Making Sense - Sam Harris - Special Episode： Recipes for Future Plagues

00:00:00.000 Welcome to the Making Sense Podcast. This is Sam Harris.

00:00:25.180 Well, Russia has invaded Ukraine, so we have the first major land war in Europe in decades.

00:00:38.580 So, that seems like a very big deal. It certainly deserves its own podcast at some point. I think

00:00:45.780 I'll wait to see how things evolve for a little while here. It remains to be seen how bad this

00:00:51.800 war will be and what else might happen as a result. So, I will reserve comment at this point,

00:01:00.680 apart from echoing the nearly universal sentiment that Putin's actions are despicable,

00:01:08.560 as is the support for him that came dribbling out of the mouth of our former president.

00:01:14.760 Anyway, as chance would have it, the topic of today's podcast is also scary. This is another

00:01:22.240 PSA, and in some sense it's a continuation of the podcast that my friend Rob Reed did for me

00:01:28.220 in April of last year. That episode was titled Engineering the Apocalypse, and it was a four-hour

00:01:37.040 examination of the threat that is posed to all of us by developments in synthetic biology.

00:01:45.380 In recent weeks, Rob discovered a specific threat along these lines that seems fairly imminent,

00:01:52.880 and he's tapped Kevin Esfelt to walk him through the problem. Kevin is a Harvard-trained biologist,

00:01:59.740 scientist, and he's credited as the first person to describe how CRISPR gene drives could be used

00:02:06.140 to alter the traits of wild populations in a way that was evolutionarily stable, and he is currently

00:02:12.860 a professor at MIT. As you'll hear, there's a call to action at the end of this episode, and the call

00:02:20.500 is to get the attention of USAID, which is currently running the program of virus hunting that poses such a

00:02:29.340 concern. Anyway, I won't say any more about this. Rob does an impeccable job at exploring the issue,

00:02:35.980 and I hope you will join me in making noise about it once you come to understand the nature of the

00:02:43.280 problem. Thanks for listening.

00:02:52.060 Today's conversation will be an episode of two different shows, the After On podcast, which I host,

00:02:57.560 and Making Sense, which my podcasting colleague Sam Harris hosts. We're doing this because we both

00:03:03.440 find the subject extraordinarily important, and also timely enough that we want to get it out there

00:03:08.820 fast. And since I've done a bunch of research that's connected to the subject and also know the guest,

00:03:14.360 Sam and I thought the quickest thing would be for me to conduct the interview, and then for both of us

00:03:19.100 to distribute it. For Making Sense listeners who don't know me, my name is Rob Reed, and I'm a

00:03:24.720 venture capitalist turned tech entrepreneur turned science fiction author turned science podcaster

00:03:30.140 turned venture capitalist once again, one who still podcasts and scribbles a bit on the side.

00:03:36.840 My voice may be familiar because I was on Sam's show several months ago, when we spent almost four

00:03:41.880 hours examining how very awful the next pandemic could be, and how we can prevent that awful pandemic

00:03:48.540 if we get our act together. I spent about a thousand hours, literally, including over 20 scientific

00:03:54.320 interviews, researching those subjects, and our episode had a fairly unusual format as a result.

00:03:59.780 I'm not a scientist myself, but over the past five years, I've gotten pretty deep into pandemic-related

00:04:04.980 issues. It started with research for one of my novels, which later led to writing a few articles on the

00:04:10.600 subject, which led to several episodes of my own podcast, then appearances on other shows, and then to a

00:04:16.500 TED talk that I gave right before COVID, and then to that big episode with Sam, and then quite a bit more.

00:04:22.660 Which brings us to today, and my conversation with MIT's Kevin Esvelt, a highly regarded evolutionary

00:04:29.220 engineer, who I met in the wake of some congressional testimony that he made on a subject that interests us both.

00:04:36.200 Kevin's going to present something that will probably shock you, which you may find

00:04:40.520 hard or even impossible to believe. But before that, he's going to lay down a fairly deep foundation

00:04:46.700 of concepts, definitions, and a bit of history, which should give you a sophisticated basis for

00:04:52.420 deciding whether you buy the fairly shocking points he's going to make in the second half of our

00:04:57.500 conversation. Now, I think it'll be useful for you to know the full context while he's laying that

00:05:02.920 foundation. So here comes a spoiler, which is that Kevin is going to argue that a small, new,

00:05:07.840 and very well-intentioned U.S. government program could, completely inadvertently, cause the deaths

00:05:14.360 of millions or even hundreds of millions of people or even more, despite what Kevin is certain are the

00:05:21.620 entirely good intentions of the people behind it. The program is called DeepVision, which I believe is

00:05:27.720 pronounced Deep Vision, and it's part of USAID, the agency which distributes most American foreign aid.

00:05:33.960 Kevin believes Deep Vision is on a path to posting assembly instructions for what we can only call

00:05:40.220 weapons of mass destruction to the open internet for anyone to download. Specifically, the genomes of

00:05:47.320 previously unknown pandemic-grade viruses that we have no defenses against. Viruses that tens of

00:05:54.620 thousands of people in dozens of countries could easily build from scratch, as soon as they're given

00:06:00.280 the genetic code. Now, if this all sounds a bit bonkers to you, I get it. I struggle to believe my

00:06:06.820 own ears at first. But if you listen to Kevin, you'll quickly realize that this isn't some COVID-era

00:06:12.520 conspiracy theory, even if he doesn't ultimately persuade you that things are as dangerous as he

00:06:17.660 thinks. I'll add that I wouldn't have interviewed him for this if I thought he had even the faintest

00:06:23.460 partisan agenda. I actually have no idea what party, if any, Kevin affiliates with, and I couldn't care

00:06:29.460 less. Because every point I've heard him make on the subject has been rooted in science, not politics.

00:06:35.440 Now, one thing that made this story especially hard for me to credit at first is that it's coming out

00:06:41.380 of USAID, which is dedicated to economic development and human flourishing in poorer countries, and which

00:06:48.380 puts its resources into so many great projects. Now, like any program that's put out hundreds of

00:06:54.100 billions of dollars across many decades, USAID has had its share of blunders and scandals and lousy uses

00:07:00.380 of funds. But I admire the agency on the whole. And that's partly due to personal experience, because

00:07:06.160 right after college, I spent a year in Cairo on a Fulbright grant and met lots of USAID people.

00:07:11.740 And they were funding things like irrigation projects, schools, and technical assistance in places that

00:07:16.740 really needed them. And they were all smart, committed, and working for a fraction of what

00:07:21.560 they could have made in the private sector. On top of that, USAID is run by former U.S. ambassador to

00:07:27.940 the U.N., Samantha Power. I can't say that I know Samantha, but we have a fair number of people in common

00:07:34.120 and had a couple great conversations at the TED conference, and I know how smart and ethical she is.

00:07:39.660 Plus, she literally wrote the book on not killing millions of people. It was a searing denunciation.

00:07:46.740 of genocide, which you may be thinking isn't the most controversial or risky position for an author

00:07:52.520 to stake out. But it won the Pulitzer Prize, so it was no puff piece. And there's zero chance that

00:07:59.140 Samantha would deliberately imperil millions of lives. All of this is to say that there are two sides of

00:08:05.580 this story, each of them held by very smart ethical people. And as you'll hear, the other side sincerely

00:08:11.780 sees Deep Vision as an invaluable weapon against pandemics. And I'm sure this is the side Samantha

00:08:18.220 hears from the most, because the program's creators work for her. And since Deep Vision is literally less

00:08:23.940 than a tenth of one percent of USAID's budget, and she basically inherited the program, this just isn't

00:08:30.260 something she manages directly. I should also point out that some of Deep Vision's plans could help us

00:08:36.300 fight pandemics. But after many hours of researching this, and talking to various people in academia,

00:08:42.900 philanthropy, security, and entrepreneurship, I believe that most of its agenda is appallingly

00:08:48.560 misguided. And I'm not going to hide that, because I'm not an actor, and it would be dishonest.

00:08:54.360 Now despite that, in the second half of the interview, I'm going to present all of the best arguments I've

00:08:59.440 heard or have thought of myself in favor of Deep Vision. Because this subject is way too important

00:09:05.040 for you not to hear the other side. I also want to convey why some very smart and ethical people

00:09:10.460 sincerely think this is a great idea, and show that thinking this doesn't mean that they're not

00:09:16.200 smart and ethical, because I'm sure that they are. Now, someone close to Deep Vision could surely do a

00:09:22.260 much better job of presenting this side. And in an ideal world, we might have structured this as a

00:09:28.160 debate between Kevin and a Deep Vision proponent. But I am an interviewer who has never once

00:09:34.440 moderated a debate. And though I could probably become a decent moderator with some experience

00:09:40.260 and guidance, this is happening right now. And we want to get the story out right now. Because the

00:09:47.320 program is so new, it's barely underway. In fact, it may not even have quite switched on yet. Which means

00:09:54.360 it should still be possible to alter its focus, or even redirect its entire budget to the countless

00:10:00.580 world-positive programs that USAID supports. And you may be able to help with this yourself,

00:10:07.280 as Kevin and I will discuss toward the end of our conversation. Which will start right now.

00:10:17.580 So Kevin, thanks so much for joining us today. And why don't we start out with a quick overview of

00:10:23.940 your professional background and what you spend most of your time doing in the lab these days?

00:10:29.280 Well, thanks so much for the invitation. I'm an evolutionary engineer and professor at MIT,

00:10:33.620 where my lab specializes in building synthetic ecosystems to what we call direct the evolution

00:10:39.200 of new molecular tools. And we also work a lot with communities to safely and controllably

00:10:44.560 engineer populations of wild organisms and associated ecosystems.

00:10:48.480 So we evolve cool tools in the lab. And then we talk a lot about whether, when, and how to use those

00:10:55.560 tools to change the environment. Evolution, of course, doesn't know anything. It's just a natural

00:11:00.760 process. And that means you can harness it to create things when you don't know how they work.

00:11:06.340 And that's true of a lot of biology, honestly. We still don't know the details of how it works,

00:11:10.740 of how to make things fold just the right way to cause something to happen within a cell.

00:11:15.720 One way to deal with this is to take something that does something reasonably close to what you want,

00:11:22.180 make a billion variants, throw them all at the wall and see which ones stick best.

00:11:26.520 Then take those, make a billion variants of those, throw those at the wall and see what sticks.

00:11:30.760 And do this over and over again until you get something that does exactly what you selected for.

00:11:35.620 And you do some related work with phages in your lab.

00:11:39.780 Yeah. So since we're interested in applying evolution to create molecular tools, it's useful to

00:11:45.780 work with the fastest evolving things in nature, which tend to be bacteriophages,

00:11:51.820 viruses that infect bacteria. And these viruses can replicate once every 10 minutes.

00:11:59.100 So that's a really short generation time. So what we do is we engineer the bacteria

00:12:04.780 so that the phage only gets to reproduce if it does the molecular trick that we want to see.

00:12:11.780 So we put a gene we want to evolve or multiple genes onto the genome of the virus. And we take away the

00:12:18.240 pieces that it needs to reproduce. We move those into the host cell. So the host cell will produce

00:12:23.460 this critical factor for virus replication in proportion to how well the virus manages to perform

00:12:30.940 the molecular trick. So that way, all of these populations of a billion viruses compete with one

00:12:35.840 another, creating mutations in those genes that we put on there. And the ones that perform the

00:12:42.080 molecular trick best produce the most offspring. And of course, this is a continuous process,

00:12:46.780 generation after generation, for hundreds of generations in the lab until it spits out the

00:12:52.560 thing that we want to see, all to create a better platform for making useful and hopefully safe biotech

00:12:58.680 tools. Now, just a couple clarifying questions. First of all, phages, they are viruses, but they're

00:13:05.680 minuscule and they can only infect bacteria. There's no way a phage is going to infect a human and

00:13:12.620 inflict disease. Is that right? That's right. Their machinery just would not function within

00:13:17.420 our cells. They're optimized for bacteria, very different context. Got it. And to edge toward what

00:13:23.180 I understand to be the practical application of this is you can train this ecosystem that you've put

00:13:27.580 together to create a very good version of a complex thing that you want with all these billions and

00:13:34.020 billions of shots on goal. And then having done that, you can make some useful product. In your case,

00:13:40.800 it's generally for pretty sophisticated biological applications, right? Like big pharma, biotechs,

00:13:46.720 other labs. For the most part, yeah, a platform for creating useful molecular tools can really

00:13:52.540 accelerate biotech research because there's no way to have a big impact like accelerating everybody

00:13:58.680 else's work and empowering others. So you've been doing this for quite a while, lots of experiments.

00:14:03.920 These things replicate every 10 minutes. You've got robotic tools accelerating it. How many genetically

00:14:09.880 distinct phages do you estimate you've produced in your lab since you started doing this work?

00:14:17.460 Genetically distinct is a hard question because we do crank up the mutation rate very high,

00:14:21.780 so they evolve much more quickly. Probably somewhere between 10 to the 13th,

00:14:26.180 maybe even 10 to the 15th at most. So that's a quadrillion.

