whether or not it’s a more traditional respiratory pandemic or something more exotic like mirror bacteria, I think the strategy is the same: you need to protect people, you need to protect their lungs. And respiratory protection I think is actually the weakest link in this chain. There’s a question of whether or not these masks would be sufficient. I think there are a lot of uncertainties as to how much mirror bacteria would actually be in the environment if a catastrophe like that were to occur. I think there is a decent chance that two orders of magnitude would be sufficient to keep people alive and protected, so this mask might be sufficient. There’s also some chance that there would be higher concentrations of mirror bacteria in the environment, and you would need to get additional orders of magnitude. I do think there could be ways of doing that, but to get that at scale, you would need to improvise. So you could do things like hook up vacuum cleaners to air filters and basically create positive pressure and then have that over this, which would then give you additional logarithms of reduction or additional orders of magnitude of reduction. Rob Wiblin: That sounds like a significant step up in complexity. Andrew Snyder-Beattie: Potentially. I think that’s why we need people to be doing this early research, and thinking about this, and engineering things to see what works and what doesn’t work — so that in an emergency, we’re not trying to figure that stuff out. Rob Wiblin: In a worst-case scenario, are people also going to need goggles basically, or something to protect their skin from getting exposed? Andrew Snyder-Beattie: Yeah. So this is really interesting. The respiratory pathway is, hands down, the weakest link. If you compare the surface area of your eyes to the surface area of your lungs, it’s a factor of 10,000. The surface area of your lungs, there’s this common factoid that it’s basically the size of a tennis court. Then you compare that to the size of your eyes and actually the respiratory route is much more vulnerable. Now, that’s to aerosols that are kind of ambiently in the air. There could be droplet spread, in which case — Rob Wiblin: Like touching your eye. Andrew Snyder-Beattie: Yeah, touching your eye. Exactly. So I think there are other reasons to think you would want to protect your eyes. But the other important thing to know is it’s much easier to improvise protection: you just wear glasses, goggles, plastic face visors, stuff like that. Whereas improvising respiratory protection is a lot more difficult. So this is the type of thing we want to have in advance. Rob Wiblin: And we’re imagining that most of these pathogens wouldn’t be able to get into people through the skin? Andrew Snyder-Beattie: Your skin is really good. Hundreds of millions of years of evolution have made your skin good against fighting pathogens. It’s extremely dry. Yeah, it’s good stuff. Rob Wiblin: All right, let’s push on to biohardening of houses and offices and things like that. How do we make safe living spaces for hundreds of millions of people when they’re not doing this sort of essential work where that forces them to go outside? Andrew Snyder-Beattie: So there are a lot of possible things that we could do here. One would be things like ultraviolet light, which sterilises pathogens and could be quite good, although we don’t have that widely deployed right now. So the question is, what would we do in an emergency if there was something that happened tomorrow? One interesting option is various vapours that could potentially sterilise pathogens. So this right here is propylene glycol. There are other vapours as well, like triethylene glycol. I learned about this from a professor at Johns Hopkins at a happy hour in DC. It turns out that this is a chemical, it’s in fog machines at Broadway shows, it’s in vapes. The interesting thing is that we produce a huge amount of triethylene glycol. It’s used in natural gas processing, which means that we have enough of it to cover basically all industrial floor space in the US, plus a wide variety of residential space, 24/7. And there are a number of studies in the ’40s done on this in military barracks and things like that that showed that it got basically a factor of 10,000 reduction of pathogens in the air. It’s a chemical that kind of disrupts the membranes of pathogens. It also dehydrates them. And you know, there are plenty of chemicals that we could put in the air that would kill pathogens. The problem is that most of them are also quite nasty for you too. And this is quite interesting, because it’s extremely safe. Rob Wiblin: I guess fog machines aren’t dangerous, but I wouldn’t want to necessarily live next to a fog machine constantly. But you’re saying even then, it basically does almost nothing to the lungs? Andrew Snyder-Beattie: Yeah, that’s right. Probably we should be doing even more studies for chronic exposure. But my understanding is that it’s just very safe. Like, just outrageously safe. So this is one example of a chemical that’s already very widespread. We already have a lot of it. So this is the type of thing that you could do if you’re in a situation where you need to take your mask off, but you still need to be reducing the concentration of pathogens in the air. Rob Wiblin: Yeah. How much of it do you need? I guess that would only cover a very small room. Andrew Snyder-Beattie: Interestingly, it works when it’s totally invisible. So there is propylene glycol outside of the fog that’s covering us. So the fog is actually mostly invisible when it’s working. Rob Wiblin: Yeah. For people listening who can’t see, it’s a small handheld device that’s producing, I guess it kind of looks like a vape. Or it’s producing about the same amount of smoke that someone would if they were breathing out through a vape. And I guess it’s literally exactly the same chemical. Andrew Snyder-Beattie: Yeah. Rob Wiblin: How does it kill viruses and bacteria that are in the air? Andrew Snyder-Beattie: We think it works by basically disrupting the membranes of these pathogens, and it also works by dehydrating them. The reason that it kills pathogens at a higher rate than it damages you is that you already have a tonne of water in your lungs, so it comes into your lungs and it’s not going to affect anything. Whereas for pathogens, getting dried out just a tiny bit is going to kill them. Rob Wiblin: And it affects all of them? There’s no bacteria or viruses, common ones that are resistant to it? Andrew Snyder-Beattie: Yeah, this is interesting. Basically all biological organisms share certain vulnerabilities here in terms of membranes. So interestingly, this works against envelopes, and it also works against non-enveloped viruses. I think it’s doing a lot of different things against them. Rob Wiblin: I guess it’s a little bit like asking if there are any animals that can walk through an open flame without dying, or sit in an oven. Andrew Snyder-Beattie: Right. Yeah. Rob Wiblin: Because bacteria and viruses are just so small that even tiny amounts of this stuff is just very large relative to them. Andrew Snyder-Beattie: Yeah, exactly. And the same thing is true of ultraviolet light. Interestingly, you can think of this from first principles: if there’s a certain amount of energy that’s getting put into a very small organism, that’s going to start breaking stuff. It’s not specific to just DNA. So this general “introduce energy to a small object to destroy it” should be fully generable all the way up into crazy nanotech or something like that. Rob Wiblin: OK. So the plan is to get this chemical that we previously were using in gas extraction, and we’re going to distribute it to all of the different houses and offices. And then do they need a machine like this? Andrew Snyder-Beattie: Yeah. Interestingly, there are a tonne of machines like this in the US; there are like tonnes of humidifiers and things like that. So I think we have actually enough to basically cover every US household. Rob Wiblin: Does it just evaporate? Andrew Snyder-Beattie: It also just evaporates. So I could douse my towel in this and just hang up a towel. I wouldn’t need a fancy nebuliser like this. Rob Wiblin: And it just hangs around in the air until…? Andrew Snyder-Beattie: Until the air moves out. I think that’s actually the major constraint. So although it gets four orders of magnitude pathogen reduction in a fixed environment, the fact that there’s air moving in and air moving out means that realistically, this is probably only getting you maybe about a 1.5 order of magnitude reduction, maybe about 30x. Which is still good. And importantly, we already produce enough of this, so we wouldn’t need to on ramp production. I think if there was going to be a project here, it would be someone looking into the supply chains and getting to know people, and figuring out, in an emergency, what are we actually going to do? This might work in theory and on paper to help out a lot, but what are you going to do in a real emergency is still an unsolved problem? Someone needs to start tackling that. Rob Wiblin: Has this ever been used in hospitals or other environments to stop infection before? Andrew Snyder-Beattie: Yeah, it was experimentally used in these military barracks briefly. I don’t know about hospitals though. I’m not sure. Rob Wiblin: Does this work for surfaces? Is this just cleaning the air? Andrew Snyder-Beattie: Yeah, this works for surfaces. In fact, for surfaces you have a lot more options, because you don’t have to breathe in the stuff — so you can use just ethanol, use bleach. The United States produces an obscene amount of ethanol due to the farm subsidies, so we actually already have enough ethanol to basically sterilise all of the major surfaces in the US in residential, at least once a day or something like that. Rob Wiblin: Maybe we should have paused and said, what is the actual threat to people who are staying at home? Say that they’re not going out as essential workers: is the risk that it comes in on some object that they have to get delivered in order to survive? Or if it was mirror bacteria, you could imagine it coming through air vents, or coming in through the soil, or coming in on insects that come inside. I guess that’s a really nasty case. Andrew Snyder-Beattie: Yeah, exactly. With the standard respiratory transmission pathway or something that’s human-to-human, you don’t need to worry about this. But with the things that are environmentally persistent, there’d be this concern that you need to eventually take off your mask: you need to eat, you need to sleep, you need to drink water. This is why it’s so important to have physical space where people can live that’s free of pathogens. So that’s what the second pillar is basically focused on. Rob Wiblin: Yeah. So if someone was an essential worker, they’re going out to operate a power plant, they’re getting exposed potentially to mirror bacteria or whatever else. And then they have to come home to their family. How do they make sure that they don’t bring home the pathogen on their clothes or on their body or something like that? Andrew Snyder-Beattie: One thing is that, in a situation like that, you might have limited movement. If you’re working in a power plant, probably it’s safest just to stay at the power plant overnight and just be living there. I think in a really worst-case scenario, there’s going to be a lot less movement of people in and out of houses. But let’s say you have to move in and out. I think there are still a number of things you can do. You can basically take off your clothes and sterilise them. You could have an improvised airlock-type situation. But I’m not even sure you would necessarily need that. I think as long as you have lots of filters set up in your house, the air changes could make it so that that’s like not as big of a deal if it’s like temporary. Rob Wiblin: And we’re going to sterilise surfaces and clothes just using all of the ethanol that we have? Andrew Snyder-Beattie: Yeah, ethanol. This is actually another really interesting thing. So this is hypochlorous acid. You know, you spray it. Hypochlorous acid is also an interesting method of sterilising things. You can buy it online. It’s actually advertised as an anti-acne thing. It’s pretty safe. You don’t want to get it in your eyes, but it’s totally safe for skin. You can put it on your face or hands. The thing that makes this really interesting is that you can make this at home using salt water and electricity. During COVID you might remember when there was this big hand sanitiser shortage. We were producing enough ethanol in aggregate, but it took time to reallocate that supply chain to making hand sanitiser. This stuff, you can make it at your own house with salt water, basically. Rob Wiblin: Do you get salt water and then run a current through it? Andrew Snyder-Beattie: Yeah, exactly. Rob Wiblin: For very long? Andrew Snyder-Beattie: I think not that long. I mean, it’s kind of crazy. You don’t think of salt as having chlorine in it, but that’s the chemical formula: NaCl. So basically you separate out the chlorine ions and then it becomes steriliser. Rob Wiblin: Has this ever been used before? Andrew Snyder-Beattie: Yeah, this is used all the time. It’s an antibacterial thing. Rob Wiblin: And I guess people don’t produce it at home because there’s never been any need? Andrew Snyder-Beattie: Yeah, it’s really cheap. Another fun fact is that the chemical reaction that this uses to kill pathogens is actually the same chemical reaction that your white blood cells sometimes use. Rob Wiblin: We’ve come full circle. Andrew Snyder-Beattie: I know, right? Rob Wiblin: OK, so you’ve got the gas, you’ve got the surface disinfectants that it’s