00:14:30.340 I believe that would be, to contextualize it, the number of stars in roughly 5,000 to 10,000

00:14:37.040 galaxies the size of the Milky Way. So that is a lot. Now, you are often described as the inventor of

00:14:44.160 the gene drive. What is a gene drive?

00:14:47.060 Gene drive. So gene drive is a naturally occurring phenomenon that originated maybe a billion years

00:14:53.780 ago. And it happened when some genes realized that they could replicate more often if they changed the

00:15:00.260 odds that they'd get inherited by the offspring. So one way they do this, in organisms that have sex,

00:15:06.280 so they inherit one copy from the mother and one copy from the father, a gene can cut the chromosome

00:15:12.420 that doesn't have it, which causes a cell to copy itself over. And that means that instead of half

00:15:18.820 the offspring inheriting the particular gene drive system, all of them will. That's a huge fitness

00:15:24.600 advantage. You're going from half of offspring inheriting you to all of them, at least for the

00:15:28.360 case of heterozygotes that have one copy of each. And so gene drive systems can just sweep through

00:15:34.480 populations of sexually reproducing organisms very, very quickly.

00:15:38.120 Just to clarify, that means that the trait that this particular gene confers can go from rare or

00:15:45.500 even one-off because it was a mutation to saturating the population in a certain number of generations.

00:15:52.600 The trait changes. We don't end up getting clones of the individual.

00:15:57.300 That's right. And Austin Burt, who is one of the earliest gene drive researchers in the modern era

00:16:03.120 at Imperial College London, first proposed that we use these things by engineering genes to cut

00:16:09.480 sequences that we want in order to change the population. The problem is we really didn't have

00:16:15.500 the tools to make this happen. Well, I played a minor role in developing CRISPR, which is a genome

00:16:21.840 editing tool that is basically a set of molecular scissors that can be programmed to cut whatever sequence

00:16:26.940 we want. A few months after we had been one of the first groups to show that CRISPR worked in

00:16:32.540 mammalian cells, I was looking around outdoors at a bunch of ducks and I saw a turtle that day in the

00:16:38.640 water. And I was thinking, hey, are we ever going to edit wild organisms with CRISPR? And I thought,

00:16:43.300 probably not, at least not effectively, because when we edit an organism, we're changing it to do what

00:16:49.180 we want, which means we're diverting its resources away from survival and replication in nature.

00:16:54.000 And that means natural selection is going to act to remove whatever changes we make.

00:16:58.980 But then I thought, well, wait a minute. What if instead of just using CRISPR to edit the genome,

00:17:04.200 what if we use it to insert our engineered trait and we also encode the CRISPR system that can make

00:17:11.040 that change? Then you get recursive genome editing. That means in principle, we could build a system that

00:17:18.000 would spread the engineered trait throughout the whole population, even if the trait was harmful.

00:17:23.180 And because I'd read the literature, as soon as I thought of this, I thought, wait a minute,

00:17:27.160 that's a gene drive. Didn't I see a paper where someone had said we should use the genes like the

00:17:31.960 one in yeast to engineer populations? And maybe we could take out, say, malaria. What if we crashed

00:17:37.360 populations of malarial mosquitoes? We might be able to help eradicate malaria that way.

00:17:42.880 With CRISPR, it's so versatile. You could cut whatever gene you needed to at whatever site you needed

00:17:48.620 to. And you could even do it in multiple places, which could make it evolutionarily stable or fairly

00:17:52.760 so. So it was tremendously exciting thinking we might be able to help get rid of malaria and

00:17:57.040 schistosomiasis and so forth.

00:17:58.960 Okay. So you had this idea that was obviously very, very novel and exciting on many levels,

00:18:04.640 but it also kind of terrified you, didn't it?

00:18:08.280 Well, the first day I admit was pure euphoria, thinking up all the applications. Because I mean,

00:18:12.780 malaria is an exceptionally horrible disease. Kills nearly half a million people every year,

00:18:18.500 most of them kids under five. And there's things like schistosomiasis, which cognitively stunt

00:18:23.360 tens of millions. So I spent the first day kind of euphoric and of course doing research to see

00:18:29.320 whether it would actually work. But the next day I woke up in a cold sweat because I was thinking,

00:18:33.780 wait a minute, if just about anyone who can edit the genome of a sexually reproducing species with

00:18:39.640 CRISPR can make a gene drive system to spread their change through the population. Well, what about

00:18:45.900 misuse? Sure, you can engineer a mosquito so it can't carry a disease. Could you engineer a mosquito

00:18:51.900 so it would always carry a disease? How much of a problem would misuse be? And it seemed rather

00:18:59.980 frightening because you could imagine people weaponizing the natural world. So I spent quite

00:19:05.900 a long time before I even told my academic advisor at the time about this idea, because I wanted to

00:19:12.040 learn whether or not that was a likely outcome. And fortunately, I concluded that it's not. Gene

00:19:17.880 drive spreads only vertically, parents to offspring. So that means it's fairly slow. It can't do more than

00:19:23.780 double every generation at absolute most in terms of frequency in the population. It's fairly obvious

00:19:29.540 in that it works in sexually reproducing organisms, which don't have CRISPR in nature. So this is a

00:19:35.760 signature that can't be hidden. It will be present in all gene drive systems. And that means you can

00:19:40.700 always find it if it's there. And most importantly, if you see a gene drive that you don't like,

00:19:47.420 it's trivial to reliably build the counter. You can add some extra instructions for CRISPR telling it to

00:19:54.640 cut the original rogue version. And your version, what we call an immunizing reversal drive,

00:20:00.100 could still spread through the wild population, just like the rogue. But whenever it encounters the

00:20:04.880 rogue, the immunizing reversal drive will cut the rogue and replace it with itself.

00:20:09.620 So there is a reliable off switch. If somebody does something bad that imperils human society or the

00:20:15.460 ecosystem, this immunizing reversal drive is a reliable off switch. And did you think through and

00:20:22.700 basically kind of map out the immunizing reversal drive before you told the world about the gene

00:20:27.940 drive?

00:20:29.080 Long before, yes. And in fact, before I even told George.

00:20:32.620 Oh, before you even told your advisor?

00:20:34.760 Yeah.

00:20:35.400 So you did not let this new idea leak into the world until you knew you had an off switch.

00:20:42.440 Well, I mean, an off switch is perhaps a little much. Just because you can reliably overwrite the

00:20:47.460 unwanted change doesn't mean that it wouldn't have made changes in the ecosystem before you managed

00:20:53.200 to do that. But anything that is slow, obvious if you look and easily blocked is not a terribly

00:20:59.960 great threat. And so with that understanding, I then approached George and said, I think there's a

00:21:05.720 lot of good that we could do this, especially if we do it openly and transparently. And I don't think

00:21:10.600 there's much potential for misuse for this reasoning. What do you think?

00:21:14.000 And when you first told me this story, you said the words that we're very lucky that gene drives,

00:21:20.560 quote, favor defense. What exactly does that mean?

00:21:25.580 Well, anytime you have a technology that is accessible to many people, and gene drive isn't

00:21:31.180 that accessible. There aren't that many labs that use CRISPR to edit the genomes of organisms other than

00:21:36.540 currently fruit flies, worms, and mice are sort of the big ones. If there's a lot of people who can

00:21:43.500 use the technology, then there's a lot of people who could misuse the technology. And so favoring

00:21:49.480 defense just means that if someone does misuse it, everyone else can prevent it from causing much

00:21:55.400 harm. Anything that's slow, obvious, and easily blocked isn't much of a weapon. But things that are

00:22:01.960 fast or are subtle or are unblockable are another story. And most obvious example of something that

00:22:09.820 can't easily be countered is a nuke, right? A nuclear warhead on an ICBM is something that

00:22:15.320 we haven't figured out how to counter, despite many decades of wishing that we could. The way we

00:22:20.860 have to deal with that threat is by minimizing access and deterring those who can actually wield

00:22:29.240 that weapon. Because if they actually use it, there's nothing we can do.

00:22:33.720 On the subject of nukes, this could be a good time to define the terms information hazard and

00:22:41.080 attention hazard, because we'll probably use them later in the conversation.

00:22:45.700 Yeah. So the history of nuclear weapons, it's remarkable how they were originally developed

00:22:51.520 because researchers were afraid of the consequences if malicious people got a hold of them. Specifically,

00:22:58.600 Nazi Germany was the concern of Leo Szilard and Albert Einstein when they wrote the letter that

00:23:04.180 launched the Manhattan Project. But what's not so well known is that once Germany surrendered in World

00:23:10.460 War II, Szilard launched a petition within the top secret Manhattan Project in which he argued that

00:23:17.680 the United States should not use the bomb on Japan. Not because it wouldn't save many American lives by

00:23:24.160 preventing an invasion, but because doing so would call attention to the power of nuclear weaponry.

00:23:32.000 He pointed out that at the time, the notion that any adversary could militarily threaten the

00:23:37.300 mainland American city, Los Angeles or New York, was just laughable. There is no way that any adversary

00:23:43.560 could threaten the continental United States. But if you show the world that there is a single bomb

00:23:50.040 that can destroy the bulk of the city, that would change. You would be advertising the existence of

00:23:56.500 this kind of weapon and thereby incentivizing other countries to gain access to it, which would then

00:24:02.000 imperil your own cities. So he said, if you use the bomb on Japan, the Soviet Union will get it that

00:24:07.560 much more quickly and probably other nations as well. And you will make the United States less secure.

00:24:13.060 And so he spent the rest of his career advocating for a nuclear non-proliferation,

00:24:16.880 pointing out that the fewer actors who had the ability to wield this kind of power, the safer humanity would be.

00:24:23.700 And so in that sense, Hiroshima was an attention hazard. It informed the world that this is possible.

00:24:32.100 This thing you might not otherwise have bothered with because it's going to cost billions of dollars to

00:24:37.340 develop and distract a lot of your top scientists without that attention to the fact that this is possible.

00:24:44.160 That's right. So there's different kinds of information hazards. One is like the blueprints.

00:24:48.920 So in a way, using the bomb on Japan, the isotopic signature told everyone what the Americans had

00:24:53.620 managed to do.

00:24:55.000 So without the blueprints of the bomb, they could look at the radioactive signature of Hiroshima

00:24:59.300 and get a lot of hints.

00:25:01.700 Yeah, that's a conceptual information hazard. But the big one is, as you said, an attention hazard. It says,

00:25:06.760 this is a weapon powerful enough to destroy cities that could determine the future course of warfare.

00:25:11.440 That's a giant neon blinking sign in the sky to anyone who wants to acquire power,

00:25:17.220 that they need to acquire this kind of weapon in order to matter in the future.

00:25:21.920 Now, with all the background of the last several minutes, I'd like to now jump topics pretty

00:25:27.060 significantly to the Spanish flu, which was quite literally extinct for 80 years. But then that

00:25:33.920 changed. How did that change?

00:25:35.660 Well, some influenza researchers at the CDC were concerned that something like 1918 influenza might

00:25:44.420 come again. So they wanted to know, well, what about it was so bad? Because no other flu strain has

00:25:51.000 killed anywhere near as many people. So they went to museum specimens that had samples from victims of

00:25:58.220 the 1918 pandemic. And they also got samples from someone who had died of the flu and been buried in

00:26:05.880 the permafrost in Alaska. And remarkably, they were able to extract the RNA genome of the virus from

00:26:12.540 these samples and sequence it, and thereby learn how to make the virus using recently discovered

00:26:19.980 techniques for virus assembly. So they resurrected this extinct deadly virus.

00:26:26.340 So that was a well-oiled machine that came back into existence after having vanished

00:26:33.160 into obscurity. I'm sure it was a very seductive idea scientifically, but was that a reasonable

00:26:38.880 thing to do from a global security standpoint?

00:26:42.700 Well, there's arguments back and forth as to whether or not it was wise. And there's good

00:26:47.560 arguments on both sides. But what bothers me from a security standpoint is that they posted the

00:26:53.540 genome sequence online. And it's there for anybody to download, as I know, because I was able to find

00:26:59.860 it very quickly via Google myself, which means literally anybody in the world can get this thing.

00:27:05.800 Now, I'm pretty sure I know how you're going to answer this. But what is bad about this genome being

00:27:11.460 online? Well, posting the genome sequence of a virus online, in this case, gave exact blueprints

00:27:18.360 for a pathogen that once killed maybe as much as one in 30 living humans to anyone, including

00:27:25.060 lone terrorists and bioweapons programs who have the skills to assemble it, which doesn't seem like

00:27:31.360 a great idea. Important qualifier. We should be grateful that it's pretty unlikely that releasing

00:27:37.860 that virus again today would cause such a deadly pandemic. And the reason is that virus has been

00:27:43.380 called the mother of all pandemics. That is to say, most modern flu strains are much less deadly

00:27:49.180 descendants. But because most of us have been exposed to at least one of those, we tend to have

00:27:54.980 immunity to that kind of flu virus, especially those of us who caught the flu during the 2009 pandemic,

00:28:01.360 which was the same kinds of molecular coding. Second is we do have antiviral drugs that work against it,

00:28:08.820 even though we probably can't produce enough of those quickly enough to really matter in a fast

00:28:13.560 moving pandemic. But what we do have are antibiotics. And there's some pretty good evidence from autopsies

00:28:21.660 and going through the records of autopsies from 1918, that it killed people at least as much by causing

00:28:28.640 secondary bacterial infections of the lungs as it did directly causing damage, that is the virus itself.

00:28:35.020 And we definitely have enough antibiotics to dose the entire world population already available.

00:28:42.100 Now, this is obviously completely impossible to know precisely. But taking those mitigating factors

00:28:47.860 into account, what would you estimate back of the envelope, the plausible ranges are of a death toll

00:28:55.980 if this thing gets out?

00:28:57.700 I'm not so sure of the lower window just because of, you know, unlikely to take off at all. But if you

00:29:02.140 presume that it does take off, then I would probably say somewhere between 200,000 and 10 million people.

00:29:11.340 And that's just because it would probably be worse than most modern day strains. But it's not going to

00:29:18.620 be nearly as bad as it was back then. So I think that's a probably reasonable 10 to 90% confidence

00:29:23.840 interval. But honestly, I haven't sat down and thought about it for 20 minutes. Of course, on the other hand,

00:29:28.720 just because it's highly unlikely doesn't mean it's no chance. And when there's a non-trivial chance of

00:29:35.020 millions or even tens of millions of people dying, well, some discretion would seem prudent.

00:29:41.300 Now, I've actually known for years that the genome is freely available online, but I actually

00:29:46.060 only learned that the CDC put it up there pretty recently. And at the time, the only explanation I

00:29:51.480 could imagine for that was that it was so impossible for anyone to assemble a virus from

00:29:57.760 scratch back then in 2005 that they just failed to realize how soon the publication of this genome

00:30:04.320 would enable labs in dozens of countries to basically restart the engine of one of history's

00:30:09.560 worst pandemics. And that was a scary enough idea to live with. But then a couple weeks ago,

00:30:15.320 you came along and told me that they actually did realize it would soon become possible to

00:30:21.360 synthesize it, and they were fine with it. And since this makes my head spin, I'd love it if you

00:30:26.980 could give us a quick sense of where the science of virus synthesis was in 2005 when they posted this.

00:30:35.000 Well, the first instructions, the first detailed protocol for assembling influenza viruses

00:30:39.980 was published in 1999. And the first virus was synthesized from synthetic DNA in 2002. And that

00:30:50.520 was poliovirus, which is probably comparable to maybe slightly easier to assemble than influenza.

00:30:58.700 How long after the publication of the genome would you say it became scientifically feasible to

00:31:05.540 synthesize it? Immediately? That is to say, probably a good hundred or so people had the skills at the

00:31:12.840 time to follow that protocol, maybe a few hundred people, to use the protocol to assemble 1918 influenza

00:31:19.000 after publication of the genome sequence. What they may not have anticipated was just how cheap

00:31:24.540 synthetic DNA would become. Because in the last 20 years, the cost of assembled DNA, that is assembled

00:31:32.200 into large gene-length fragments like you would need in order to boot up the virus. The cost has

00:31:37.300 fallen by a factor of a thousand. And while presumably they knew that a good couple hundred people could

00:31:43.820 assemble it immediately, they may not have understood how that number would grow. Because most of the folks

00:31:49.820 involved were virologists, they weren't involved in biotech, and they weren't familiar with exponential

00:31:54.160 technologies. Okay, so we've gone from a hundred-ish labs that could create this more or less

00:31:58.900 immediately after it was published. What would you say that number has grown to today?

00:32:04.380 I'd estimate somewhere between 20,000 and 50,000 people.

00:32:08.000 Could do that right now.

00:32:10.580 Which is pretty alarming, even given that it's unlikely that it would actually cause a pandemic.

00:32:15.560 So is it fair to say that we're kind of relying on the goodwill of half a stadium to a full stadium's worth of

00:32:23.300 people not to put this back out in the world?

00:32:27.000 I mean, frankly, most people aren't that evil. And anyone who is that malicious, if there's only a

00:32:32.080 small chance that it might actually take off, then why would they bother? But you can do the math and

00:32:37.900 say it still looks like a pretty bad bet given the expected death toll, even if you assign a very small

00:32:43.380 chance that someone would, and a small chance that it would actually cause a pandemic. The numbers don't

00:32:47.920 look good when you calculate it from that perspective, especially because it's going to remain

00:32:52.720 accessible to future generations. That is, in five years, there's going to be even more people who

00:32:57.580 could do it. And in 10 years, quite a bit more. The number is only going to grow.

00:33:02.740 Okay, even though 1918 flu would probably be less destructive than many of us would fear if it was

00:33:10.540 released today. The decision to put it online will probably boggle most people's intuitions. It's

00:33:18.080 certainly boggled my own. But the people at the CDC are really smart, and by definition, they're hugely

00:33:23.620 concerned about public health. So what benefits might they have thought would accrue from publishing this?

00:33:31.120 And what perspective might they have that I lack that would cause them to prioritize those benefits over

00:33:38.220 the fairly obvious downsides? There is a mindset difference, which says something along the lines

00:33:44.320 of, if we understand how the world works, then we can come up with better cures and treatments and

00:33:50.760 interventions that we couldn't necessarily predict in the absence of that knowledge. That is another way

00:33:56.480 to say it is, you got to do the basic blue sky research without being able to point to a particular

00:34:01.360 benefit, or you will, of necessity, lose access to all the benefits that you couldn't see before you did the

00:34:07.620 research. That really is what drives folks in these fields, the belief that we can come up with

00:34:15.320 innovations that will make life sufficiently better to be worth the risk. It's important to point out

00:34:21.480 that the scientific benefits that did accrue from resurrecting the 1918 influenza virus were largely

00:34:28.960 accomplished by a relative handful of specialized laboratories doing research with it or with pieces of it.

00:34:36.260 We didn't actually need to publish the genome sequence in order to gain virtually any of those benefits.

00:34:43.960 Given that the number of people with access to 1918 flu has grown by a factor of 300, and it's

00:34:50.460 quite plausible that the CDC didn't realize it would be quite that large quite that soon,

00:34:57.720 do you imagine the people who made the decision to post it regret that decision now,

00:35:03.440 particularly in the post-COVID era? Well, some of the folks who are in charge or highly influential

00:35:08.660 at NIH today have been around for most of the last 20 years, and they've been among some of the

00:35:13.120 strongest supporters of taking highly lethal viruses that aren't very transmissible and engineering and

00:35:19.740 evolving them to become much more transmissible. Since that's simply identifying things from the past,

00:35:26.060 to which most people have immunity today, it's fair to say that they don't regret it.

00:35:30.000 What other scary genomes are there online? Is the horse out of the barn when it comes to

00:35:38.320 terrifying things being posted online, or are we currently in okay-ish shape?

00:35:43.800 You know, a lot of people seem to think that it's too late, but it's really not, because again,

00:35:50.040 1918 isn't that likely to take off itself. There is something that would take off, which is variola virus,

00:35:56.860 smallpox, which is much more lethal and about as transmissible.

00:36:01.280 And also online.

00:36:02.340 And also online, put there by the CDC. But with variola virus, the United States has 350 million doses of vaccine

00:36:10.000 ready to go. And variola is much harder to make as a virus. I would estimate that maybe 100 people could make

00:36:17.380 variola. And it doesn't do the asymptomatic transmission thing the way COVID does. So I think we'd have a much

00:36:24.380 better chance of getting under control, especially since we have so many vaccines. It would require

00:36:28.740 the U.S. to actually donate some to the rest of the world to stamp it out. But the world, in living

00:36:33.180 memory, has experience of vaccinating people against smallpox and eradicating it. And other than that,

00:36:38.700 people speculate about a few things, but there's nothing else out there that is really all that

00:36:42.740 nasty in terms of potential pandemic pathogen.

00:36:45.440 Right. So, of course, the Ebola genome is online, 50% case fatality rate, but relatively not very

00:36:52.560 contagious compared to COVID. SARS, 10%, demonstrably far less contagious than COVID, because we know how

00:37:00.080 many people died in the SARS outbreak, and it was less than 1,000. One that always frightens me is MERS,

00:37:06.060 Middle East Respiratory Syndrome, another coronavirus like SARS, but also not very transmissible. So unless I'm

00:37:13.760 missing something, we're in pretty good shape for now.

00:37:17.200 That's right. I'm not super worried about all the viruses that are out there in terms of

00:37:21.900 causing sufficient devastation to really, certainly not to threaten civilization or anything like that,

00:37:27.060 probably not even the casualty levels that we've just seen with COVID. Folks can reasonably disagree

00:37:31.800 on this, but from my perspective, we're actually doing pretty okay when it comes to blueprints for nasty

00:37:38.080 plagues online.

00:37:39.360 Now, has anybody heard the last episode I did on Sam's podcast? One thing that really worries me a great deal

00:37:46.320 is the deliberately artificially modified H5N1 flu that was created in Madison, Wisconsin, and Holland,

00:37:54.700 two labs, in 2012. It scares me because it has a 60% fatality rate, even worse than Ebola. And in nature,

00:38:02.860 it's barely contagious at all. But that experiment led it to being transmissible, at least in ferrets.

00:38:10.720 And the idea of something with that kind of case fatality rate ravaging the population freaks me out.

00:38:17.000 But you're not as worried about that as I am, correct?

00:38:20.760 That's right. Both of them, yeah, they're respiratory transmissible in ferrets. That doesn't

00:38:24.720 mean they're transmissible enough to cause a pandemic in ferrets, let alone transmissible enough

00:38:28.800 to cause a pandemic in humans. And subsequent studies of them suggest that there are reasons

00:38:34.620 to think that even if they were continually passaged to be even more transmissible and it translated to

00:38:40.780 humans, the virus would be much, much less lethal. The same mutations affecting transmissibility would

00:38:46.880 also reduce the lethality. So those are bad, but are still in the bucket of probably not.

00:38:52.860 So unless we're quite unlucky with H5N1, at this point, we don't really have much to fear

00:39:00.860 from the genomes that are currently on the internet unless we keep posting new and worse ones to the

00:39:08.980 internet. And because it's more future risk than current risk, it's really alarming that you previously

00:39:14.980 said that quite a few people seem to be operating under the presumption that the horse is already out

00:39:20.280 of the barn and therefore why not post everything? Like who's saying that?

00:39:24.320 Well, a lot of people, in fact, to the extent that I've been trying to raise this concern saying,

00:39:28.860 you know, we perhaps shouldn't share blueprints for what amounts to an arsenal of plagues online

00:39:34.220 by doing new research for trying to find things that would actually be that bad and would be likely to

00:39:39.480 take off. A lot of people say, eh, we're already past that. And I think that's because some people

00:39:45.300 are referring to COVID, which is not really what we're talking about. There's no reason not to post

00:39:49.720 COVID genome online when it's already causing a pandemic. That's different from the scenario where

00:39:55.520 it hasn't actually infected humans because that means that it might never do so, at least unless

00:40:01.260 humans cause it to do so. But a lot of scientists seem to take the view of there's lethal viruses

00:40:07.060 online, therefore there's no reason not to put more of them online. And that attitude is quite worrisome.

00:40:14.460 To give you just one example of this, so Dennis Carroll is a huge figure in the field of finding

00:40:19.600 and identifying unknown viruses. In a December article in The Intercept that we were both

00:40:24.500 interviewed for, said that Dennis basically acknowledged that yes, a scientist could use

00:40:30.140 the genetic sequence of a dangerous virus maliciously. But he said that risk already exists.

00:40:36.340 And they quote him directly, saying,

00:40:37.860 we don't need to find some new virus in order to elevate that risk. And I just can't imagine why

00:40:44.760 he would say something like that. Now, to be fair, unlike many folks in this field, Dennis

00:40:49.940 does seem to invite us to hold him accountable in the case that he's wrong. He says, and again quoting,

00:40:55.320 if you go out and cavalierly begin collecting and characterizing these viruses, there is inherent risk

00:40:59.640 attached to that, and you have to be accountable for that risk. So I applaud him for saying that,

00:41:04.680 but I just couldn't disagree more when it comes to the risks of putting new viruses online.