relatively easy to scale up massively. In your notes, you talked a whole lot about what’s more like a biocontainment strategy where you’re putting duct tape over all of the windows. You were talking about using, I think, air blowers or something, to just blow air into the house through a filter. Is that like plan C or plan B in this? Andrew Snyder-Beattie: Possibly. It turns out that just filtering air is sometimes the simplest and most effective thing. So actually, just getting air filters is probably the best thing. One interesting thing is that people will have these facilities that grow laboratory mice that don’t have immune systems, because you need to run experiments on immunocompromised mice. So they will have these facilities, these kind of clean rooms — and basically all it is is like plastic and air filters, and people being very careful about where they go and touching things. So that’s like one example of a proof of concept that you can use of keeping animals alive without immune systems using very simple techniques. And the cost of one of these immunocompromised mice is only about 10x the cost of a regular lab mouse. Rob Wiblin: I guess we all know how much that costs. Andrew Snyder-Beattie: It’s about $300 versus $30 — at least according to Gemini. Rob Wiblin: For one mouse? Andrew Snyder-Beattie: Yeah. Rob Wiblin: I guess if this was a respiratory virus — something more like the flu or COVID — then we’re not really worried about it blowing in through a window. Andrew Snyder-Beattie: Yeah, you don’t need any of this stuff. Rob Wiblin: You probably don’t need almost any of this. Andrew Snyder-Beattie: Basically PPE is the only thing you need. Temporarily, to be clear. Rob Wiblin: Yeah. So this is all kind of focused on the more dramatic mirror bacteria or something along those lines scenario — where you’re worried about it being ambient in the environment and coming in through all kinds of different surfaces, so you just want to be vigilant all the time? Andrew Snyder-Beattie: Some of it is. I would say that some of the glycol vapours are still very useful in the human-to-human transmission pathway. I think having ultraviolet light could be really helpful if you’re in hospitals and you need to be extra safe, or you’re dealing with people that are coughing and generating even more aerosol. So all that stuff is still super useful. You don’t need the exotic mirror bacteria stuff in order to think that this stuff is important. But in those scenarios it’s strictly necessary rather than just like icing on the cake. Rob Wiblin: Yeah, yeah. I guess you haven’t talked about UVC. There’s UVC lamps that I think can disinfect surfaces and disinfect the air as well, but I guess we don’t have many of those. They’re quite expensive, and I suppose very difficult to scale. So we wouldn’t see that as a core part of the plan? Andrew Snyder-Beattie: I think it should be part of a longer-term plan. I think there needs to be more studies on it. I would like to see people experimenting with it and adopting it more. But I don’t think it’s the type of thing that we could get everywhere within two years — whereas these other options might be very scalable and tractable on very short turnaround. Rob Wiblin: OK. I guess I was all ready to give you a bit of a hard time about the idea of turning these leaf blowers into air filters in every single house and trying to make them almost air contained, like a BSL-2 facility. What is the weakness of using this glycol suspended in the air, and these surface disinfectants? Does this actually just solve the problem? Andrew Snyder-Beattie: In many scenarios it might be sufficient to solve the problem. I think in situations in which the outdoor concentrations of some sort of pathogen are very high, then they might not be sufficient — just because you have some vapour in the air, but it’s getting blown out of the house or out of the building that you’re protecting. So you want to be also, in the worst of the worst-case scenarios, protecting against air coming into the building. The way you can do this is you can just have basically a fan and a filter and you can blow air in. This is what they do for standard hospital clean rooms and other things like that, or the facilities where they grow mice without immune systems. This is a very common thing. There’s a question of how many people could you cover with a strategy like that? Rob Wiblin: Yeah. How many people could you cover with a strategy like that? Andrew Snyder-Beattie: Maybe most people. This is really interesting: about 60% of houses in the US have furnace fans that are powerful enough to push air through a HEPA filter. You can also use things like leaf blowers or vacuum cleaners or other improvised things — the power consumption starts to add up, but you could potentially do this 24/7. Then you might ask, well, we don’t have enough HEPA filters to cover huge populations. But interestingly, HEPA filters… So the first HEPA filters were made during the Manhattan Project. They were worried about radioactive dust getting into workers and hurting them. So they had these fibreglass fibres, and they stacked a bunch of those together, and that was the first HEPA filter. And it worked. And it turns out that the fibreglass in household insulation are even better than the ones that were used in the Manhattan Project. So potentially, you could just rip out the insulation in houses and kind of improvise HEPA filters and stack those up. Again, this is all still working on paper, and I think it would be really interesting to have a team of people that’s actually trying this in the real world and figuring out what the weaknesses are. So this is still early days, but at least on paper it looks like maybe we could, even in the worst of the worst-case scenarios, protect a lot of people just using these improvised methods. Rob Wiblin: Yeah. So I think that there’s two different reasons why you need to be letting air in. One is just that we need some additional oxygen to breathe and to get rid of the carbon dioxide. And I guess you want as clean air as possible to be coming in. But in your notes you also talked about the benefit of having positive pressure, where you’re actually actively pumping air into the house. Andrew Snyder-Beattie: That’s just to prevent wind and stuff from getting in. So there are inevitably going to be some cracks in the building: you’re never going to fully seal something; that’s just impossible. So if you have a little bit of positive pressure, that makes sure that stuff can’t just blow in easily, because everything is blowing out. And the stuff that’s blowing in is going through the HEPA filter. Rob Wiblin: So if you are able to maintain the air being pushed into the house, then you can think that maybe you would basically stop anything blowing in through any cracks that are remaining? Andrew Snyder-Beattie: Right. Rob Wiblin: Would this be as foolproof as it sounds if you could actually pull it off? Andrew Snyder-Beattie: Someone needs to test it. Rob Wiblin: It sounds mental on some level. Andrew Snyder-Beattie: But the thing is, because it sounds mental, there are zero people working on this. This was a small fraction of one of our researcher’s time just looking into the raw materials and what you could use. I’m not saying that there should be a massive effort to look into this, but there should at least be one full-time person looking into this as a backup option. Rob Wiblin: Yeah. What is the work that has to be done now? I suppose it’s figuring out whether any of this stuff works, and then also writing a guide to say, “Should the time come, here’s all the things you might have in your house that you could use to seal it up”? Andrew Snyder-Beattie: Exactly. The ideal kind of two-year goal would be: you have a team of people, they’ve tested out all of the vapours, all of the different improvised methods, and they’ve run it through really rigorous tests. And then at the end of that, you could have a guidebook that explains how to do this in an emergency. Or you could imagine an LLM that’s trained, and you could take pictures of things in your house and the LLM is telling you how to tighten up the cracks and adjust the filter or things like that. So this seems doable. It’s a research project that’s taking advantage of the fact that there could be a lot of abundant materials all around the world that could protect us even against the worst-case scenarios. Rob Wiblin: Yeah. As we can remember in the early stages of COVID, once it really hit home what was happening, everyone was spending almost all of their time learning about this thing and trying to figure out how to react to it. So you’d have a lot of people who would be willing to try to figure out how to protect their house at that point. They’re not going to be willing to do anything right now, but if you can figure out how to repurpose all the stuff that they happen to have lying around anyway, then you would have a huge labour force basically to scale it up. Andrew Snyder-Beattie: Yeah, exactly. I should also mention it might not necessarily be the case that houses are the best thing. Maybe you want to be converting office buildings or malls or other things like that to protect as many people as possible. But yes, basically you take advantage of the fact that, in an emergency, there’s a lot of motivation and people are going to be using the information that’s already been produced. Rob Wiblin: Yeah. So you’re hiring someone to lead on this task of trying out all these different things. This sounds like a pretty cool job. Andrew Snyder-Beattie: I would like to find at least one person. It could be a cool job. There needs to be at least one full-time person thinking about this. It’s crazy that for all of humanity there’s not one person who’s digging into this. Rob Wiblin: That we can’t scrounge up one human. Andrew Snyder-Beattie: Yeah, yeah. So I’m looking. Rob Wiblin: All right, let’s push on to pillar three, which is detection. What is the point of detecting these things a little bit sooner? Eventually we’re going to figure it out, because everyone will be dying. But what’s the benefit of doing it a day or a week earlier? Andrew Snyder-Beattie: I think the benefit could be huge. In a lot of scenarios, you might want to be able to stop the spread of something. And even instances in which you’re not able to stop the spread, if you really imagine a truly catastrophic or existential risk, it’s important to think about how much governments are willing to spend, and willing to go all-in when they have their back up against the wall. One of my favourite statistics is: at the end of World War II, Japan was spending 76% of its GDP on the war effort, and the United States was spending about 41% of its GDP on the war effort. So if you take the current US GDP — it’s like $30 trillion now — and you say that the US spends half of its GDP fighting some sort of catastrophic risk, I think that comes out to like $100 billion per day of acceleration in countermeasures. [Correction: this $100 billion figure isn’t accounting for only half of GDP spent, so would be closer to $50 billion. —ASB] Now, that’s like the hypothetical best-case scenario, where governments are taking things really seriously. But basically if you don’t know that something’s spreading, you can’t do all these really dramatic countermeasures. Rob Wiblin: I guess the earlier you find out, the sooner people will be able to turn that GDP to focusing on this issue. And also, at an earlier stage there’s some options that might be available to you that the door might be closed a bit if you waited too long. There’s also one case where you really do have to find it out, because you might not figure it out for a very long time: if you have a disease that’s spreading that doesn’t really show any symptoms for a very long period of time. More like the HIV case, right? Andrew Snyder-Beattie: Yeah, exactly. That’s where the detection really helps a lot, because you might be able to contain it. And that also allows you to get started working on medical countermeasures, which might take a bunch of time. You might need that time for the medical countermeasures. Rob Wiblin: All right, so what should we do on detection to find out about new diseases soon enough? Andrew Snyder-Beattie: I think there’s a lot of stuff to do on detection. One of the things that Open Philanthropy is funding is pathogen-agnostic metagenomic sequencing. I can get into that in a moment. Briefly, the way we detect diseases now is that people show up in a hospital, and they have symptoms, and maybe there’s a cluster of people. First of all, in some instances, if the symptoms look like the flu or COVID or something, people just get sent home. So in Seattle, during the early days of COVID it was peak flu season. People were showing up to the hospital or clinics, and doctors would just be sending them home, saying, “I’m sure you have the flu,” and not testing. But let’s say that there’s a cluster of people, and they’re showing some symptoms that the doctors can’t quite explain. Typically, the doctor will run a panel of tests. If those tests come back negative, then the doctor is like, “Gosh, this is a bit weird. Maybe we should look into this.” They will then probably take a sample, send it to the CDC. The CDC will then run metagenomics over it to sequence all the stuff that’s in the sample to determine if this is some new pathogen that we haven’t seen before. I should say, in some situations, the system actually works surprisingly well. With COVID, in Wuhan, the virus was sequenced within two weeks of the first cluster showing up. So that’s pretty good. The thing that’s dangerous though is if there’s something spreading that does not show symptoms. For