00:41:10.700 Speaking of Dennis, this is actually a really good moment to transition to talk about Deep Vision,

00:41:14.600 the new USAID program that concerns you so much, because Dennis ran the immediate predecessor to it,

00:41:20.520 a program called PREDICT, which was basically the template for Deep Vision. And that's just strange

00:41:26.580 to me that PREDICT was able to spawn a follow-on program. In light of some of the terrible scandals

00:41:34.660 frankly, it has been implicated in. At the center of which is the fact that PREDICT directed

00:41:39.720 a fair amount of money by way of an intermediary called EcoHealth Alliance to fund research on

00:41:46.280 coronaviruses in Wuhan. And not just that, but in labs that were known to have very shoddy safety

00:41:54.180 practices to the US government. We know that beyond a shadow of a doubt from some declassified

00:41:58.580 diplomatic cables. So that seems like an incredibly radioactive error in judgment. This was all out

00:42:06.360 in the open by the time Deep Vision was funded. And on top of that, we have to add the significant

00:42:11.760 possibility, not a certainty, but possibility that that work may have resulted in a leak that led us

00:42:18.340 to COVID. While there was a time when a lot of well-regarded scientists were putting the odds of a lab

00:42:23.760 leak at zero, that's a long, long time ago. At this point, substantially nobody who has

00:42:29.460 significant scientific depth in this would put the odds of a lab leak at zero. Some put it as high

00:42:35.300 as 90%. I'm sure some put it in single digits. But it is undeniable that PREDICT may have had a hand

00:42:42.320 in that. So for all those reasons, this is politics. It's irrelevant to your discussions, Kevin, so we can

00:42:47.660 just leave it there. But I just find it a little bit unfathomable that PREDICT spawned an encore.

00:42:53.620 Now, one of PREDICT's main activities was something called virus hunting. And that's also going to be

00:42:59.300 a major activity of Deep Vision. What is virus hunting?

00:43:04.900 Well, you probably have this vision of someone who puts on some kind of suit and cavalierly goes

00:43:10.880 spelunking in a bat cave to collect samples and take them back to the lab, see what kinds of viruses

00:43:15.620 you can pull out. Yeah, Indiana Jones, National Geographic kind of vibe.

00:43:20.220 Now, you've got to keep in mind that's the media distortion. That's the glamorous presentation of the

00:43:24.240 virus hunter. Most virus hunting is probably more like going to those wet markets with various wild-caught

00:43:30.840 creatures and taking samples, or going to bushmeat markets in other parts of the world, or contracting

00:43:37.560 with local hunters to bring samples of various wild critters back, or even just taking general

00:43:43.280 environmental samples out there from a wide variety of environments and isolating whatever

00:43:48.460 viruses you can find from those samples. In the modern day, now sequencing them all, sharing the

00:43:54.120 genomes online, and then potentially doing more than that with the viruses that you've isolated.

00:43:59.320 The idea is we want to know what viruses are out there.

00:44:03.000 Having done a bit of digging into this, some basic stats on PREDICT. They found roughly 1,200

00:44:08.360 previously unidentified viruses. They identified them, they sequenced them, and they were active,

00:44:14.860 I think, in 20-something countries, 60 foreign laboratories. And what was interesting to me,

00:44:21.140 in the 1,200 viruses, all of them had the potential to erupt into pandemics. That was the presupposition

00:44:28.120 for selecting them. I got this from the LA Times, including more than 160 novel coronaviruses.

00:44:34.940 So that was PREDICT's haul. What's interesting to me about Deep Vision, they're actually talking

00:44:41.560 about 10,000 to 12,000 viruses over five years. So something like 10x the number of viruses in half

00:44:49.120 the time. This is with 125 million bucks instead of 200. It's a big explosion in efficiency. And to

00:44:56.080 the extent that this work poses risks, it's a big explosion in the risk. What's driving this

00:45:01.540 acceleration? And if we continue to do programs like this after Deep Vision's legislated five-year

00:45:08.600 life, will that explosion continue to the point that at some point we might be doing this for

00:45:13.760 many tens or even hundreds of thousands of viruses?

00:45:17.540 So one big chunk of the cost is taking all of these isolated virus samples and sequencing them.

00:45:23.220 That's creating the detailed blueprints. Because once you have the genome, then we can use DNA

00:45:27.200 synthesis to reconstruct that virus using these virus assembly protocols that folks have worked

00:45:33.360 out and made tremendous improvements in over the last few years. But the main cost reduction for

00:45:38.880 programs like Deep Vision is that sequencing is just getting cheaper and cheaper and cheaper.

00:45:42.800 And that means it's easier and easier to collect more and more viruses. What isn't appreciably easier

00:45:48.720 is taking them back to the lab, culturing them in the lab, and performing the set of studies that tell

00:45:54.160 you, is this one likely to actually cause a pandemic?

00:45:58.000 So what you're talking about now is essentially step two. We have found the viruses, we have virus

00:46:02.320 hunted, now they're back in the lab, and what you're talking about now is called characterization, right?

00:46:07.280 Yeah. The whole point of the program is to identify which viruses might spill over and cause future

00:46:13.760 pandemics in the hope that we could do something to prevent it. There's four key classes of experiments

00:46:20.560 that you perform on an animal virus to determine whether it's a good candidate for causing a

00:46:26.320 pandemic in humans. You want to know how tightly does this virus bind to human target cells? How

00:46:34.000 readily does it actually infect those cells? How readily does the backbone of the virus replicate

00:46:40.880 and churn out new copies of the virus in relevant human tissue types? And because you can't test it in

00:46:47.680 humans, of course, how readily is it transmitted in animal models that are chosen for their similarity

00:46:53.520 to humans, whether naturally or because in the case of mice, they've been engineered to express

00:46:58.960 human receptors and have human-like immune systems? So to play this back to you to make sure I understood,

00:47:03.840 there's four sets of experiments. Experiment one is, does this virus bind to a receptor on a human cell?

00:47:10.640 Like, can it find the door? Step two, having found the door, can it get in? Does it infect the cell? So does it

00:47:17.360 find the door? Does it have the key? Step three, having gotten inside, can it hijack the mechanisms

00:47:24.080 of the cell to replicate? And then step four is, if it does those first three things, does it seem to be

00:47:31.600 transmissible in animal models that replicate as closely as possible transmissibility in humans? Did I get

00:47:38.480 that right? That's exactly right. Great analogy. And if you answer those four questions, you'll basically

00:47:44.880 determine which of these 10,000 yet undiscovered viruses actually have major pandemic potential.

00:47:51.600 There's going to be a tiny minority of them. And this is something you'd have zero knowledge about without

00:47:58.080 taking these steps. So having done this characterization or pandemic identification process,

00:48:05.440 you will know something that humanity would otherwise never have known, which is that these viruses could be

00:48:11.840 incredibly dangerous, and these are not. Now, while I know this is, of course, an unknowable number,

00:48:18.560 and the best we'll probably do is a confident range, roughly how many pandemic-grade viruses would

00:48:26.240 you estimate Deep Vision is likely to find over its five years if it does, in fact, net 10,000 to 12,000 mystery viruses?

00:48:34.720 It's a good question. That's about 10 times as many as PREDICT found. And PREDICT didn't find

00:48:40.400 any that looked particularly likely in and of themselves. It found a bunch that looked worrisome

00:48:44.880 in one or another property on some of those tests, but it didn't find any that checked all the boxes,

00:48:49.520 so to speak. We can also just look back and say, well, how many viruses are plausibly out there?

00:48:54.080 How many pandemics do we normally see? And then here's the real big one. What fraction of viruses

00:48:59.520 out there are actually going to spill over and get into humans in the first place? Because we're

00:49:05.120 still discovering new species out there. There's plausibly a lot of viruses out there that have

00:49:09.280 never seen a human and never will. So some researchers say there's probably way more pandemic-capable

00:49:15.120 viruses out there than will ever cause pandemics, because most of them are just never going to come

00:49:18.560 in contact with humans. Let's try to drill this down and say the low-end estimate for how many

00:49:23.280 viruses might be out there in mammals is around 40,000, of which they estimate that maybe 10,000

00:49:31.440 are capable of human-to-human transmission. That's one rigorous estimate that's out in

00:49:36.000 the literature. That's the low-end estimate. Others in the field have estimated as many as

00:49:40.000 half a million in mammals. They didn't give an exact number for how many in humans, but you dig

00:49:44.560 into the other papers, they suggest that maybe 20% of those. So somewhere between 10,000 and 100,000

00:49:49.920 viruses that could plausibly human-to-human transmission. But human-to-human transmission

00:49:55.040 is not the same as causing a pandemic. There's lots of viruses that can spread human-to-human,

00:50:01.920 but on average each person infects fewer than one other person. So they die out, like SARS-1 and MERS.

00:50:09.040 So how many of them could actually cause pandemics? That's really hard to know, but if deep vision

00:50:15.120 finds 10,000 and you assume the lower-end estimate, then that would be as much as a

00:50:19.280 quarter of all the viruses out there. So let's say 20% of those are going to be capable of

00:50:24.960 spreading human-to-human. It's 2,000 that we expect human-to-human. What fraction of those 2,000

00:50:30.400 spreading human-to-human could actually cause a pandemic? It's not really known because the ones

00:50:36.000 that are endemic in humans are a biased sample from all of history. So it's really hard to say. But if

00:50:41.280 we're guessing like one in 100, then that would mean 20 pandemic-capable viruses. But that doesn't

00:50:46.640 make sense statistically because PREDICT found 1,100, 1,200 viruses and didn't find anything

00:50:52.480 that clearly looked like a pandemic-capable virus. Which makes logical sense, and I do generally

00:50:57.280 believe that probably that means 8-ish is a high-end plausible outcome. But a quick question. Is it

00:51:03.200 possible that the targeting of the hunted viruses or the accuracy of some of the laboratory steps

00:51:11.920 have gotten substantially more efficient, that perhaps more will be caught, maybe predict missed

00:51:18.320 things that deep vision will not miss because it's being done in 2022 instead of 2009?

00:51:25.040 The main difference is that now there are much better computational tools for matching virus to

00:51:32.160 host species by receptor similarity and the like. So when deep vision decides a virus is worth testing,

00:51:39.120 from the sequence data they're going to have a much better shot at identifying at that virus being

00:51:44.640 high risk than PREDICT was. Okay, so to play this back, the upper ceiling could be 8. It could be

00:51:52.000 higher because of improved technology. Bottoms-up estimate gets us into the possibility of it maybe

00:51:57.840 even being low dozens. We don't know exactly. But it's important to know that it is possible that you will find

00:52:03.840 zero pandemic-capable viruses, in which case the program cannot by definition trigger any of the

00:52:10.720 terrifying things that we're about to talk about that it could trigger. So this is not a certainty

00:52:14.880 that there's a huge risk here. But zero feels like a low-end estimate because the organizers of this

00:52:22.640 program presumably know a lot about the state of the science and what can be done now and what can be

00:52:27.040 expected. And they wouldn't be spinning up this program if they thought it was a real possibility

00:52:32.000 that there would be zero because then by definition the program will fail. So zero is possible. And then

00:52:37.680 the high-end from the bottoms-up and an intuition and a top-down, we're getting into quite plausibly

00:52:45.200 mid-to-high single-digit number possibly but not likely into the dozens. Is that reasonable ballpark?

00:52:51.280 Yeah, I think that's, frankly, a much better way of estimating it than I was giving.

00:52:55.920 Okay, thank you. Now, the characterizing work is really interesting to me for a bunch of reasons.

00:53:02.640 How hard and expensive is it to do this characterizing work at scale? And how likely

00:53:10.400 is that kind of work to happen with many thousands of viruses in the absence of this program?

00:53:16.640 From what I understand, the characterization aspect is the most expensive component of a program that

00:53:24.240 is not limited to virus hunting and characterization. That is, Deep Vision is also about monitoring the

00:53:30.880 human populations that are closely exposed to a bunch of the animals that are most likely to pass

00:53:35.520 on viruses to them, so as to identify outbreaks early and give us a better shot at containing them.

00:53:42.000 So this program does a bunch of really good stuff, too.

00:53:44.960 Yeah, that's an unadulterated good thing, what you just mentioned.

00:53:48.160 But my understanding is that the bulk of the cost is actually taking the

00:53:52.400 viruses back to the lab and running the experiments, because that requires a bunch

00:53:55.600 of trained virologists with skills that are much more specialized than following a protocol

00:54:02.080 to assemble a virus from the sequence. That is, there's a whole lot more people that can

00:54:06.080 make a virus from a genomic blueprint than can search through this haystack of wild viruses

00:54:12.560 looking for ones that could plausibly cause a pandemic.

00:54:15.680 Okay, so we virus hunt, we characterize, we haul in a record-shattering number of viruses over the

00:54:23.040 five years. At the end of which, humanity knows something that it would otherwise not have known,

00:54:29.440 which is that these, let's call them six, previously undiscovered viruses pose a very,

00:54:35.600 very significant pandemic risk. Now, I could see very good arguments for us not wishing to know

00:54:44.080 this information, but I could also see arguments for us wishing to know this information. But what's

00:54:49.600 most significant to me is step three in this process, which so flabbergasted me when I learned about it

00:54:58.480 that my head is, frankly, still spinning a little bit. And I want to be very clear, and I don't want

00:55:03.760 to exaggerate, what step three is. And so in order to put this out there in Deep Vision's own words,

00:55:10.800 so to speak, I'm going to quote from their so-called NOFO, which is a notice of funding opportunity.

00:55:17.760 This is something that USAID and other agencies put out in the world and say,

00:55:21.520 we want this program created. And Deep Vision's NOFO is over 100 pages long, so it's very,

00:55:26.560 very specific. And then universities and other people can bid on doing this piece or that piece,

00:55:32.880 and all the pieces are eventually assembled by a program officer inside of USAID. So this NOFO,

00:55:38.960 this 100-plus page document in the public sphere, a bit of Googling, you can find it,

00:55:43.200 is as clear and detailed and certainly official statement of Deep Vision's intentions as exists.

00:55:49.760 So what I'm going to do right now, Kevin, is I'm going to read a few snippets. And for those who

00:55:54.960 find the NOFO, these are scattered a few phrases and sentences from the bottom of page 16 to the top

00:56:00.000 of 17. And I'll ask you to explain a couple terms and meanings as I go along. So first of all,

00:56:06.160 from the NOFO, since USAID expects that data generated by Deep Vision will be publicly available,

00:56:13.760 and I'm going to skip ahead a few words for efficiency, Deep Vision will assist in linking

00:56:19.520 in-country data with global systems, e.g. GenBank and GIS-AID. What are GenBank and GIS-AID,

00:56:28.240 and what does it mean to link the data with them?

00:56:31.840 So that's referring to the standard open data practice of sharing the genome sequences of

00:56:38.400 everything that you find in a given research program. So GenBank is the main repository hosted

00:56:43.920 by the U.S. government of all genomic information that everyone has sequenced and submitted to GenBank.

00:56:49.920 GIS-AID is the repository specifically of virus strains. Not everyone sends the sequence of every

00:56:56.640 strain to GenBank if there's already a sequence that is very similar there. GIS-AID is for the

00:57:02.960 detailed virologist viral evolution specialists who want to see and map out how different are the

00:57:10.160 mutational variants of a given virus. So basically it is a wildly standard

00:57:16.320 scientific practice now to put any sequence you find up to the cloud before you have any notion of

00:57:22.000 its significance. So everything they find by definition is going to be made publicly available,

00:57:26.880 long before they characterize it. Is that roughly correct?

00:57:29.600 That's exactly right. The principle of open data is you collect data in a scientific experiment,

00:57:34.480 you should really put it online for everyone to see.

00:57:36.320 Okay, a couple words after that last snippet of text, pointing again from the program's NOFO.

00:57:42.400 Knowledge gained by deep vision on novel viruses assessed to be zoonotic and significant epidemic

00:57:49.600 slash pandemic threats will be immediately available to the in-country owners of the data and will be

00:57:57.200 expected to be available expeditiously to policy makers, the private sector, and implementing partners.

00:58:04.720 So I think I know what that means. Could you just play back what that amounts to?

00:58:09.280 If you think there's a virus that poses a major epidemic, that is local outbreak risk or,

00:58:14.560 gods forbid, a pandemic risk, you need to tell the country that you found it in immediately so that

00:58:19.840 they can take action. And it's expected that they will then share it with the international community

00:58:25.040 as soon as possible.

00:58:26.080 So bottom line, if deep vision succeeds, everyone on Earth will get the recipes for

00:58:35.200 the dangerous viruses that we find. They will be put online just as the 1918 flu virus genome was put online.

00:58:42.560 That's right. Because remember, the intent of this program is to identify which natural viruses might

00:58:49.120 spill over and cause pandemics, if they spill over into humans, so as to better target interventions

00:58:55.120 to prevent that from happening. And the way they hope to do that is by creating a list of viruses

00:59:02.320 rank ordered by perceived threat level, which they've already done for the viruses that are highly lethal

00:59:07.520 and don't spread particularly well human-to-human, but are known to have spilled over. That's the

00:59:13.200 product of a related USAID program called Stop Spillover, which is generally highly admirable with the

00:59:18.560 perhaps exception of this list. Although, as we discussed previously, there's not a whole lot that's

00:59:22.000 dangerous on that list so far. But what deep vision apparently wants to do is, as soon as they find

00:59:27.520 anything dangerous, they're going to alert the world to it and put it on that list, presumably in the

00:59:32.320 number one position. So bottom line, if and when deep vision succeeds, based on what they've stated

00:59:38.880 themselves in general practices, the genomes of potentially very significant pandemic viruses

00:59:46.000 will be put on the open internet for anybody to find. Correct. And given the base of scientists and

00:59:52.080 individuals who are already able to conjure certain viruses from whole cloth in the lab, tens of thousands

00:59:59.440 of people will be in a position to follow those recipes as soon as they're posted and animate that

01:00:06.400 virus, even though they don't personally have a sample from the bat that it originally came from.

01:00:11.600 Did I get that right? That's exactly correct. And to tie it back to what we were talking about a

01:00:16.640 little while ago, the current population of frightening genomes on the internet is probably not all that bad.

01:00:24.800 But this action could substantially change that by saying, yes, world, 1918, probably not as bad as you

01:00:32.720 think. H5N1, pretty good chance it's not transmissible. But holy cow, people, these, whatever, six viruses

01:00:40.000 can really do the trick. That's going to be out there as a result of this program. And tens of thousands

01:00:46.000 of people at first will be able to conjure these things, followed by we don't know how many in the

01:00:52.160 future. Is that an excessively alarmist interpretation? No, it's really not, I'm afraid.

01:00:57.920 And it's worth comparing, COVID has killed many more people than any single nuclear detonation.

01:01:03.920 There are nine commonly acknowledged nation states with access to nuclear devices. So as soon as viruses

01:01:14.000 likely to cause pandemics are publicly identified with freely available genomes, the number of actors in

01:01:21.600 security terms capable of inflicting a million plus deaths in expectation will rise by roughly a

01:01:28.720 thousand fold. So the vulnerability of the world to a deliberately inflicted pandemic will rise

01:01:38.080 tremendously if deep vision succeeds. I mean, that's what I get from this. And again, please tell me if I'm

01:01:46.400 being unrealistic in my interpretation. Frankly, I wish you were, but that seems pretty accurate.

01:01:53.280 I'm going to add, it's a little bit worse than that actually, because a lot of people

01:01:56.960 seem to assume a pandemic is a pandemic, right? It's the same virus. But nature throws on average

01:02:02.800 four or so pandemics at us per century. If someone deliberately causes a pandemic, if the list has eight

01:02:10.800 viruses, they could make and release all eight at once. The idea of facing eight different COVIDs,

01:02:19.600 none of them with vaccines or even tests, all at the same time, it just boggles my mind. That's

01:02:26.160 just such a horrifying thought, given how hard it was to deal with this one.

01:02:29.680 And it's worse than that, I'm afraid, because a lot of people see our best hope as being vaccines,

01:02:35.200 because obviously that's what got us out of this one, right? Well, as we're talking about this,

01:02:40.000 it's been 102 days since the Omicron variant of COVID was first sequenced and the genome shared with

01:02:47.680 the world. The goal that the White House has put forward is that we should have vaccines available

01:02:53.280 within 100 days of the genome sequence being identified. But by a couple of days ago, Omicron had

01:02:58.720 infected a very large fraction, maybe not 50 percent, but in some places, definitely 50 percent of all

01:03:05.280 humanity. So 100 days is too slow, even though Omicron arose in one place in the world and spread from

01:03:11.920 there. And that's our stretch goal is 100 days. And I agree with you, as you know, I've been aware of

01:03:17.280 that number for a while. And it's just, it's dangerously timid. But it's worse than that.

01:03:22.560 Oh, good. If Omicron can spread that fast from a single point of release,

01:03:27.440 anyone malicious enough to deliberately cause a new pandemic, let alone several at once,

01:03:33.360 would almost certainly release them in multiple travel hubs, say airports throughout the world.

01:03:39.600 So it wouldn't be a single point of release you have a chance to notice because lots of

01:03:42.640 people are getting sick. It would be people getting sick in cities all over the world all at once.

01:03:48.000 Wow.

01:03:48.960 It would substantially diminish our response time. And one would imagine thereby substantially

01:03:54.880 increase the resulting death toll. So even if it's just one virus, a deliberate release is worse

01:04:00.800 than a natural spillover or an accident because both of those are spread from a single site.

01:04:05.680 Yeah, that's a really interesting and important point. So COVID emerges in Wuhan and it's two and

01:04:12.160 a half months before it shows up in Detroit. That was nowhere near enough lead time for us. And it could

01:04:17.520 be chopped by 99 percent by a malevolent actor. That's scary.

01:04:21.360 And there's one more thing I really want to emphasize, which is some people might be saying,

01:04:24.560 well, yeah, but no matter how those four experiments turn out, even if they're all four

01:04:29.280 positive and give values comparable to an endemic human virus of the same family,

01:04:33.280 you might say, well, maybe that's just a 50 percent chance that it would actually cause a pandemic.

01:04:36.880 Because again, the transmission studies are in animals, not humans. And to be fair,

01:04:40.320 I just pooh-poohed the H5N1s, mutants, as being actually likely to cause pandemic in humans.

01:04:46.560 So let's say that, okay, they find five viruses that they think could cause pandemics. Let's assume

01:04:52.720 each of them is actually 50 percent likely to cause one. Well, if someone were to assemble and

01:04:57.760 release all five, they'd have a 96 percent chance of causing at least one pandemic and an 81 percent

01:05:04.480 chance of igniting two or more simultaneously, again, presumably across multiple travel hubs.

01:05:09.680 That's pretty terrifying. And if you assume that deep vision is only the beginning and the goal is to

01:05:16.640 find all pandemic-capable viruses out there in nature and put them all on a list in order to

01:05:21.760 prevent as many natural pandemics as possible, then you're raising the possibility that someone might

01:05:26.720 do it and then you'd see copycats. Because that's what happens with mass shootings. It's a socially

01:05:32.800 contagious behavior. Once one person sets the example, then many more people who are mentally

01:05:40.800 ill or captive to a horrific ideology would be more likely to do it now that they know that it's

01:05:46.560 possible. In general, I'm not super worried about civilization falling apart because of a single

01:05:52.080 natural pandemic virus. But if you're talking releasing eight at once, that might be another

01:05:58.800 story depending on how bad they are. I mean, remember, if essential workers aren't willing to

01:06:03.040 go out there and risk their lives, then you start having problems in food and water and power

01:06:08.240 distribution. And if those fall apart, then everything falls apart. So putting those genomes online

01:06:14.800 looks pretty risky given that the best case scenario is that we would prevent all of the natural pandemics,

01:06:21.360 which is historically around four per century. It seems a lot to say it's definitely a good deal

01:06:28.720 to prevent four natural spillovers per century in exchange for giving tens of thousands of people

01:06:34.720 the power to launch more pandemics than that simultaneously across multiple travel hubs.

01:06:40.800 In light of all of that, I don't know how you could argue against the statement that if deep vision is

01:06:48.880 spun up and succeeds, the exposure humanity faces to malevolently inflicted pandemics skyrockets,

01:06:57.440 even if deep vision finishes its job and none of this work is ever, ever done again. Which is,

01:07:04.880 first of all, a completely naive assumption. I don't think anybody would make that, given that

01:07:08.640 there's momentum to do more and more of this rather than less and less of this in USAID of all places,

01:07:14.320 which we'll discuss in a bit. And that brings me to my last quote from deep vision's NOFO.