HIV, for example, typically once you’re infected, you’re not going to be showing symptoms until eight to 10 years after you were infected. And it took us a very long time to discover HIV. It wasn’t discovered until the ’80s, so it had already been spreading for many decades. So if there was something that was spreading quickly, that was airborne, that would be far too slow. So there’s this question of, why don’t you just skip to the last step and start doing some metagenomic sequencing on presumably healthy people, just to make sure that they’re not harbouring some sort of pathogen that has a long latent period? And this is something that people are doing. We’re funding the Nucleic Acid Observatory. They’re one of our grantees, and basically they’re swabbing people in the Boston subway and sequencing that. They’re also analysing a lot of wastewater. Lots of times, if you’re infected with something like a virus, it’ll come out in your wastewater. Then all of that goes to the water treatment plant, and they take a sample out of that and they sequence all the viral RNA and DNA in that sample. So they can pick up on interesting things that people have missed by doing that. Rob Wiblin: There’s going to be an enormous number of DNA sequences in people’s faeces, but I guess they just get a good sense of what the baseline is, and then they can just flag anything that they’ve never seen before? Andrew Snyder-Beattie: That’s right. In fact, this is actually one of the big engineering challenges, and I should say there’s a very small team of like 10 people working on this specific problem — so I think more talent directed to this problem would be good. But yes, there’s a huge amount of DNA in these samples, so typically what they’ll need to do is filter it out. So like bacterial DNA that’s with very common bacteria, they’ll want to get rid of that. Human DNA, they obviously want to get rid of that. You can just run it through a filter that also gets rid of the chunky bacteria, so that you’re left with the viruses and free DNA. So there are a lot of sample preparation steps. Granted, these sample preparation steps make it so it’s slightly less fully pathogen-agnostic. So if there was some crazy bacteria that was spreading and it looked a lot like a regular bacteria, you wouldn’t necessarily pick that up using this method. But that’s something that the team is looking into: ways of making the method more generalisable to a full range of threats. Rob Wiblin: I would have thought this is a hell of a bioinformatics problem, because all of these things are evolving all the time. There’ll be new shit showing up constantly. How do you pick out the one that’s the super bacteria that’s going to kill everyone? Andrew Snyder-Beattie: Totally. Yeah, yeah. If you have a bioinformatics background and this sounds like an interesting problem, you should maybe be working on this problem. Rob Wiblin: We are getting a lot better at going from a sequence to figuring out what would the protein or the enzyme do. I think we kind of can do that to a reasonable extent now — just say, here’s a particular genetic sequence in an animal that we’ve never seen before, and we can kind of guess what this protein is probably for. Does that help us? Andrew Snyder-Beattie: It might help us. It might also hurt us. So there’s this interesting question about the offence/defence balance of detection. So imagine you had a perfect tool that would enable you to generate some protein function using a wide variety of different sequences. One thing you could potentially do is engineer a virus that has a very different sequence, but it basically is functionally the exact same as the virus you’re trying to get. If you’re an attacker, this would be very useful. First of all, to get around DNA synthesis screening mechanisms, because when you’re ordering that DNA, they might not recognise that as a dangerous pathogen that you’re ordering. And then second, if it’s spreading and they sequence it, it might not look like any other virus that people have seen before, even if the virus is in practice something that we kind of know what it is. So this might be a reason why really good tools like this could help the attacker. But I think there’s this interesting argument: if you’re able to redesign a virus to that extent, the defender ought to be able to check, and it ought to be in some sense cheaper to check than it should be to create. Kind of like a verification versus creation, like a P versus NP type thing. Rob Wiblin: Sure. I would think most of the time, most genetic sequences are similar to other genetic sequences in other species, because most of this stuff is conserved and reused. If you came up with a completely new protein from scratch using some AI-driven tool that could figure out exactly how the protein would fold and what it would do, then it would stick out like a sore thumb, basically. It would never arise naturally. Andrew Snyder-Beattie: Possibly, yeah. But I think that you’d have to design your bioinformatics to detect stuff like that, and have a really strong baseline, and do other things. Rob Wiblin: OK, so the Nucleic Acid Observatory, they’re going around and swabbing people in Boston. Why are they doing it just in Boston? When I’ve heard this suggested before, it’s always been that you should do it in airports or you should be grabbing stuff off of aeroplanes. Andrew Snyder-Beattie: Yeah, sorry. They’re looking at over 20 wastewatersheds across the US — so it’s much broader than just Boston. The swabbing is an early pilot study that they’re looking at. They’re not sure whether or not to scale that up yet. Rob Wiblin: Ideally probably you’d be sampling from around the world to catch things earlier, but I suppose that’s just insanely more difficult to get permission to do all of that. Andrew Snyder-Beattie: Yeah. And aeroplane waste is interestingly apparently really good, because A, it’s really cold, and B, there’s a lot of detergent in it. So when there’s human waste that goes in, it kind of just gets preserved in the way it was. Because a lot of times in waste, there’s bacteria that grow and they explode in numbers, so then the signal-to-noise ratio gets messed up. So having cold, sterilised waste is ideal for metagenomic sequencing. Rob Wiblin: Aeroplane wastewater: underrated. Andrew Snyder-Beattie: That’s how humanity gets saved. That’s right. Rob Wiblin: Is there much interest in doing this work of detecting what pathogens are out there for just more mundane public health reasons? Maybe you could bring in funding that’s not focused particularly on this. Andrew Snyder-Beattie: Yeah, I think the metagenomic stuff is actually promising for a lot of different reasons. One interesting result that the Nucleic Acid Observatory found is there are a number of different types of flu. So there’s flu A and flu B, but there’s also flu C, which not many people have heard about — and that’s because it’s thought to be really rare and not even worth testing. But interestingly, because they were using this metagenomics approach where they were searching for basically all human viruses and all possible pathogens, they actually found that there was one city in Missouri that had flu C levels that were basically the same as the flu A levels. So yeah, I think we’re learning interesting things that could be applicable more broadly for public health using these methods as well. Rob Wiblin: How early would we be able to detect things using this kind of method? Andrew Snyder-Beattie: It depends on how many areas you’re sampling from, and how deeply you’re doing the sequencing. Right now, by the end of the year, they should be able to hit something like detecting something before it infects, say, 1% of a population cumulatively — which is not very good, to be clear. Rob Wiblin: That’s pretty late. Andrew Snyder-Beattie: Yeah, it’s quite late. So ideally they could drive this down a few more orders of magnitude and get other possible signals. But I should say that even the 1% still could be good in a kind of stealthy scenario — where that might be the difference between most people catching it versus stopping it still relatively early. Which again, we need to improve that, and I think there’s a lot of important work that needs to be done, but I think they’re on track. Rob Wiblin: The challenge with this detection stuff has always been that to catch it when 0.1% of people have it, rather than 1% of people, you kind of need to be scanning 10 times as much stuff to get to have the same probability of doing that, so it’s like 10 times more expensive. So each doubling time that you want to do it earlier, the costs escalate pretty massively. Andrew Snyder-Beattie: Yeah, I do think this is a big problem, and I think it’d be good to supplement this with other detection methods. I do think this is a weakness of this approach. Rob Wiblin: Is that still the main reason why people think it’s going to be very hard to use these techniques to discover a new disease so early that you could just contain it and cordon it off and ensure that it doesn’t spread? Andrew Snyder-Beattie: Yeah, I don’t think the metagenomic strategy is going to be the way you contain an outbreak. I think the more traditional approach is probably going to be better there. Rob Wiblin: Scanning people coming into hospitals or something like that? Andrew Snyder-Beattie: Yeah, something like that. Again, that’s relying on the thing not having like a long latent period, or — Rob Wiblin: Or I guess symptoms that are masked, because they look exactly like something else. Andrew Snyder-Beattie: Yeah. Looks like the common cold or something. That would be bad. Rob Wiblin: Cool. Is there much more to say on detection? Andrew Snyder-Beattie: No, just that again there’s a broad community of people working on disease surveillance, and I think they’re doing really important work there. I think the number of people working specifically thinking about these long-latent things is very small. Rob Wiblin: There is one more thing we should definitely talk about. How do you detect mirror bacteria? Because they’re completely different, right? Andrew Snyder-Beattie: Yeah. Unfortunately I don’t think there’s going to be any detection needed. I think in a mirror bacteria scenario, the forests are going to be getting destroyed, there are going to be dying animals, cities getting destroyed. It’s not going to be subtle. Rob Wiblin: What would it look like at the earliest stage when it started infecting people? You’re saying you’d basically get sepsis because it would get into your blood, potentially? Andrew Snyder-Beattie: I think it’s actually really unclear. You might die of stroke because the bacteria accumulate. You might die of sepsis because it grows out of control. And then your immune system finally might catch some glimpse of certain things. It’s actually not clear how you would die, but generally speaking, having an uncontrolled growth of something in your bloodstream generally leads to death. So that’s a cheery topic. I laugh about it all the time too. Rob Wiblin: All right, moving on from detection… Andrew Snyder-Beattie: Yeah, great. Rob Wiblin: Pillar four: we’ve got to get out of this. We’re imagining a scenario, in the worst case, where you’ve got everyone wearing masks in order to go to work and keep society functioning. Everyone else is hiding in their homes, occasionally seeing their loved ones. I guess the detection phase is well and truly over. How are we going to get back to some sort of normality where civilisation can resume? Andrew Snyder-Beattie: I think the way we dig ourselves out of this hole is going to have to be through some sort of medical countermeasure eventually. That’s basically what we did with COVID: there were lockdowns, eventually we had the vaccine, the vaccine allowed us to get back to normal. I think that’s going to be similar for even more catastrophic events. On the medical countermeasures thing, it gets a little bit more complicated when you’re thinking about an intentional adversary that might be designing things to especially bypass your medical countermeasures. So maybe I’ll talk a little bit about the weaknesses of medical countermeasures before I talk about… I think the reason they’re good is obvious, in some sense. We all love vaccines. Rob Wiblin: People often have the reaction these days that pandemics aren’t going to be such a severe problem because we’ll just do an mRNA vaccine and that will basically solve it. Why isn’t that a reliable strategy? Andrew Snyder-Beattie: I mean, my hope is that that will be a reliable strategy for a lot of different threats, and I think the work that CEPI and other groups are doing is just absolutely essential. But I do worry that in the scenarios where there are adversaries that are intentionally engineering things, that might not necessarily be sustainable or ideal. And then there’s the more common critique, which is that medical countermeasures take a long time to produce. With COVID it took almost a year. The 100-day mission is a thing that people are excited about — and I think is really exciting, but that’s still 100 days to get to have a vaccine. If you compare the speed at which, say, the omicron variant went through China: it infected something like 80% of the Chinese population within six weeks. So 100 days in some sense is still far too slow. That’s why I think we need the protective equipment and the other things in order to slow down the spread and make sure that we can keep things running while we’re doing the medical countermeasures. Rob Wiblin: But do we think that mRNA vaccines, you can make one of those against most of these different threats? Andrew Snyder-Beattie: You might be able to make an mRNA vaccine against a lot of