01:07:20.400 The USAID seeks to assist a limited number of countries with a focus on Africa, Asia, and Latin

01:07:30.640 America to establish capacity to detect, characterize, and disseminate information and findings regarding

01:07:38.960 previously unknown viruses that have originated in wildlife. Correct me if I'm wrong, but basically,

01:07:45.200 he says, in addition to doing this stuff ourselves, we are going to be training lots of foreign countries

01:07:51.920 to also do this. And it's noteworthy that if this training, which is currently specialized,

01:08:00.080 and will probably enjoy many breakthroughs as a result of $125 million of funding over five years,

01:08:06.640 you're going to have a lot of smart people concentrating on characterizing better, cheaper,

01:08:10.080 faster, which wouldn't otherwise happen without this program. So if deep vision succeeds and

01:08:15.760 society continues tripling down on this stuff, this work will be as a consequence done on foreign soil,

01:08:23.120 where the US government has absolutely no sway or say, will eventually be done with tools that are

01:08:28.720 far better than what we've got right now, tools and techniques. So this is the horse out of the barn

01:08:34.560 that doesn't actually yet exist on the internet with the relatively limited set of sort of dangerous

01:08:42.320 viruses that are there. Now, there is one factor that I should probably point out, which is suppose

01:08:49.760 we're concerned about state bioweapons programs. If they were to say, don't mess with us, we have the

01:08:56.400 ability to launch new pandemics, no one would believe them. I would laugh and say I can fabricate data

01:09:03.440 too. That's cute. But if it's done by deep vision, by deep vision funding, a bunch of independent labs

01:09:11.360 who are well-meaning because they want to prevent natural pandemics and they just haven't thought

01:09:16.080 about the security risk. If they publish that data, I believe it. If it's done by independent labs who

01:09:23.600 don't have a motive to have a pandemic in their pocket, it's going to be believable in a way that it

01:09:28.880 won't be if malicious actors try to do it themselves. And I think this is one of the most powerful and

01:09:34.800 original arguments that I've heard about the danger of this kind of genomic information being

01:09:39.920 out there. And I hadn't thought about it until you first mentioned it to me on the phone before. If, I

01:09:44.560 don't know, an environmental extremist movement or a frightening anonymous source on the internet,

01:09:50.800 or even a state actor like North Korea, were to say, hello world, I am going to inflict a devastating

01:09:58.000 pandemic unless you meet my demands, it would be laughed off. It would freak people out, it would

01:10:03.840 probably get a lot of coverage on CNN, but the scientists who are in a position to inform the

01:10:09.440 national security apparatus in various countries about whether or not to take this threat seriously

01:10:14.160 will say this is just not feasible. That totally changes when a genome that the world would not

01:10:21.440 have known otherwise is definitively blessed and publicized by a brilliant scientific group as being

01:10:29.280 that thing. Once that work has been done, which wouldn't be done otherwise, now all of a sudden we know

01:10:35.520 30,000 and perhaps quite a bit more people can follow that recipe overnight. And what that means

01:10:43.840 is we could suddenly go through a terrifying series of hijackings of the attention and the stress

01:10:52.560 levels of the world with all kinds of people issuing those threats and they don't even have to have

01:10:57.680 access to one of those 30,000 people that can make that virus. They just have to say it because it's

01:11:03.600 suddenly credible. Nobody will be able to deny that it's impossible for one of those 30,000 people to be

01:11:08.800 under the control of this terrifying anonymous source online or this terrorist group or this rogue

01:11:14.640 actor or whatever. And that alone is just wildly disruptive. It's one thing if a kid pulls the fire

01:11:22.560 alarm in their school to get out of their science exam or whatever. But if that kid calls the school and

01:11:29.200 says, I have a hydrogen bomb that's ready to go off, no one's going to get that alarm. The creation of this

01:11:34.800 credibility empowers so many more people than, quote unquote, just the 30,000 people who could act on it

01:11:43.200 to do awful things in the world. And it's probably even more likely because the people who might make

01:11:49.440 that threat may in their inner conscience say, I'm actually not going to do it. I don't have a gun.

01:11:53.920 You know, I'm just holding up the liquor scorer with a squirt gun. No one's going to get hurt.

01:11:57.760 Might actually make them more likely to move ahead with that kind of thing than it would be for a state actor.

01:12:05.040 to do something. Yeah, I'm actually at least as afraid of not just non-state actors, but even just

01:12:11.520 individuals, mentally ill or otherwise. So Seichi Endo was a member of this apocalyptic terrorist cult,

01:12:17.680 Aum Shinrikyo in Japan. Aum was responsible for making and releasing chemical weapons that killed

01:12:23.760 a bunch of people in Japanese cities in the early mid-1990s. But before he joined the cult,

01:12:30.080 Endo was a graduate-trained virologist. And he sought to obtain samples of Ebola for use against

01:12:37.680 civilians and was unsuccessful. James Holmes, who is a convicted mass murderer, the Aurora shooter,

01:12:43.600 quit his life sciences PhD program not long before he opened fire in the theater. And of course,

01:12:49.680 pre-al-Qaeda, the most famous terrorist was arguably the Unabomber. Ted Kaczynski was a brilliant

01:12:55.280 mathematics professor who referred in his manifesto to the immense power of biotechnology,

01:13:01.280 even though he wrote it decades ago. It's really hard to imagine someone like that,

01:13:05.440 who wanted to bring down the industrial system, would not have used that power if given access

01:13:11.760 to pandemic virus genome sequences and modern virus assembly protocols. And that's not even getting

01:13:17.200 to groups like Daesh, ISIS, and other kinds of terrorists. Folks who might actually be tempted to

01:13:24.160 use it, even if it would hit their own people. Yeah. And the omnicidal factor, if we can call it that,

01:13:31.680 is something that I think people very frequently miss. Because it's easy to rule out

01:13:35.520 a majority of bad guys who normally dominate international headlines. We can safely say that

01:13:44.480 Putin, Xi, indeed North Korea, are wildly unlikely to inflict a pandemic on the world because they have

01:13:53.120 so many people to protect and so much to lose. And if we're in the mindset that it takes a major

01:13:59.280 state actor to do such a thing, there is a bit of mutual assured destruction built into that.

01:14:05.680 There's radical deep ecologists who don't think much of humanity in general and might think that

01:14:10.720 the world is better off with a whole lot fewer humans. There's folks with those persuasions who

01:14:16.000 have called for it. Certainly it's unclear whether a group like Al Qaeda would actually have unleashed

01:14:21.360 something that would hit their supporters as well, but quite possibly. I mean, they did put out a call

01:14:26.640 for brothers with skills in microbiology in 2010 to make biological weapons of mass destruction. And then

01:14:34.000 maybe one of the most haunting ones is folks with the mindset of the German wings pilot who decided

01:14:39.840 to commit suicide. He was mentally ill, didn't disclose his mental state, and decided to end it

01:14:46.320 all by flying the plane into a mountain.

01:14:49.360 Yeah, in the United States, I'll add, we have an average of more than one mass shooting per day.

01:14:54.560 So that suicidal mass murder instinct is out there, and the omnicidal instinct is out there. And then

01:15:00.480 there's just thinking out loud, like, there could be other people who don't realize quite how deadly the

01:15:05.760 thing they're doing is, and might have some sort of boneheaded cunning plan. Like, I could imagine

01:15:12.320 somebody saying, ooh, I remember the markets crashing when COVID came out. I'm going to release something,

01:15:18.000 short the market, make a pile of money, and stockpile gas masks. It sounds moronic and absurd, but there's

01:15:26.800 seven billion people in the world, and moronic, absurd ideas occur to at least some of us each and every day.

01:15:33.280 Okay, now I want to get into the final, and in some ways, maybe even the most important part of our

01:15:40.720 conversation, a little bit of a preamble. It's obviously abundantly clear that I am horrified by

01:15:47.760 the agenda and the prospects of this program called Deep Vision. But I'm going to try to do my very best

01:15:54.960 to push back with every argument that I have heard of or have dreamt up on my own in favor of Deep Vision,

01:16:02.200 because there is another side to this story. And the people behind this program are extremely smart

01:16:08.280 and well-intentioned. And I'll also add that when I personally first heard about PREDICT,

01:16:14.760 it was when the news broke about it being shut down. And that actually happened either right before

01:16:19.240 or right as the pandemic was starting. And at the time, I thought this was insane, because zoonotic

01:16:26.040 spillover has obviously happened before, it will obviously happen again. And how can we defend ourselves

01:16:31.240 against it if we don't study our enemies? I'm actually very sympathetic to these arguments.

01:16:36.040 So let's go back and forth. And I'm really going to try to make every argument I'm capable of in

01:16:42.920 favor of this. So as much as the other side as we can possibly deliver. So my first question to you,

01:16:51.400 which is the question that was in my mind when this thing was first shut down,

01:16:54.280 shouldn't we want to know who the bad guys are and where they live before they strike? Shouldn't we

01:17:01.880 want to know where the dangerous pandemics dwell and know exactly what they look like? And isn't that

01:17:07.960 a bit like putting wanted posters all over the Wild West?

01:17:12.360 It's an extremely intuitive and compelling rationale. And most of the time it's true. The problem is

01:17:22.280 that it's kind of like saying, here's a wanted poster for this particular device, which happens

01:17:28.920 to be the detailed schematics for a hydrogen bomb, and peppering the world with different blueprints for

01:17:35.640 hydrogen bombs or equipment necessary to make hydrogen bombs in order to identify people who

01:17:42.680 might be making hydrogen bombs. It's not a perfect analogy, but it's hard at the top level to say,

01:17:49.720 okay, best case scenario, we prevent all natural pandemics. Is that worth giving these tens of

01:17:55.800 thousands of people the power to release more pandemics than would normally occur in a century

01:18:01.480 all at once? That just doesn't seem like a good trade. And when you put it in that context,

01:18:07.000 then you kind of have to look back at your basic assumption that it's always good to know more about

01:18:11.960 a threat. Because that intuition doesn't encapsulate the costs of knowing. That is,

01:18:18.360 the value of information can be negative. That's the whole point of an information hazard. There's

01:18:23.640 actually a whole field of modeling and information theory on calculating the value of information.

01:18:28.280 When should you run an experiment to learn more? And it tends to assume that information is always

01:18:33.080 positive. And the question is, is the cost of running the experiment worth reducing your uncertainty

01:18:37.480 about what would happen in the world? But it's also possible for the value of information to be

01:18:41.480 negative. And you can imagine an extreme case. Suppose you figured out some way of creating a

01:18:50.120 singularity on earth that would devour the planet, to give the absurd example. Would we want

01:18:54.920 humanity, anyone in humanity to know that that is possible? Probably better off if no one ever

01:19:00.520 knows that that's possible. And because of the credibility issue and the difficulty of doing this

01:19:05.320 kind of research, it looks a lot like if we don't go there, then there won't be credible pandemic

01:19:11.720 capable virus blueprints online for quite some time. That's not going to last forever. We will eventually

01:19:19.320 lose. They will eventually go up there, whether through this kind of route or another one. But

01:19:25.720 the longer we can push it off, the more time we have to build actually effective defenses.

01:19:31.560 And it's important not for listeners to come away with a sense of doom and gloom, because it's also

01:19:36.200 true that even for the scenario where people release multiple pandemic capable viruses all at once in

01:19:43.160 multiple airports. If we have sequencing-based monitoring systems in place everywhere, we'll pick

01:19:49.240 them up nearly immediately, certainly before they spread too widely through air travel, and be able to

01:19:57.400 put on our protective equipment. And if we actually tried, we could probably build comfortable,

01:20:03.000 even stylish equipment that keeps you from getting infected with viruses. And we could make it available

01:20:08.200 to everyone who is required to keep food and water and power flowing for as long as it takes to stamp

01:20:15.000 out the virus entirely. And if we had that kind of equipment available, and the threat was that salient,

01:20:20.920 we could do it. Even after COVID, even after our manifest failure in so many ways, with that kind

01:20:27.160 of technological advantage, I think we could do it. But we can't do it today. So it's really hard to make

01:20:34.360 the argument that the threat of natural pandemics could ever be enough to justify creating that risk

01:20:40.680 of deliberate misuse that could be so much worse because humans make better terrorists than nature.

01:20:46.600 Even if we can't make viruses worse than nature, we make better terrorists. I think the main difference

01:20:53.080 between folks who think that this kind of attitude towards understanding the threat as thoroughly as

01:20:59.080 possible, no matter what, assume that all technologies favor defense. That no matter how bad it is,

01:21:06.920 if we just know more, we can come up with some kind of effective defense. But that's just not how the

01:21:14.120 world necessarily works. And we know that from nuclear weapons. But also, frankly, we should know that from

01:21:20.440 COVID. Because we weren't able to effectively defend against COVID, even though we arguably should have been.

01:21:26.600 Certainly some nations did much better than others. But right now, you can't say that understanding

01:21:32.840 viruses better, especially given our difficulties reliably making vaccines quickly enough, is

01:21:39.080 plausibly going to mitigate the damage from any pandemic enough to warrant the kind of offense that

01:21:46.840 you're giving to individuals. One individual can launch pandemics, and the entire world has to

01:21:53.080 frantically make vaccines, test them, approve them, manufacture them, and distribute them,

01:21:58.600 if we can even do that. And then the offense can do it again. It takes so much less effort if you have

01:22:06.920 a list of many different pandemic-capable viruses to choose from.

01:22:10.120 Okay, you mentioned vaccines. So that gets to another argument that I've heard in favor of this

01:22:16.520 approach, and one that I certainly harbored myself for quite some time, which is, if we do identify

01:22:23.640 the spillover bad guys before they spill over, don't we get a huge head start on creating vaccines and

01:22:29.960 therapies to counteract that spilled-over pandemic? And if that's the case, don't we have a great potential

01:22:37.880 to really mitigate an awful lot of deaths? And we need to put that in the benefits column as we also

01:22:43.800 look at the potential cost column. In order to have a head start in vaccine production,

01:22:50.120 you need to be able to establish whether it works. And if you're trying to develop a vaccine against

01:22:56.680 a virus that has never infected a human before, the only way to test efficacy, that is to run what's

01:23:02.840 called a phase two clinical trial, would be to deliberately infect a bunch of people with a virus

01:23:08.840 of unknown lethality that we think might cause a pandemic? Now that's what's called a challenge

01:23:14.360 trial, and we weren't willing to do that even for COVID until well over a year into the pandemic.