different things, but I don’t necessarily think you could make an mRNA vaccine against anything. Mirror bacteria is a good example. You need to have a special conjugate vaccine against mirror bacteria, and I don’t think mRNA would be suitable for that. In fact, bacterial pathogens in general are much harder to vaccinate against. So like an anthrax vaccine, you need to take five different doses — and generally speaking, the antibiotics are much more efficient. So depending on the biological threat, the vaccines might be more or less effective. And I think that gets into the second point, which is that it’s not obvious that you’re going to be able to make a medical countermeasure against any possible threat. There are several examples here. One would be: there were a number of researchers in Australia studying mousepox — which is like the mouse version of smallpox — and they inserted an immunoregulatory gene into the virus because they were trying to sterilise the mice for some experiment, and that ended up killing even the vaccinated mice. It was highly lethal even in the vaccinated mice. This was just one gene that they stuck into this virus. Again, if you look at the Soviet programme, they had thousands of scientists figuring out how to make their bioweapons overcome vaccination, medical countermeasures — you know, plague that was resistant to 16 different kinds of antibiotics. So I don’t think that I’m that optimistic about finding a medical countermeasure in advance of a threat, because there’s so many different options that an attacker could pick. You’re probably going to have to do the medical countermeasure in a reactive sense, once you know what it is that’s spreading that you need to counter. And even then it’s not obvious that a medical countermeasure is going to work quickly in any given situation. HIV is a good example of this: HIV is a virus that infects the immune cells that are needed to mount a vaccination response in the first place. This is why it’s been so difficult. We’ve been working for 40 years to try and get an HIV vaccine and it’s been very difficult. And hopefully there are some promising signs recently. But the human immune system is in some sense kind of a fixed target — so as technology gets better and better, it’s less clear that you couldn’t find vulnerabilities that break the immune system in pretty fundamental ways. Rob Wiblin: You’re saying if you’re actively malicious and trying to maximise the damage that your designed disease does, basically it’s very likely that you would choose a virus or bacteria that immediately goes and just damages the immune system as much as possible, so that then an immune response is not possible. Andrew Snyder-Beattie: Right. Yeah. Rob Wiblin: So that breaks the mRNA approach. I guess you could still do the antivirals, like the specific chemicals or the antibiotics that target without going through the immune system. Andrew Snyder-Beattie: That would be the hope. Those are currently quite slow. There are not many people working to design new ones, and I think there needs to be a lot more money going into antiviral and antibiotic development. And Open Philanthropy has funded some good work on new methods of making antivirals. Rob Wiblin: If we assume that helping the immune system with a vaccine or mRNA or something like that is off the table, then it sounds like we’re in a pretty difficult situation — because we don’t have an example of quickly, in 100 days or even a year, turning around a new antibiotic or a new antiviral that really effectively hits some new arbitrary disease that we just discovered. I think nothing like that. And we’ll be doing this in a very difficult time, under massive duress. Andrew Snyder-Beattie: I don’t think supplementing the human immune system should be off the table, but I just don’t think it’s necessarily a sure thing, so we need to have backup options. But I think there are some theoretical reasons to think that medical countermeasures might end up being kind of defence-biased in the longer run. This is still a somewhat speculative hypothesis, but it might be interesting to talk about. The hypothesis is called the “wrench hypothesis.” This came about because I was thinking about nanotechnology and the grey goo scenario where you have this little nanobot that’s spreading and eating stuff and killing people. And the question is, could you develop some sort of countermeasure that would stop the nanobot? I think the answer should be yes. I think probably there should be things that you could do there. Rob Wiblin: Basically, you have to find a molecule that fits into some part of the nanobot and jams it up, but doesn’t do that for humans. Andrew Snyder-Beattie: Yeah, exactly. That’s precisely what the strategy would be. Rob Wiblin: There’s got to be a shape like that. Andrew Snyder-Beattie: Yeah, yeah. We call this the wrench hypothesis because it’s like a wrench in the gears of a machine. For those of you who are not familiar, this is exactly how antibiotics work. You can think of a bacteria as being a little machine that’s made of lots of little tiny machines inside of the bacteria. The way antibiotics work is they go in, and they’re a molecule that just sticks to the machine and gums up the gears, and it just prevents that machine from working. And if you break enough of the little machines in the bacteria itself, the bacteria ends up dying. So there’s this question of, could you ever have some self-replicating machine where it was impossible to find a little wrench that broke it? I think the answer is no, basically. If you’re self-replicating, you have to be taking in nutrients, you have to be pulling in molecules. There are going to be delicate things that are responsible for that. And in the limit, you can imagine a nutrient with another thing attached to it that just ends up breaking it. So it’s hard for the thing to discriminate between malicious molecules and the ones that it needs to grow and replicate. This is also true of viruses, I should say. So viruses use a lot of your machinery to self-replicate, but all the viruses have at least one or two little machines that they themselves make, that are different from you. And that’s, again, what most antivirals are targeting: they’re blocking and breaking those little machines. Rob Wiblin: Yeah, OK. So in principle, we think that if our technology was sufficiently advanced, the defender would win here. I guess one reason is that they get to move second. So someone’s got to design the pathogen, they’ve got to design the nanobot. And then it’s not able to adapt all that much, not all that quickly. Then you just get to choose whatever you think is the weakness. Andrew Snyder-Beattie: And maybe you can choose 10, right? There’s no reason to stop at one if you get really good at designing them. Rob Wiblin: OK, yeah. So what should we do in order to take advantage of this fact? Is there much we can do to potentially have medical countermeasures that are much more general and much faster? Andrew Snyder-Beattie: I think this is still early days, so this is still kind of a sci-fi technology. But you could imagine in the future, if models get really good, basically being able to produce antibiotics and antivirals very quickly. You can imagine this kind of sci-fi technology where you get a new sequence, you put it into your AlphaFold+++ that shows you what all the machines are doing, and then you can design a giant library of molecules that stick to those machines and break them. And then you run that through a toxicity screen, and you’re pretty confident that it doesn’t break any of the human machines or break down into some other harmful molecule. Then you also might filter for things that you can manufacture at scale really quickly, and then produce a lot of it and test it. This sounds good in theory. In practice, there are actually a lot of bottlenecks. The thing I’m describing is very difficult. This is not the type of thing that I think we could do in two and a half years. But we’re in the early stages of thinking this through — and we have one researcher who’s looking at the different bottlenecks, but I think there should be more people thinking about this and working on it. Rob Wiblin: I guess there’s probably broader interest in it. Across all biomedical research, I mean: there’s been a huge effort to figure out how to solve the protein folding problem. Not just because we want to stop biological catastrophes, because it’s incredibly useful. Andrew Snyder-Beattie: Yeah, exactly. There’s a very wide community of people working on things like this and thinking through it. Rob Wiblin: Are there any particular missing pieces that you think Open Phil could fund to speed this stuff up? Andrew Snyder-Beattie: I think we’re in the early days of looking into this, and ideally we would hire someone who could figure out whether or not there’s a way that funding could actually accelerate this kind of future. Rob Wiblin: As you were saying, you can just find arbitrary stuff to throw at any nanobot or any bacteria or virus. It’s also occurring to me, couldn’t you just simulate the human body and then find arbitrary numbers of specific poisons that happen to break all of our machinery? Why doesn’t that put us in a pretty dim situation? Andrew Snyder-Beattie: Yeah, I think you could do that — and basically what you’d be doing is generating a big library of chemical weapons. The thing is, the chemical weapons don’t replicate. Rob Wiblin: It’s hard to distribute them to everyone. Andrew Snyder-Beattie: Yeah. You might make a really good poison, but the reason biological weapons are scary is because they self-replicate and they get everywhere. Whereas there are tonnes of molecules that already kill people. Finding new molecules that kill people — Rob Wiblin: “I found a poison you wouldn’t believe!” Andrew Snyder-Beattie: Right. So that’s like the main argument for why this biases the defender. Rob Wiblin: OK, let’s zoom out and consider the four pillar programme as a whole. There’s a lot to like about it. I think even someone who was sceptical would probably think some parts of this might work. But in order to really be secure, we need these four different things to be all working somewhat in order: we need to detect the thing early enough; then we need to ensure that we roll out the PPE to tens or hundreds of millions of people, ideally billions of people around the world; then we need to be biohardening the offices of the things that spread and things that get into the house; and then as everything is maybe falling apart and people are struggling to survive, we’ve got to do the best biomedical research as quickly as possible and then manufacture enough to get to millions of people. It’s quite an effort. Do you think that all of these things would be able to work together? Or would probably one of them break and that would be the thing that would wreck us? Andrew Snyder-Beattie: So I don’t necessarily think you need all four of the pillars. It depends on what kind of threats you’re facing which ones you would need. Like if you really had the amazing sci-fi medical countermeasures thing, where you could produce 10 new antibiotics against a thing instantly, then you wouldn’t actually need anything else other than protection. Rob Wiblin: Just need masks for a little bit. Andrew Snyder-Beattie: Yeah, or something like that. But again, I don’t want to be banking on that kind of sci-fi tech — which is why I think just the really robust physical defences — physical sterilisation; simple, cheap masks — is just the way to go to buy time for the other things. Rob Wiblin: OK, so the four pillars is meant to be in aggregate. It could defend us against the worst-case basic story, but many things will fall short of that and then maybe muddling through would be enough to at least prevent extinction. Andrew Snyder-Beattie: Right. Yeah. And I think all of the four pillars basically benefit, because I don’t think future technologies break them. As technology gets better and better, all four of the pillars ought to bias the defender and get better and better faster than the attacker is getting good. So that’s part of the idea behind them. Rob Wiblin: So what is the biggest weakness of the plan? If it didn’t work, what’s the reason? Andrew Snyder-Beattie: I think there are two arguments. One would just be lots of human error. We have yet to actually test a lot of the stuff. It’s working on paper, but we need people to actually run this stuff to ground, and figure out how well the masks fit after eight hours or 10 hours. I’ve worn them for a while, but I have not worn them for like a full work week. And probably I should do that at some point to really get a taste of my own medicine if I’m telling other people to do this. Actually we have a team week planned where we’re going to do some of this. Another big weakness is if you’re worried about an AI scenario where an AI is deploying biological weapons, it might not be just the biological weapons: maybe you also have to be worried about cyberattacks or drones that are picking off people, and that might make the defences substantially more complicated. The other argument is that, again, we’re imagining just a biological catastrophe. Normally when you think about catastrophes, it makes sense to not think about correlations between them. Like the probability that we have a pandemic and an asteroid at the same time is obviously low. But with biological weapons, like with the Soviet Union, the plan was that they would first nuke and then they would use the bioweapons afterwards to like — Rob Wiblin: Mop people up. Andrew Snyder-Beattie: Yeah. In some scenarios you can imagine there being a tight correlation between things like nuclear weapons or