01:23:20.600 Would we really be willing to do it for a virus that was isolated from animals and might never

01:23:26.360 actually infect a human or certainly ever take off as a pandemic? What if we discover a couple dozen

01:23:32.920 viruses? Are we going to do it for all of them? Then you have to take into account how fast mRNA vaccines

01:23:37.640 can be designed basically a day. I mean, Moderna's was famously in less than 48 hours. I'm sure we can

01:23:42.680 do it within 24 now. If you can design the vaccine within a day, and you already have production

01:23:47.720 facilities that allow you to churn out tons of doses very quickly, because again, you're just

01:23:51.640 making RNA, you're just changing the sequence. It's very easy to specify it for a new virus. Then

01:23:57.080 there's no reason why you can't run a combined phase one and phase two trial immediately. Because one of

01:24:03.880 the best things that NIH has called for and is now working on with White House support is a program

01:24:09.080 to develop one vaccine for a virus of every family. Because if you do that, and we actually have some

01:24:14.760 idea of this from many viruses, they're already ongoing, then you know roughly what dose you need

01:24:19.880 to use for your mRNA vaccine against that family. If there's a pandemic already going, you should be

01:24:26.360 making your mRNA vaccine candidate and getting it in people's arms who are high risk in order to protect

01:24:32.040 them as soon as possible. Because you already know that an mRNA vaccine for a related virus at

01:24:37.240 a given dosage was safe and effective. So given that, you wouldn't save even a single full day

01:24:46.360 of vaccine development by knowing the virus in advance.

01:24:49.480 Very few people know how quickly Moderna was developed, and I think it's an important

01:24:53.960 factual point to lay out there without any editorializing. This is just fact. I think it was

01:24:59.080 something like 342 days between when Moderna and most of the non-Chinese world got the genome for

01:25:07.160 COVID. It's 300 and something, mid to lower 300s of days before the vaccine came out. But what few

01:25:13.720 realize is that roughly two of those days were the total vaccine development timeframe, and the rest of

01:25:20.760 it was testing, safety, regulation. And the point well taken that it doesn't take long at all to make

01:25:27.160 one of these vaccines. But it's really important to note the development of the formula for the vaccine

01:25:32.920 is much briefer than the time necessary, as we've seen with COVID, to create 7 billion of them to

01:25:39.080 vaccinate the world, or even 350 million of them to vaccinate our own country. That just took many months

01:25:45.080 in the case of the US until we got to the point where anybody who wanted a vaccine could get it.

01:25:49.240 And we're nowhere near that point with the world yet. So wouldn't it be beneficial to do the deep

01:25:54.360 vision work, to find the plausible bad guys, and to just stockpile billions or hundreds of millions,

01:26:01.080 or whatever the appropriate number is, of those vaccines so we are in a position to snuff it immediately?

01:26:06.120 Well, what you just described sounds a lot like spending an awful lot of money

01:26:10.440 stockpiling doses of vaccines that we're not actually sure work yet, so that you can do what

01:26:15.960 amounts to a phase one plus phase two trial of ring vaccination to try to stamp out an epidemic before

01:26:22.520 it comes a pandemic. Define ring vaccination.

01:26:25.240 So ring vaccination is what we use to get rid of smallpox, for example, and it's where you have a case

01:26:30.440 and you essentially give everyone in the area who might plausibly have come in contact with them a jab,

01:26:36.680 just in case. You do some contact tracing if you can, but it's more like list everyone you know,

01:26:42.440 plus everyone who lives or works within 10 blocks of your home or workplace, respectively.

01:26:47.080 We're just going to vaccinate everyone to a couple of degrees of contact out, or even whole cities, if need be.

01:26:54.600 But that's still actually not that many doses. What you described sounds a whole lot more expensive

01:27:01.080 than just building the capacity to make mRNA vaccines in bulk very quickly in various places throughout

01:27:07.640 the world. I mean, we're going to have these factories for making mRNA vaccines against other

01:27:11.640 things. We're definitely going to be developing mRNA vaccines against other pathogens and probably

01:27:16.520 mRNA versions of existing vaccines because it looks a little like the mRNA versions may well be better.

01:27:21.560 Chickenpox is probably going to eventually be an mRNA vaccine because it's probably better.

01:27:25.560 And if you have factories making all these other vaccines, you can immediately switch those to be

01:27:30.840 making vaccines against some new zoonotic agent that has just jumped. And that's a heck of a lot

01:27:36.680 cheaper than having to stockpile all those doses in advance for a bunch of viruses that are probably

01:27:43.480 never going to spill over anyway. My immediate rebuttal to that, I think I know your answer to it,

01:27:49.080 but let me just make it is, if that's so easy, why COVID? Why didn't we just do ring vaccination

01:27:55.240 with a vax that took a day to make and stuff that one out in Wuhan?

01:27:58.840 Well, that would have required the Wuhan officials to actually inform the Beijing central government

01:28:03.160 that there was a problem in a timely manner, and for them to have had mRNA vaccines available,

01:28:08.680 which they still don't have today.

01:28:10.120 I can think of another immediate rebuttal to the ring vaccine strategy, but the rebuttal is so obvious

01:28:16.120 that even though I'm trying to be a good devil's advocate, I'm just going to lay it out there. I

01:28:19.560 mean, yes, it takes a day, but it took 340 days of testing and approval. And so how can we really

01:28:27.080 do this ring vaccination strategy? The obvious rebuttal to that is you're going to have the same

01:28:32.520 problem if you have 350 million stockpiled copies of a vaccine that itself, by definition, has not yet

01:28:39.240 been tested and approved because there was no pandemic against which to test it. So that rebuttal has a

01:28:45.640 built-in rebuttal. But I do actually want to add something here, and I'd like to just hear your take on

01:28:52.200 it. My personal belief is now that we do have mRNA vaccines, and we'll have more in three years, and

01:28:57.320 they'll be everywhere in some amount of time. And now that the safety profile of mRNA vaccines has been

01:29:03.960 well-established, my own feeling is when something scary emerges, we need to be able to access emergency

01:29:10.600 regulations. We can't have this 340-day test period, which, by the way, was record-settingly

01:29:17.000 quick. We can't have that because, as you pointed out earlier in the podcast, 100 days is nowhere near

01:29:23.480 enough time if something diabolical is on the march. And so this is something I've thought about on my

01:29:28.520 own. I'm wondering if you think it's a crazy idea to have ready-to-go, pre-approved, pre-thought-through,

01:29:34.360 pre-debated, pre-protocolized emergency provisions that if something really awful starts to happen,

01:29:42.920 society X, whether it's us or a country in the hot human-animal interface, can basically flip a

01:29:48.820 switch and say, as soon as we have a high-confidence vaccine, very high confidence in safety and pretty

01:29:54.960 damn good confidence in efficacy because we've been doing mRNA for X years now, we can at least allow

01:30:00.960 people to take that voluntarily rather than waiting 340 days before they can take it. What do you think

01:30:08.480 of that as a tool that whether we take the stockpiled approach or the ring vaccination approach, I kind of

01:30:14.540 feel like we need that tool? I mean, what you say makes so much sense that it hurts, and it hurts

01:30:19.900 especially knowing that so many of those 300-odd days could have been avoided given appropriate

01:30:27.180 institutional incentives that we're sadly lacking at both FDA and CDC. But it's not really to single

01:30:32.720 out those agencies in particular, because it's not like international agencies did that much better.

01:30:38.480 And to be fair, mRNA vaccines were new. It took time for the manufacturing scale up. It just wasn't there.

01:30:45.660 So even though a lot of lives could have been saved by accelerating the regulatory approval, which could

01:30:49.800 then let the companies in confidence build up even faster than under Operation Warp Speed, there is a limit to

01:30:56.700 how fast we could have done it when mRNA vaccines were new. And in future, we'll be able to do it much

01:31:01.680 faster, which means that the regulatory approval is the sticking point even more. Having a set of

01:31:07.860 people who are authorized in the event of nascent epidemic to just go ahead with a combined phase

01:31:13.460 one, phase two trial in a ring vaccination format, using an mRNA vaccine targeted to the new agent,

01:31:19.360 using doses similar to those identified for viruses of the same family in the past.

01:31:24.580 I think that's just got to be on the books as something you can do and negotiate internationally

01:31:29.320 to get approval to do it everywhere. If we agree to do it, if our FDA agrees this is okay and we can

01:31:33.940 do it, then it'd be a lot easier to get regulatory agencies in other countries to agree and just have

01:31:38.680 that as the plan. That would frankly make a heck of a lot more sense, your idea, than what nations are

01:31:45.120 currently arguing about in the World Health Organization for so-called pandemic preparedness.

01:31:48.940 Okay. I think another powerful argument in favor of knowing the precise genome of a potential bad

01:31:56.600 guy in advance is monitoring the hot interfaces between the human world and the animal world,

01:32:03.720 that human-animal interface. Those are in fairly narrow parts of the world. And it seems that if we

01:32:10.420 do identify the likely spillovers from a particular region, we can put a lot of muscle into that interface

01:32:18.760 specifically targeted at this one bad guy that might emerge from there. And we're going to inevitably

01:32:26.420 put much more muscle in there into that early detection in that geographically specific place than we

01:32:33.280 would if we never did the work that deep vision proposes to do.

01:32:37.780 I think that's inarguable that you could. It's not clear how much better it would be than just looking at

01:32:44.600 the animal-human interface again without looking for which specific viruses you think would actually

01:32:51.160 cause pandemics were they to spill over. And instead saying, which animals cause the most spillover

01:32:57.460 events in which communities? And can we work to prevent those?

01:33:02.200 So you're arguing that those hot interfaces between the animal and human world can be carefully

01:33:08.000 monitored even in the absence of the precise genome. How would that happen? And how easy is

01:33:13.820 that to do? And does it require new technology or enormous budgets?

01:33:19.260 Well, so that's what the technology that's already made deep vision's job easier is doing. That is,

01:33:24.380 they're going out there and they're monitoring people who are often exposed to animals and checking to see

01:33:29.760 which animal viruses they have been exposed to. Or even they're just getting a lot of samples from

01:33:36.220 the animals that people are most likely to contact and sequencing them. And that would let you create

01:33:41.900 a model of which creatures are highest risk. And the thing is, identifying that, say, a particular kind

01:33:49.380 of bat is high risk because of the suite of viruses to which is exposed and how often viruses from it

01:33:56.720 end up in people in regions nearby, that doesn't give anyone blueprints that could be used to cause

01:34:04.120 a new pandemic. But it does let you target interventions in communities, ensuring that

01:34:08.640 anyone who might have been exposed to a bat gets much prompter medical care and diagnostics to see what

01:34:15.140 it is they might have been infected with and resources to contain that potential outbreak before it actually

01:34:22.360 happens. Okay, next rebuttal, which I actually think is a very strong one, and which a number of people

01:34:27.960 have put to me when I raise this issue. The United States can control whether or not deep vision does

01:34:34.380 this work. But we can't stop the rest of the world from doing it. Based on what you told me about the

01:34:39.220 level of expertise and budget that it would require, there's probably not a lot of actors out there who could

01:34:44.440 do this. And there's obviously no economic incentive for any private actors to do it. But nothing's

01:34:51.020 stopping China, for instance, from doing this work. And wouldn't it be bad for China to do this work

01:34:56.980 under the cloak of darkness, for them to identify six pathogens that we know nothing about, and then

01:35:03.580 we have this information asymmetry? Isn't the danger of that frightening enough that we just can't let it

01:35:11.020 happen? And we're kind of dragged into almost an arms race? Well, you have to ask, what do we really

01:35:16.660 lose? And what do they gain from that scenario? I mean, these are viruses that are going to kill

01:35:21.660 their people as well. They're not strategically useful to great powers the way nuclear weapons are,

01:35:27.380 because they can't be effectively targeted. And you could say, well, what if they hypothetically

01:35:32.380 tried to develop vaccines in advance and vaccinate their people in advance? And there I would say,

01:35:37.800 I think it's pretty hard to vaccinate a billion people without intelligence agencies noticing that

01:35:43.980 you're doing it, and presumably getting a sample of whatever it is, or at least finding evidence of

01:35:49.780 it. And even if you somehow manage to accomplish that feat, it's going to be awfully suspicious when

01:35:55.440 it ravages every other country, but even your citizens abroad somehow never get it. That just seems

01:36:02.220 quite a reach. I think normal deterrence really operates just fine in that scenario. And again,

01:36:09.900 what do we gain from identifying it in advance? A day, when it comes to vaccine development.

01:36:16.200 If we really do have that capability of make mRNA vaccines very quickly, which is, I certainly hope we

01:36:23.160 have, and frankly, even if the government fails to invest in it, it looks a lot like the private sector

01:36:27.940 is interested in doing that anyway, because market force is to the rescue, doesn't really look to me

01:36:32.360 like we lose anything. So two responses to that. A, it would make no sense for China to do this,

01:36:38.580 unless B, they start mobilizing in plain sight. To which I'll say, governments do stupid things all

01:36:45.520 the time, even though they shouldn't. And as we can see with Russia and Ukraine right now,

01:36:51.520 governments even marshal their forces in plain sight and de facto tell the world what you're going to do

01:36:57.260 about it. So I'd feel better about the arguments you just presented if I believe that there was

01:37:04.520 a plausible path to, say, the United States deciding vehemently against deep vision, and then

01:37:12.260 basically evangelizing that viewpoint to the rest of the world successfully. Could that happen

01:37:19.100 realistically? Is there any shot of that? I think actually we are probably the hardest audience for

01:37:24.940 that one. We'd be the hardest people to change the minds of, you mean?