other infrastructure-destroying things combined with the biological weapons. And the plan is not robust to that. Rob Wiblin: Trying to pull all of this off while also having just been through the nuclear apocalypse. Andrew Snyder-Beattie: That seems really rough. I mean, some of the stuff like PPE still should work. You know, you stockpiled it. Rob Wiblin: People have it at home already. Yeah. Andrew Snyder-Beattie: And it also would help you against the fallout. But yeah, that’s obviously a much more rough situation. And similarly with the mirror bacteria scenario: you’re not just having to protect humans; your agriculture might be getting destroyed at the same time. So that compounds the problem substantially. I think there’s just something about having a layered defence, and making sure that these systems are robust enough that they can be done even in a really stressful, bad environment. Rob Wiblin: So it sounds like you’re saying you don’t have a plan for how we can protect against all of these things happening simultaneously, but other people should try to make sure that there’s not a nuclear war. Other people need to do their jobs too. Andrew Snyder-Beattie: Something like that. Rob Wiblin: If this is such a good plan — and, I guess with the benefit of hindsight, somewhat obvious plan in some ways — why hasn’t anyone else proposed it? When you’ve gone and shopped this around with people in government or people in the broader pandemic control area, and you’d be like, “Why aren’t we already doing this?” what’s their reaction been? Andrew Snyder-Beattie: The short answer is actually we haven’t been talking about the plan much, and this is the first time we’re talking about it publicly. So we’re going to do more writing, we’re going to do more to get it out there. On some of the specifics, I think the elastomerics are very well received. Smart people in government do look at that and they think, “Yeah, that actually just makes a lot of sense.” Rob Wiblin: Is there any plan for the government to buy up a whole lot of those, or is anyone really buying them? Andrew Snyder-Beattie: Not that I know of. The US Department of Defense put in an order for a number, and I think that makes a lot of sense and that’s good. But I don’t know of any other big [orders]. Rob Wiblin: So what fraction of the US population do you think would survive here? It sounds like, in principle, it could be almost everyone? Andrew Snyder-Beattie: If people don’t make mistakes or something like that. Rob Wiblin: Would there be a big gain going from having stockpiled 50 million of these masks to protect the essential workers who have to go out, to having 300 million for the US and like 10 billion or something for everyone? Andrew Snyder-Beattie: I think so, yeah. You know, the allocation is never going to be perfect. It’s going to be really rough if you’re going to have to be triaging a situation like that. Ideally you want to be operating from a place of abundance, where you have more than enough for everyone and everyone can get a mask. That would be the ideal situation. So yeah, I think there are still returns to getting more and more people covered. You could also imagine in the crazier scenarios, where you’re also trying to fight some AI takeover, maybe a lot more people need to be going to work than you thought. Like people dealing with the cyber stuff, and people going to shut down data centres or whatever. There might be a lot more that needs to be done there. So ideally you could kind of fight a war against an AI in your sleep, without the bioweapons ruining everything. Rob Wiblin: Let’s say that you did manage to get a bunch of money into this plan, and you managed to get the 50 million masks stockpiled somewhere. Presumably this isn’t going to happen everywhere else in the world. How big a problem is it that there might be significantly more fatalities outside of the United States, or at least outside of rich countries? And supply chains across the world are like, the economy is just enormously contracting at this time. Seems like that could make it harder to get a supply of all kinds of other materials you need to keep things running, to have the scientific base you need to do the best-ever medical countermeasures. Andrew Snyder-Beattie: I do think this could be a big concern. If you look at some of the COVID vaccines, they had supply chains that I think were over 200 components from a lot of different countries. So yeah, you might need complicated supply chains. That might mean you need to protect a very large number of people — even setting aside the humanitarian reason to obviously protect as many people as possible. Rob Wiblin: Yeah. I guess you’re focusing on the US. That’s where you are. That’s something that is perhaps within budget, roughly. But having convinced the US to have enough — Andrew Snyder-Beattie: I don’t think we’re necessarily focused exclusively on the US. There are some reasons to think the US is a good initial place: it’s relatively autarkic, it has enough food to basically cover everyone, energy independence. There are a lot of things that make the US relatively robust to catastrophes. But I don’t think we should be stopping there. I think we should be doing more research on other countries that we’d want to be covering. And generally, we want to be saving as many lives as possible, getting this as widespread [as possible]. Rob Wiblin: So if you wanted to get a mask for every single person on Earth, it costs about $50 billion. That’s a lot of money. Andrew Snyder-Beattie: Assuming we can actually drive the cost down to $5, which is not a sure thing. So let’s conservatively say $100 billion. Rob Wiblin: Call it $100 billion. That’s what fraction of global GDP? Like 0.1% of global GDP? One-off, basically. Or every 10 years. Andrew Snyder-Beattie: Every 20 years. Rob Wiblin: Every 20 years. OK, so it’s like 0.05. Is it 0.005% of GDP on an ongoing basis to have everyone have a mask like that? Seems very doable. Andrew Snyder-Beattie: Yeah, seems maybe like we should do that. Rob Wiblin: I don’t know how much we spend on ice cream globally, but it’s going to be a significant amount. Andrew Snyder-Beattie: Totally. Rob Wiblin: Do you want to hire people to potentially be pushing this overseas? Andrew Snyder-Beattie: Yes, definitely. We have a new nonprofit that we’re setting up. It has an interim CEO. We want to find a great team of people. We just got it started. We want to find a great team of people that’s excited about the personal protective equipment problem. We need people who are manufacturing experts, we need people who are logistics experts, global health experts. We need just a really big team of people to be moving this idea forward, manufacturing it. Thinking through whether or not we should be doing a philanthropic strategy and fundraising from a collection of donors, or whether or not we should be getting governments to do this. Or both. Rob Wiblin: Earlier I was talking about how if there’s other countries that are not covered, that would be damaging to supply chains. Could it also be destabilising if people can see that some countries are protected and some countries are not, that they could potentially turn to violence? Or potentially they just think that the writing is on the wall, they’re all going to die? Andrew Snyder-Beattie: Yeah, you might be worried about that. Interestingly, I do think in catastrophes people actually tend to be quite cooperative. This is slightly different than the situation you’re talking about, with one full country versus another — and that I could see turning violent. But I think the scenarios where people are killing their neighbours and stuff is actually quite unrealistic. You see in actual disasters people are actually really altruistic. Rob Wiblin: I guess inasmuch as there’s a hostile attacker that is creating these diseases, are we assuming that at some point they’ve been killed, or they’ve been stopped from just keep coming every month with a new one? That’s a pretty bad situation. Andrew Snyder-Beattie: Yeah, I am assuming that. Rob Wiblin: Is that realistic? Andrew Snyder-Beattie: It depends on what the threat you’re imagining is. If it’s a state bioweapons programme and you’re fighting a war, then I’m assuming that the more powerful countries are going to step on that country. Rob Wiblin: I guess it would become a very clear target. Andrew Snyder-Beattie: Yeah. I think in the instance of a terrorist who’s using this, the whole world’s resources would be focused on this — and even if they could hide, it’s going to be hard to continue the work. If you’re worried about some AI system, that might be scarier — because maybe the AI system is hidden across like a lot of different servers or something, and it’s telling people that they’re making countermeasures and ordering them around, and turns out they’re making the next generation of weapons or something. So you can imagine scary scenarios like that, but overall I think it’s a reasonably safe assumption that you can stop the attacks once they start. Rob Wiblin: Let’s talk a little bit more about the interaction between these catastrophic bio threats and AI in particular. I’ve heard people make the argument before that it’s a bit silly to be working on bio if you have a picture where AGI is going to come soon, and it’s going to be enormously powerful. Because if we produce an AGI and it’s really aligned with human interests, then it’s going to be able to come up with technologies and a better plan than what we’ve got here in order to protect us. If we come up with a misaligned AGI that wants to kill us all, then it’ll be doing the thing where it releases 10 of these diseases all very quickly, and this is not going to be really sufficient to protect us, or we won’t be able to dig our way out. Why do you think of that as kind of perhaps too extreme in either direction, and maybe there’s a middle ground? Andrew Snyder-Beattie: On the AI making defences, I think one weakness of this argument is that a lot of the defences might be physical manufacturing. Maybe you just need to physically create a lot of masks, you need to physically create a lot of air filters and chemicals and stuff. And AI benefits a lot of different things, but it kind of is biased towards things that are information-heavy versus physical manufacturing. Rob Wiblin: Especially early on. Andrew Snyder-Beattie: Exactly. So if you have a number of different AIs, some of which are trying to protect you and some of which are trying to hurt you, I worry that the AIs that are trying to hurt you might be able to generate lots of biological weapons before the AIs that are trying to help you can generate physical stuff that can actually protect you — just because it’s faster to generate a small snippet of biological code than it is to mass produce protective equipment and protective buildings and structures and stuff like that. That’s one argument for why this still could be scary. And interestingly, I think this could still be scary even if you have smarter systems that are on your side, even against slightly dumber systems — because it might be that the slightly dumber system is still able to make arsenals of biological weapons, whereas the smarter system has a harder task of physically manufacturing lots of things. Although if we’re right about the four pillars, then maybe we only need human-level technology to manufacture enough defences, and it’s just a matter of getting those in place really quickly. I think that’s why I’m excited about doing it on such a short timeline. Rob Wiblin: I’m kind of imagining a superintelligent, aligned AI. We go like, “What is going to be our plan for protecting ourselves from diseases?” And it’s like “Well, obviously you should wear masks. I really can’t do much better than that for you guys. We’ll do the medical stuff later.” Andrew Snyder-Beattie: “Wish we had made the masks. That was bad.” And then in the opposite direction, I think you could argue that in a lot of the scenarios the superintelligence or whatever is going to have plenty of options. But imagine you’re a misaligned AI, and you’ve managed to escape from the laboratory, but you’re not wildly superintelligent. You have all these humans that are doing their thing, and you also have other AIs at other labs that are getting developed that are going to be more powerful than you in the near future. So you might be willing to take a lot of risky gambles to try and gain power or otherwise do things. What you might want to try and do is create some way of surviving, even if most of the humans have been killed, and then release a lot of biological weapons in order to knock down all of humanity and stop the other AI labs from doing it. Even if it has a relatively small chance of success, this might be the thing that you might be incentivised to do. So I don’t think we should be assuming it’s all or nothing. I think we should be working on the margin, where there could be AIs that are in positions where they’d want to do this. I also think how much probability do you put on a multipolar world, where there are lots of different AIs that have lots of weird motivations, or lots of people with lots of motivations that are using AIs, and so — Rob Wiblin: Well, also just AIs that aren’t necessarily superintelligent, but just are super erratic. That’s one thing that we’ve seen with a lot of models lately is that they just aren’t doing necessarily the thing that their operators want them to do in all kinds of crazy ways. And they can be given random instructions, and the open-weighted ones, any rando can potentially give them a random goal, alter them to have a different mission than the original one, and then give them a bunch of compute and see what they can do. It could be much more random. Andrew Snyder-Beattie: Yeah, like ChaosGPT. People are like, “What’s the probability that there’d be an AI that’s intentionally trying to kill humans?” People just make that. People just make that for fun. It’s crazy. Rob Wiblin: Yeah, we should expect it. For people who don’t recall ChaosGPT: in the early days back with GPT-4, someone immediately made a model whose goal was to cause human extinction. And it was a bit comedic on some level, because it wasn’t able to do it. But at some point, people might do that with an open-weighted model that was actually in a better position to do some real harm. Yeah, I interrupted. Andrew Snyder-Beattie: No, we need to defend against that. That’s crazy. And again, the number of people that are working to prevent these worst-case scenarios is tiny, like fewer than 100 people. Rob Wiblin: Yeah. You’ve talked about this “window of vulnerability.” Which is the idea that, as technology advances, and I guess AI and all kinds of medical technology advances, there’s a window between when it’s possible to create an incredibly dangerous biological weapon — which perhaps that window of vulnerability is already open — but we haven’t yet reached the point where the defensive technology has been created and scaled up, such that basically it’s no longer possible for those really to succeed. And we’re basically just trying to bring forward the point at which the defensive technology closes that window of vulnerability. Andrew Snyder-Beattie: Exactly. That’s the plan. We want to close the window. We want to close the window quickly. And the way we’re going to close the window is these four pillars. That’s the hope. Rob Wiblin: It sounds like you could potentially do quite a lot of these different pillars without necessarily having to have the government do it, more or less. You can distribute the masks; it’s not so expensive. You can disseminate information about how to harden your home. The Nucleic Acid Observatory doesn’t require the government to operate it. Andrew Snyder-Beattie: That is one of the advantages of the four pillars plan. If the governments are doing it, that’s the best world to be in. And I think governments ought to want to do things like this. But yeah, I think a group of philanthropists could basically do this, and potentially do it well. Rob Wiblin: Up until now we’ve been thinking basically exclusively about biological catastrophes that kill human beings. Sometimes they’re spread from human to human, sometimes they’re spread from the environment to human beings. But we could imagine other ways that humans could go extinct from biological catastrophes — like if you had something like mirror bacteria that killed all of the crops, or some other super disease that destroyed agriculture more or less, such that we just couldn’t feed ourselves and things progressively fell apart. And people have theorised about how you could have some sort of bacteria or virus that killed some natural environmental process that we relied on to survive, like not having enough photosynthesis to create enough oxygen for us to breathe. Why are you not spending very much effort on those potential threats? Andrew Snyder-Beattie: The good news is I think these two risks are substantially lower than the others. As you mentioned, basically you can divide all biological risks into one of three categories: things that target the environment, broadly speaking; things that target agriculture; or things that target human bodies. Interestingly, when I first started at Open Phil, I was quite worried that maybe we would put all of our investment into things like good protective equipment and better vaccines and stuff like that, and then it turns out that this whole time we should have been worried about an agricultural threat or something like that. So we had a researcher look into this, and the task that we gave them was: imagine a worst-case scenario. One way you could approach this research question is you could try and generate a list of all the horrible things that you could do to agriculture, and then look at the list and think, “Is this that scary or not?” But that generates a lot of information hazards, so you don’t necessarily want to be doing something like that. Instead, there could be another way of approaching the problem, which is: just go ahead and assume a worst-case scenario, and then ask how many people could we save in that worst-case scenario. And the worst-case scenario that we gave them was: imagine that all crops die instantly at the worst possible time in the harvest cycle, and that you can never grow crops ever again. And we initially gave this research prompt as that’s the most extreme example; obviously there’s no way we could survive that, so then we’re going to titrate and make this scenario slightly easier each time to then figure out what the threshold is. It turns out that even in that worst-case scenario — where all the crops die instantly and you can never grow crops ever again — you can feed the entire US population for… guess how long? Rob Wiblin: I know the answer, but it’s more than you would think. Andrew Snyder-Beattie: It’s more than you would think. People typically say, how much food do we have stockpiled? And we have about 18 months of food stockpiled. But the actual answer is 500 years. The way you do this is you have basically bacteria that eat natural gas. And using only about 15% of US natural gas production, about 6% of US electricity production, and about $200 billion [Correction: this would likely be closer to $500 billion. —ASB] worth of infrastructure, we could feed every single person in the US for 500 years. And obviously you could increase the capacity of that if you wanted to — you know, feed the rest of the world as well. And that’s the worst-case scenario, where all of the crops instantly die, no time to adjust, you only have the food in your stockpile as an adjustment period. Rob Wiblin: What does this actually look like though? Andrew Snyder-Beattie: I mean, it’s not a pleasant future. You know, you’re going to be eating bacterial sludge. Rob Wiblin: I think I’ve played this computer game. So you’re producing bacterial sludge, basically. Andrew Snyder-Beattie: That’s right, yeah. Rob Wiblin: All of the plants have died. We’re not going for walks outside. Andrew Snyder-Beattie: That’s not a great future. Rob Wiblin: But what we’re doing is we’re getting natural gas out of the ground, and we’re bubbling it through these tanks with bacteria living in it. I guess this requires some electricity. There’s these specific bacteria that eat natural gas, and then we filter out the bacteria from the water, and that’s what everyone is eating. Is this a complete diet? Andrew Snyder-Beattie: It is, yeah. Or I think there might be some minerals and vitamins you’d then supplement with. But basically it has all your carbohydrate, all your macros, totally accounted for. And interestingly, this isn’t some hypothetical technology. They already use this to feed fish. So there are natural gas plants that then shunt off the excess natural gas, and then it’s used to produce fish food. So this would just be scaling up an existing technology. Rob Wiblin: Do fish like it? Andrew Snyder-Beattie: I don’t know. It’s factory farmed fish. It’s not pretty. But yeah, so this would be scaling up basically an existing technology. I think there are some things that researchers or philanthropists could do here. You could engineer strains to be nutritionally complete for humans rather than nutritionally complete for fish. I think there would be some adjustment period, and maybe doing a bit of that work ahead of time would be slightly better. But also, keep in mind this is the worst of the worst-case scenarios. Rob Wiblin: Normally we would have more of a transition. Andrew Snyder-Beattie: Yeah. If you think about all the stuff that ALLFED is looking into, you can turn trees into sugar and do other interesting things to get you stopgaps. And we were assuming the worst of the worst-case scenarios. Rob Wiblin: I think the common factor between all of these plans to feed people without agriculture is that humans actually consume shockingly little raw energy in some sense. You’re saying we could feed everyone using 15% of the natural gas — so almost all of the raw energy that we’re burning is not going into human bodies; it’s going into cars and factories and so on, and producing electricity. So there’s an awful lot of chemical energy lying around somewhere, basically, that we could repurpose for feeding humans if we’re savvy enough. But if I imagine this actually happening, I don’t think that we would feed everyone very quickly, because I don’t think that we would plausibly repurpose all of these materials to produce this bacterial sludge quite quickly enough. Do you actually think that conceivably could happen? Andrew Snyder-Beattie: Well, there is a lot of food that’s in the supply chain, and a lot of that food is going to feed animals. So in a catastrophe — Rob Wiblin: How much animal feed is there? Andrew Snyder-Beattie: I forget, but if you add it all up, it’s something like 18 months of food in the US. And that assumes that the catastrophe happens at the worst possible time in the harvest. If it happens after the harvest, you get more like two years. [Correction: this might actually be closer to six years! —ASB] Rob Wiblin: Is this something specific to the US? I guess it’s a real agriculture powerhouse. Andrew Snyder-Beattie: Unfortunately, the US is a bit of an outlier on how much food we have stockpiled. China has also stockpiled a large amount of food. In many European countries, it’s closer to six months or something like that. And the developing countries… Rob Wiblin: They don’t have as much natural gas either. So the US is in a pretty beneficial situation. Andrew Snyder-Beattie: Yeah, yeah. Rob Wiblin: But can you just take animal feed? Why do we have 18 months of animal feed? Andrew Snyder-Beattie: It’s basically in the supply chain, and because the animals consume a lot more and then you consume a lot less of that. It’s just corn, soy, wheat, you know. Rob Wiblin: I thought that there were different kinds of corn that animals eat than humans, but I suppose we would just cook it and we’ll find a way to digest it one way or another. OK, so you just think that the agricultural thing is not a problem? Andrew Snyder-Beattie: I don’t want to say not a problem. In a mirror bacteria scenario, where you have to do this and you have to protect the people, that’s a pretty grim situation to be in. But to be clear, there are a lot of other arguments for why agricultural biological threats are going to be less severe than the ones that are targeting humans. Like you can pivot which crops you develop; you can genetically engineer crops to be resistant — whereas you can’t genetically engineer new humans quickly or whatever. Rob Wiblin: You’re stuck with the humans you have. Andrew Snyder-Beattie: Yeah, you’re stuck with the humans you have. So I think there are a lot of other haircuts against the agricultural catastrophe argument. But I think this is an interesting example where, if you think about a risk window for agriculture, we might already have exited the worst-case scenario risk window of agriculture — and maybe we’re resilient to even the worst-case scenarios. Rob Wiblin: Yeah. If one country has lots of food and everyone else is kind of starving, I think a case of a disaster situation where people usually do turn to violence is actually long-term sieges where people start starving and then they really do start basically turning on one another in order to get as much food as they can. So if the US had lots of food, but everyone else literally was dying of starvation, I guess the US is hard to attack, but I think you could see international relations fraying. Andrew Snyder-Beattie: Certainly. I think that’s why ALLFED is doing interesting work here, where they are actually just trying to feed everyone, basically. There are just a lot of things you can do on top of the natural gas as stopgaps. And that was just with 15%, so the US could produce a lot more to feed the rest of the world as well. Rob Wiblin: ALLFED, as you mentioned, have a whole bunch of other ideas about other sources of calories that we could potentially use, so we wouldn’t be putting all of our eggs into the fish feed basket. What about the environmental [biorisks]? To be honest, I actually don’t even know: what are the environmental disturbances that people are envisaging? Andrew Snyder-Beattie: You could hypothetically imagine something that somehow shut off photosynthesis, or you could imagine how people talked about mirror cyanobacteria sucking out the carbon and creating an ice age that makes agriculture really hard, and stuff like that. Rob Wiblin: Explain that like I’m five. Andrew Snyder-Beattie: If you have a mirror bacteria, the viruses are not going to be attacking it. So if it’s hanging out in the ocean, it’s going to be drawing down carbon maybe at a faster rate than other organisms, because it’s not getting digested, basically. Rob Wiblin: OK, so background information here is that there’s cyanobacteria in the sea, kind of everywhere. They’re a massive driver of photosynthesis. So they’re doing a lot of work to draw carbon dioxide out of the air, turn it into oxygen. But their population levels are regulated by the existence of viruses. Andrew Snyder-Beattie: Viruses, things that eat them… Rob Wiblin: OK. But if you had mirror cyanobacteria, then they would have no natural predators. All of the viruses that have evolved to control populations or that happen to control populations of normal cyanobacteria don’t even exist. They would just proliferate to an extraordinary degree, and they would suck all — Andrew Snyder-Beattie: And the carbon that they draw down would not be digested and then spat back out. So it would sink to the bottom of the ocean at a higher rate. Rob Wiblin: Because it would be in sugars that have the opposite handedness and no one could digest it. Andrew Snyder-Beattie: It could be a concern, but I don’t think it’s a real existential concern. Basically all these scenarios take way too long. Rob Wiblin: Even if you had the mirror cyanobacteria throughout the oceans, it would just take decades? Centuries? Andrew Snyder-Beattie: Yeah, it would take many centuries, and there are pretty obvious countermeasures. You could just make a mirror phage. You could do other things. You could do basic geoengineering stuff. You’d have hundreds of years to deal with it, and there are pretty obvious countermeasures. And just more generally, it takes a really long time to mess with the Earth’s environment. So even if you had a magical button that you could hit to stop all photosynthesis — which is the worst possible thing imaginable — you’d still have 1,000 years of oxygen just hanging out for us to figure things out. So even in the worst of the worst-case scenarios, I think these scenarios are quite unrealistic. Rob Wiblin: All right. We’ve gone pretty deep on the plan and various different objections that people could raise to it, but I think I’m reasonably sold. I’d really like to see a bunch of this happen, and I imagine many listeners feel the same way. As we flagged, you’re hiring hand over fist, trying to get some really talented people to lead on each of the different pillars. Maybe we should go through all of the most important roles that you’re hiring for at the moment. Andrew Snyder-Beattie: Sure. Most important roles right now include grantmakers at Open Philanthropy. I’m growing my team, trying to figure out how to get a bigger team of people to basically deploy funding. So if you are interested in a grantmaking role to deploy tens of millions of dollars — notably not just for the four pillars plan, but also across a wide range of different biosecurity issues — I want to hear from you. Fill out the online form. Rob Wiblin: What sort of person is a good fit for a grantmaker role? Andrew Snyder-Beattie: People that are entrepreneurial, who want to talk to as many people as possible, collect information. There’s a common misconception that grantmaking roles involve just reviewing lots of applications that come in and then just giving the thumbs up or the thumbs down. That’s not at all what the role is like. And that’s especially not what the role is like in an area where the field is so small and you kind of have to create the things that you want to see. So I would describe most of the grantmaking roles at Open Philanthropy as being more similar to venture capital or headhunting — where you want to go out and find people and get them working on the most important problems. Rob Wiblin: Do people need any particular bio background or any particular domain expertise? Andrew Snyder-Beattie: This is a great question, and I think this is a really common misconception that people need a really strong biological technical background to contribute in biosecurity. It can be very helpful. Half of the people on my team have biology backgrounds, PhDs, but the other half don’t. I majored in economics, and a lot of the people that got into the field did physics, or were just entrepreneurs. Some of the highest impact people, one of them was a software engineer at Amazon and then ran a startup. So a lot of people from a lot of different backgrounds can contribute. You don’t necessarily have to have a biology background. One way of arguing this is: it’s not like you need a physics background in order to reduce the probability of a nuclear war or something like that. I think there are a lot of things that people can do, even if you kind of abstract away the biological details. Rob Wiblin: OK, so that’s being a grantmaker on your team. What’s the next most important role you’re trying to fill? Andrew Snyder-Beattie: On the personal protective equipment plan, we’re still in the stages of figuring out whether or not this is something we want to go big on, and thinking through how that could possibly work. I still think there needs to be a bigger team of people working on running all those details to ground and figuring out, like, should the mask design be a certain way, and how low can we get those costs? We have an interim CEO and she’s doing a great job, but I think we need a more permanent CEO for that. We also need people who work in manufacturing, product design, communications. There’s a lot of work on the PPE project. Right now that project has maybe three full-time people on it, and this might be one of the most important projects for humanity. So if you’re interested in that, we’d love to hear from you. You should fill out the online form. Rob Wiblin: Yeah, we don’t often have so many roles for people who are interested in manufacturing or logistics, that kind of thing. So if you’ve been listening to the show and thinking, “I wish that there was a role for me,” I think this is the place for it. Andrew Snyder-Beattie: Yeah, totally. The person actually did her engineering degree in mechanical engineering, and then ended up doing AI safety. She ran METR briefly and now she’s working on the PPE thing. Rob Wiblin: So I guess there’s improving the design of the mask, making it a whole lot cheaper, figuring out how you can manufacture it at a bigger scale, and then also figuring out how you would deliver it in the worst-case scenario, and how would you get it out there everywhere? Andrew Snyder-Beattie: Exactly, yeah. This is a very concrete problem. Maybe one thing I’ll say more generally is that biosecurity really lends itself to people that want to take a very concrete physical problem and make progress on it. I do think this is maybe in contrast with a lot of the AI work, where a lot of it is quite hard to reason about, and it’s not clear. A lot of it’s not even clear whether or not you’re net positive or something. And I think biosecurity, the problem is just in some sense a very simple problem: there could be a thing that’s spreading, and you want to stop that spread, and you want to erect physical barriers and physical sterilisation. So it’s in some sense a very straightforward strategy, so it’s more easy to measure your progress and figure out, are we actually cutting these risks? Rob Wiblin: Yeah. Where can people learn more about that role? Andrew Snyder-Beattie: Again, fill out the online form. All of this will just be in the form. Rob Wiblin: OK, what’s the next one? Andrew Snyder-Beattie: As I mentioned, glycols could be a really interesting strategy. You know how many full-time people are working on thinking about the supply chain of that and how to distribute it? Rob Wiblin: I’d guess zero? Andrew Snyder-Beattie: Correct. There are some part-time people looking at this, and that’s how we’ve run these initial numbers. But I think this deserves one full-time person who’s going to be thinking about this and working on it. Similarly for the air filtration: thinking about ways of improvising that and thinking that through, that’s another example. There are zero full-time people working on that. We need to actually validate that, think it through, figure out if this actually works. And those are roles where eventually they could be leading teams if they’re doing a good job. Rob Wiblin: OK, what’s the next one? Andrew Snyder-Beattie: On the medical countermeasures strategy, we have a researcher at Open Philanthropy, and she’s been working on this a little bit. But I think we need more intellectual effort here; we need more people thinking about what the medical countermeasure strategies should be. So also researchers thinking about that would be good. Maybe I will just zoom out and say that in general I think there are a lot of roles in biosecurity. We also have scholarships, we have fellowships. If people want to get involved in the field, that’s a really good way to start. We offer career transition grants for people that want to get into the field and don’t quite know where to start. And those have been very successful. Some of the top people in the field came in through that pathway. So fill out the form. Rob Wiblin: It sounds like the medical countermeasures one is perhaps a role where you would benefit from having some domain expertise, because you’re often dealing with quite technical biomedical questions about what’s viable and what’s not? Andrew Snyder-Beattie: Yes, absolutely. Rob Wiblin: And on the biohardening, that sounds like maybe a role for someone who’s more like hands-on engineering side of things. Andrew Snyder-Beattie: Hands-on engineering would be good, absolutely. And just experience in managing teams, managing organisations, pulling projects together. Rob Wiblin: Yeah. It sounds like in general you want someone who has a lot of initiative, and is going to be willing to go where no one has gone before on some of this stuff. Andrew Snyder-Beattie: Yeah, absolutely. Entrepreneurial people. Rob Wiblin: If someone didn’t feel like they were suitable for those roles, are there any other things you could point them towards? Andrew Snyder-Beattie: I think there are a number of organisations doing really important work. The early detection system that I was describing still has maybe 10 to 12 people working on it, and I think they’ll be hiring for more roles soon. So more people with bioinformatics experience, wet lab experience, even just logistical or government affairs experience. Generally speaking, policy kind of cuts across all these different areas and is really important. Rob Wiblin: A lot of people who would be up to doing these roles might also be considering going into AI policy, AI technical work, or some other AI-related project, which is such a topical issue at the moment. Do you think that this work is competitive or maybe even more impactful than AI-related work? Andrew Snyder-Beattie: Yeah, I think there’s a strong argument that it could be more impactful. That argument is simply: it’s more neglected; there are far fewer people working in this. In most of the subareas it’s like three to four people. It’s also very tractable. We have a basic plan that I think could cut the risk substantially. Whereas in AI I think it’s a lot less clear how successful the interventions are going to be. And finally, I don’t think it’s wildly less important. If you think there is a 1% to 3% chance of catastrophe causing an existential risk in bio versus AI, it’s probably within an order of magnitude. So the neglectedness and tractability arguments can easily mean that on the margin people are better off working in biosecurity. Rob Wiblin: And personal fit as well. If their main focus is logistics, this might well be a better fit. Andrew Snyder-Beattie: Yeah, absolutely. Rob Wiblin: Let’s talk about a bunch of tactical career stuff, opinions that you formed over the years. One you wrote in your notes is that you think people too often do work just expecting that someone is going to be later in the pipeline who’s going to be able to make use of it — and often this is just completely delusional. Tell us about that. Andrew Snyder-Beattie: So I think there are two things here. One is like, if I think about my own early career, I came across some of these existential risk arguments and I was thinking, “Oh my gosh, that’s really important. I should focus my career on that.” And then I ended up doing some very silly things. I was donating to asteroid-deflection charities, which was, in retrospect, not very effective. And I was doing my master’s thesis on the Great Filter and the probability of finding alien life and accounting for different things, because maybe that could possibly help. I think I just had this mindset of, “There are a tonne of other people focused on existential risk that are way smarter than me, and they’re going to go off and solve the problem, and I’m just going to have this little tiny drop of knowledge that I’ll put into the ocean of humanity’s knowledge to solve these problems, and that will be my contribution.” I don’t think I fully realised exactly how outrageously neglected these problems were, and how if I just put in a bit of effort, you could end up in a very important position with a lot of responsibility. Which in some sense is terrifying. But I think the other side of that is that I think people can have a lot of impact if they really take ownership of a problem. In terms of passing work off, I see this a lot in people who think that they’re going to influence policy by writing a report that no one reads, or that they’re going to do research on a problem with the hope that people are going to read the research and use the research. One interesting thing that I found is I actually think the quality of the research that me and my team is doing is actually a lot better because we’re making decisions about how to allocate money. So we have these very high-stakes decisions, and the research that we do is directly informing that decision. I think what you generally need is a really tight feedback loop between the decision that needs to get made and the research that’s informing that decision. And if that feedback loop is broken, it’s very easy for people to do research that’s quite disconnected from important decisions, or decisions that people are actually making in the world. Rob Wiblin: Why do you think it is that people find it so natural, the idea of, “I’ll do some precursor work and then expect that someone else is going to pick it up and make use of it” — even when they have no idea who those people are, or they haven’t even gone to check whether they exist? Andrew Snyder-Beattie: I’m not sure. One hypothesis is that a lot of them have not been in decision-making roles, so they don’t have a good model of what’s needed to make those decisions. I also think academia can sometimes instil this habit, where to do well in academia you have to be sort of working on the trendy topic and making a contribution there — and you can make a contribution in an increasingly sub-sub-sub-specialised field, where you are just adding a little drop of knowledge into a growing ocean. I think on certain topics it’s very important, but if you’re really trying to do good in the world, you want to be finding things that are extremely neglected, where in fact there might not be any good work on it yet. So I think that’s another big difference. Rob Wiblin: Yeah. I think among the Silicon Valley entrepreneur crowd, the conventional wisdom is that there’s always far more promising, interesting opportunities for new businesses than there are entrepreneurs who can give them a real go. I’m not sure what the underlying reason for that is. It’s possible the conventional wisdom is wrong, but I think it’s probably right. The world is big, but there just actually aren’t that many people who are trying to start new businesses, actually making a product that has never existed before. That is kind of an abnormal pursuit, and many people in the world are just not in a position to do that, so you shrink the people a lot. Then I guess technology is always changing; the frontier of what things we could have a crack at is always potentially quite wide, and it’s hard to tell ahead of time what is going to work and what is not going to work. So it could just be the case that it’s much easier to come up with ideas than it is to come up with an entrepreneur who’s going to actually give it a sufficiently solid go to tell whether it will work or not. Andrew Snyder-Beattie: Yeah. This has been my experience trying to recruit for a lot of these roles. Typically I will tell someone, “I think you should take this role. I think you’re a really good fit.” And they say, “Well, obviously there’s the second person and the counterfactual — because who’s the replacement?” Often there’s no replacement: there’s one person who I think can do the job. And people tend to be surprised by this. Rob Wiblin: Do people tend to assume that the government is doing more than it is? That there’s so many people in the government on national security, that wouldn’t they just be on top of all of this? Andrew Snyder-Beattie: Yeah, certainly. I mean, I believed this. Before COVID I assumed that the government would just be really competent at handling COVID, and the CDC would just have it sorted. And I was pretty shocked. Rob Wiblin: What do you think is going on? I guess there maybe isn’t the budget. I guess these groups aren’t always stress tested to tell whether they can really do the thing. Especially if you’re dealing with a risk that only occurs rarely, or maybe it will never occur over someone’s lifetime, it’s really hard to know whether you’re on top of it or not. Andrew Snyder-Beattie: Yeah, totally. It’s much easier to deal with problems that are more chronic and ongoing where you can get feedback loops. If you look at a lot of the public health agencies, they’re focused on things like tobacco use and HIV and stuff like that. I think they’re doing good jobs, but it’s just very different than a fast-moving pandemic where you have to make decisions under uncertainty very quickly. Rob Wiblin: Yeah. You think another mistake that people make, or another place where people struggle to make the right decisions, is choosing something that is ambitious enough that it has a big impact, but isn’t so ambitious that it’s kind of beyond what they can actually accomplish. Can you explain that? Andrew Snyder-Beattie: Yeah. I think sometimes people make the mistake of just taking a subproblem of a subproblem of a subproblem, and then just kind of assume that if they do that, there’ll be someone to hand the baton to and that will be impactful. And I think people can end up wasting their time there. But I also think sometimes people in the EA community will dream up these things like, “We should study the grand strategy of how countries interact, and then that can help inform how we should think about AI.” I think that can be sometimes interesting, but usually I’ve been quite disappointed at actual concrete results that come out of thinking like that. I think the more promising thing, at least the strategy we’ve pursued in biosecurity, is to try and carve off problems, like subcategories of things — like, “Let’s just limit our focus to human transmission. Can we solve that subcategory? What do you need to solve that subcategory?” — and then have an ambitious plan to solve that subcategory or things like that. I think this is a general tactic that maybe more people should be considering. Rob Wiblin: How do you strike that balance well? Is it something that you think, “If I built a team, we could kind of handle this. We could come up with an answer, we could come up with a project that would solve it basically, without having to necessarily rely on other people to do all that work for us”? Andrew Snyder-Beattie: Yeah. I think you can tell if you’re in the sweet spot between being too narrow and too broad if you’re actually making progress — like if the research is actually tractable, if that makes sense. Maybe my hot take is that effective altruism really pushes people to think about what’s important. I think that’s really good, but interestingly, people just kind of forget the tractability. So people spend a lot of time just being like, “Well, AI is the most important thing. I’m going to do my career in AI” — and they end up just kind of forgetting about tractability, and a lot of people end up in careers that might not be necessarily having that much impact. Rob Wiblin: Do you think that’s still the case in AI today? It feels like 10 years ago it was harder to find direction, but… Andrew Snyder-Beattie: I’m not sure. I don’t have a tonne of visibility on this, but definitely that would be my hypothesis. Rob Wiblin: I would have thought that technical AI safety is kind of clearer what the projects are. Andrew Snyder-Beattie: Yeah, I think it’s more tractable now than it was 10 years ago, or something like that. Rob Wiblin: But even if you’re doing a sensible project in AI policy or AI technical work, it still is overwhelmingly likely that it won’t matter at the end of the day, that it will end up being irrelevant. You got unlucky. Andrew Snyder-Beattie: Sure. To be clear, hopefully none of the stuff that I’m doing is going to matter, because there’s not going to be a biological catastrophe. So I think you’re always operating under uncertainty. I also sometimes get the feeling that because effective altruism is focused on the most important things that everyone kind of herds towards the most important thing. It’s maybe cynically a bit like five-year-olds herding to a soccer ball. It’s like, “That’s the most important thing!” and then everyone goes to it. And then there’s a lot less thought to where should you be positioned? Should there be other important, but maybe slightly less important things to be focused on? Should people be spread out a bit more? My guess is that yeah, that is the case. Rob Wiblin: Yeah. If people were biased towards crowding or herding into the most popular thing, what would be the dynamic there? Of course, the social dynamic, that maybe you want to be seen as part of it. Andrew Snyder-Beattie: I think that’s definitely part of it. It’s just easier to have a path that other people are helping on. Rob Wiblin: I guess it’s that there’s no coordination mechanism to ensure that the ratios across all of the people who care about existential risk are sensible. You can imagine everyone kind of chooses the thing that at that point looks most important, but then they don’t have any sense of like, is anyone going to do the second and third and fourth most important thing? Andrew Snyder-Beattie: Totally. Yeah. So this is my job. Here I am. Spread out, figure things out. Rob Wiblin: Is it possible that you’re a little bit biased towards wanting people to go into your area, take good jobs? Andrew Snyder-Beattie: Possibly. Like even climate change, I think the conventional wisdom within effective altruism is that the whole world is focused on climate change, so on the margin, one additional career focused on climate change is not going to have that much impact. And I think at a high level, that’s correct; it’s better to focus on these more neglected risks. But Hannah Ritchie wrote this great book looking at climate change through a more effective altruist lens, thinking about what’s actually effective? What’s not actually effective? Why did it take 20 years for someone to write that book? Maybe an EA could have written that book a lot earlier, because if you apply the importance, neglected, tractability framework on other problems, you can make a lot of progress. So yeah, I actually think more people should be doing weirder, more unique things. Rob Wiblin: Yeah, I had an interview with Johannes Ackva, who was working on effective-altruist-minded climate change grantmaking. And their team, as soon as they started looking at that question, were coming up with completely different proposals to what was being funded by everyone else. They were saying, “From our point of view, we don’t want to focus on solar and wind at all, because if that does solve the problem, then basically it’s already handled. So we need to be basically concentrating on the specific scenarios where that does not pan out, which is actually reasonably plausible, and then thinking, well, what would work in that case?” I guess they thought it was a question that was basically not being addressed by any other group almost anywhere. Andrew Snyder-Beattie: Yeah. How many full-time people did they think were working on this? Two or something? Rob Wiblin: I wouldn’t want to put words in their mouth, but I think it was very small, or that was a very niche issue at best. Andrew Snyder-Beattie: I think this is the case with many things in the world: more things are outrageously neglected than people think. On most of these given projects there’s a big broad community of people, but the number of full-time people that are really ruthlessly focused on the problem, it’s usually between two to five people. So if you’re interested in any of these problems, you could fill out the form. We desperately need you. Rob Wiblin: All right, so a final question is one that came in from the audience. Maybe it could be multiple questions. People were curious to know, from a just selfish point of view, what should people do in order to protect themselves or give themselves a greater chance of surviving a really bad event like this? To really solve it, we kind of need the societal response — because ultimately, if the rest of society falls apart, then you’re probably toast — but is there any useful stuff that you do personally to prepare? Andrew Snyder-Beattie: I would recommend getting a good elastomeric respirator. I like the 3M ones. I also have the EM Pro. Then I would also get enough food so that you can socially distance in a pandemic without needing to go outside for, say, three months, something like that. Rob Wiblin: Any advice on what food? Andrew Snyder-Beattie: Yeah, dried food. The shelf stable stuff that lasts for a long time seems like a pretty safe bet. Rob Wiblin: Do you stockpile water? Andrew Snyder-Beattie: I actually don’t. I actually do surprisingly little prepping myself. So if the catastrophe is bad enough that the water gets shut off or the power gets shut off, I would get on a bicycle and get out of town, but… Rob Wiblin: Should people have a plan for getting out of major cities if that’s where they live? Andrew Snyder-Beattie: Seems reasonable, but I wouldn’t necessarily trust my prepping advice. Rob Wiblin: You’re trying to save everyone, not an individual. If we were at the beginning of a biological catastrophe, like some very nasty bioweapon that was released, how do you think you’d be spending your time? Andrew Snyder-Beattie: Probably on the phone, trying to get people to care. I was doing this during the early days of COVID. In February, I was calling up these government lobbyists, trying to get people to pay attention. People were like, “You don’t want to be crying wolf.” And it was just crazy to me. Rob Wiblin: Did people ever apologise? I don’t know whether you called them back to request an apology, but sounds like no. What do you think you would be trying to push them to do? Just wake up and take the thing seriously? I mean, if the bioweapon was bad enough, maybe people would react very differently. Like you were saying, COVID in some sense was in this sweet spot, where it wasn’t quite bad enough that people really had to react. Andrew Snyder-Beattie: Yeah. There are tonnes of different things, and there are different pandemic playbooks and different things that the government should be doing. So a crash programme on medical countermeasures, like making sure the hospitals are prepared. There’s tonnes of stuff. Rob Wiblin: Implementing all of the plan that we’ve talked about. Andrew Snyder-Beattie: There’s a lot more stuff that would need to be done other than the four pillars, but I think the four pillars are the basic building blocks that other governments could build on. Rob Wiblin: Well, hopefully the response is swifter next time. My guest today has been Andrew Snyder-Beattie. Thanks so much for coming on The 80,000 Hours Podcast, Andrew. Andrew Snyder-Beattie: Thanks for having me. It’s been fun.