01:37:28.960 Honestly, yes. I think it's inarguable that if China's leadership decides this kind of thing

01:37:33.880 shouldn't happen, it's not going to happen there. True.

01:37:36.820 Whereas in the US, if we decide that this isn't going to happen and the government isn't going to

01:37:40.500 fund it, then it's actually a lot harder for us to stop the private sector from doing it anyway.

01:37:44.620 The Global Virome Project hoped to raise a couple billion dollars from government, yes, but also a lot of

01:37:50.200 it from philanthropists to do this kind of research and assemble that ranked order list of viruses by

01:37:55.440 threat level for all natural viruses using private money. It's a lot harder for the United States to

01:38:01.340 say you can't actually do that. There are some things we could do. Most notably, we could add most

01:38:07.220 viruses with a hint of pandemic potential to the select agent list, which greatly increases the cost

01:38:13.240 of working with them by requiring background checks and ensuring that physical samples are

01:38:18.380 appropriately under lock and key and so forth. That could do a lot, but we still can't actually

01:38:22.760 stop them unless we actually decided to ban the particular class of experiments required to identify

01:38:30.620 a virus as pandemic capable. That is, though, to give your example, can it find the door? Can it find the

01:38:37.840 key to the lock in order to get in? Can it take over the inside once it's there? And can it actually

01:38:44.860 take over others using the animal transmission models? If we were to say, you know what? A pandemic

01:38:51.940 capable virus can kill as many people as a nuclear weapon. We spend somewhere between two and seventy

01:38:56.520 billion dollars a year on nuclear nonproliferation. Why don't we take pandemic proliferation similarly

01:39:04.580 seriously? Internationally, there is a nuclear test ban treaty. Well, those four sets of experiments

01:39:11.720 are the virological equivalents of nuclear testing. You actually make it a national security

01:39:17.360 matter, then you treat it like a national security matter and a proliferation risk, which is arguably

01:39:21.680 greater than that of nuclear proliferation. Because again, there's nine acknowledged nuclear powers

01:39:27.920 versus tens of thousands of people that could gain access to these kinds of agents once the genomes are

01:39:33.600 online. So I think that if you can convince USAID, which I think is eminently doable, and if you can

01:39:40.700 convince NIH, which I think is much more difficult, but still possible, then we could absolutely take

01:39:46.860 the case internationally that this is in our shared strategic interest as the international community

01:39:52.380 to prevent people from doing this tiny subset, less than one percent of all virology, that is the

01:39:58.460 equivalent of nuclear weapons testing. So tell me if this is a fair summary of that detailed

01:40:05.400 response. It is definitely not in China's interest that this knowledge be discovered. They have a lot

01:40:11.080 of people to protect. This is not targeted weapons. And even though governments do stupid things all the

01:40:17.240 time, they're far less likely to do something stupid if it has been strenuously and persuasively

01:40:23.140 argued to them, hey, guys, this is stupid. Okay, here's another, and again, like, I'm not just being

01:40:29.720 devil's advocate. I think some of the arguments in favor of deep vision are pretty strong. Although you've done

01:40:35.200 a decent job of demolishing a couple of them already. But this is one that I think about a lot. I believe one of the

01:40:42.900 greatest and most securing development science could possibly conjure in the response to COVID would be so-called

01:40:51.820 pan-familial vaccines, which just brief pocket definition for those who aren't familiar with it.

01:40:58.240 The notional pan-coronavirus vaccine would immunize the lucky recipient against substantially all coronaviruses, of

01:41:07.120 which there are countless numbers. And there was talk about, and even I believe the beginning of an effort

01:41:12.500 back in 2003 in the wake of SARS, to gin up a pan-coronavaccine effort, which was understood would cost a lot of money

01:41:20.760 and take many years. But after SARS petered out and didn't even kill a thousand people, that focus was

01:41:26.280 lost. Not because it was scientifically impossible. It might be, but not because it was scientifically

01:41:30.940 possible, because it became politically uninteresting. And it haunts me to think of how different the world

01:41:37.160 would be right now had that pan-coronavirus vaccine in response to SARS been completed before MERS came

01:41:45.100 along. Several years later, another coronavirus, and obviously before COVID came along. Now, isn't it true that the

01:41:52.140 kind of virus hunting Deep Vision is proposing, isn't it going to get a lot more examples of a lot more corona and other

01:41:58.780 viruses, paramyxoviruses, et cetera, to inform the development efforts of pan-familial vaccines? Because I imagine to do

01:42:07.280 one of those things, you need as many examples as possible because you need to find the vulnerabilities that are

01:42:11.960 conserved throughout the family. That's exactly right. If you want a broad-spectrum vaccine, you need

01:42:18.220 a decent sample of the viruses within that family. What you don't need to know is which ones of those

01:42:23.720 could cause pandemics in humans. Because if you have a pan-coronavirus vaccine and it works against

01:42:29.200 a good fraction of the diversity, of the extreme diversity throughout the family, then you should

01:42:35.400 believe that it'll work for all of them. Because you're not going to find every last coronavirus out

01:42:39.060 there. You're only ever going to get a decent enough sample. So yeah, you do need to have the

01:42:43.300 genome sequences of a bunch of the coronaviruses, but you don't need to know which ones of those

01:42:48.360 viruses could cause pandemics. So here's where we will have to draw a really important distinction

01:42:52.220 between the virus hunting part, or just sequencing viruses in nature, to get an idea of what's out there,

01:42:59.840 and the pandemic virus identification, which is where you go back in the lab and you run those four

01:43:05.300 sets of characterization experiments. It's the latter that creates the problem from a security

01:43:10.000 perspective. And you can do the former without doing the latter. So if I was in charge of deep

01:43:16.480 vision, I would say, you know what? We already agreed that we would not continue to fund virus

01:43:23.760 enhancement work. Because PREDICT did fund the Wuhan Institute of Virology in not just finding a bunch

01:43:29.220 of bat coronaviruses. But they also funded research in which the Wuhan folks made chimeras of some of

01:43:36.080 the more dangerous looking ones, the ones that passed one or another of the tests, but not all

01:43:40.500 of them, and mixed and matched the pieces to see if they could make something that was more dangerous.

01:43:45.720 So deep vision, to their very great credit, has said, we're not going to fund that anymore.

01:43:50.280 And that's a very important point. And I'm glad you surfaced it. And I just want to highlight it

01:43:54.280 because there's a little bit of definitions flooding around here. Deep vision is already a step in the

01:43:58.860 right direction. They've already made one sensible step by saying we're not going to do what many

01:44:03.220 call gain-of-function. And the next step is to say the virus discovery part, the virus hunting,

01:44:10.400 is important, not just the pan-family vaccine development, whether or not it's actually

01:44:14.040 possible, it's worth a shot for exactly the reasons you articulated, but also for pan-family

01:44:19.640 antiviral development. So if I were in charge of deep vision, I would say, just like we said,

01:44:25.380 no more gain-of-function, well, we're still going to go out there and sequence a bunch of viruses

01:44:28.680 to help out the broad-spectrum folks. But we're just not going to take them back to the lab and

01:44:32.540 run those experiments to determine which ones are most likely to cause pandemics. And we're

01:44:37.000 certainly not going to add them to a list of viruses rank-ordered by threat level.

01:44:43.020 Now, just to drill down a little bit more on that fantasy situation of Kevin Esfeldt,

01:44:47.320 deep vision director, you pointed out that the characterization work is actually probably

01:44:51.520 a very, very high percentage of the budget. So if you're running deep vision and you got that

01:44:56.520 budget, what else would you do? Spend on the monitoring, not on the prediction. It's that

01:45:01.580 simple. And there's an opportunity cost to budgets. I'll just point out an obvious fact.

01:45:06.960 If they're spending 20 million of the 25 on characterization, they ain't got that 20 to

01:45:10.500 spend it on these other things. And I know that takes me a little bit out of my semi-devil's

01:45:15.540 advocate-y role, but I'm actually done with it because I've made pretty sure all of the arguments

01:45:22.760 that I've heard in favor of deep vision, some of which, as I said, aren't all that bad.

01:45:25.680 So your responses to those arguments and also your assessment of the overall situation seems so

01:45:34.680 intuitively obvious once one hears those arguments. So how is it that this program

01:45:43.400 is going forward with its stated objective of posting what I'm going to call weapons of

01:45:49.500 mass destruction to the internet? USAID deserves a tremendous amount of credit for recognizing that

01:45:55.600 one of the greatest threats to the poor comes from pandemics. And the problem came when they

01:46:01.940 took the reasonable seeming step of saying, you know what, we could target all these efforts

01:46:06.880 more effectively if we knew exactly which viruses were the most risky. I don't think anyone should

01:46:14.340 blame the folks at USAID for failing to notice this. Because first of all, USAID leadership

01:46:22.580 inherited deep vision as a program. The new director, Samantha Power, was confirmed only three months

01:46:29.280 before the announcement, which means that the program was pretty much fully established and just

01:46:33.600 needed the stamp. And what's more, even the folks who were working at PREDICT, like Dennis Carroll,

01:46:41.480 who launched the program, as far as I know, no one ever mentioned that this could pose a security

01:46:48.700 risk, let alone a proliferation risk greater than that of nuclear weapons, during all of PREDICT and

01:46:55.400 afterwards, including in folks who are super worried about pandemics and even do think about security

01:47:01.840 issues. I mean, you yourself, when you heard that PREDICT was canceled, you thought that was a bad thing.

01:47:07.340 Absolutely. So no one, I think, pointed out that this was a security risk until after deep vision

01:47:14.260 was announced. And that includes folks who do have security experience, which is not something

01:47:20.940 that anyone at USAID is expected to have, is trained to have in any way, shape, or form.

01:47:27.840 These are people who have passed up, frankly, much more lucrative salaries in the private sector

01:47:32.200 in exchange for the opportunity to help some of the most vulnerable people in the world. The poorest

01:47:39.080 of the poor, the folks who have really been left out of all the benefits that have accrued from all

01:47:44.520 of the technologies that we've developed, all of the economic growth. And they identified pandemics

01:47:50.740 as one of the things that could most harm the poor and vulnerable. And they were right. I mean,

01:47:56.800 remember, they did this before COVID, more than a decade before COVID. And they did their best to

01:48:02.880 come up with a program to prevent them. And they started out by doing the really reasonable thing,

01:48:08.260 saying, we need to know which communities are most at risk. We need to identify what we can do in

01:48:15.380 order to limit potential exposure that could lead to spillover events and cause epidemics. We need to

01:48:20.260 ensure that they have good medical care and diagnostics to identify viruses quickly, train their medical

01:48:26.320 workers in vast response, give them support for isolation protocols and everything required to

01:48:32.640 give the best chance of containing the epidemic, thereby protecting not just that vulnerable community,

01:48:38.020 but vulnerable communities throughout the world. And they did all this again pre-COVID. So that's

01:48:42.800 to their tremendous credit. So bottom line is self-evident is these arguments certainly seem to me having

01:48:49.300 heard them. They just were not self-evident until people like you started raising them. That happened very

01:48:55.400 recently. The fact that nobody, not just inside of Predict, but in society, pointed to this danger

01:49:01.900 over the 11-year history, I think, of Predict, makes it pretty clear that these are not obvious or

01:49:07.440 natural arguments to rise in the mind of somebody who is not tasked with security or even people who are.

01:49:14.260 That's exactly right. You can't expect folks who have devoted their lives to serving the poor to recognize

01:49:20.760 security risks that their work might be creating when folks who do have that kind of security background

01:49:27.200 also failed to recognize them.

01:49:29.760 Now, to wrap this up and also to bring it home in a really important way, I'm just going to point out to listeners

01:49:35.940 the reason we're having this conversation and we're getting it out there as quickly as we can is because although

01:49:42.260 deep vision has been approved, it hasn't yet launched. Is there any evidence out there that

01:49:48.040 you're aware of, Kevin, that it is underway?

01:49:50.300 It's really hard to say. There's certainly been the press release of announcing what they were going

01:49:55.500 to do, that the program existed. But there isn't anything out there suggesting that funds have been

01:50:00.080 dispersed, certainly not to begin the characterization. And remember, the characterization comes after the

01:50:05.340 virus discovery part. So even if they've begun the virus discovery, that doesn't mean they're taking

01:50:09.460 them back to the lab and running those four classes of experiments that are, again, the virological

01:50:13.980 equivalent of nuclear testing.

01:50:16.240 So it's more or less beyond a shadow of a doubt that this train has not left the station, that deep

01:50:22.140 vision's objectives might be shaped if people start thinking about them differently, or perhaps all of

01:50:28.320 its budget could be directed toward anti-malaria, bed nets, or who knows. And as a statement of obvious

01:50:34.680 fact, it's much easier to influence the shape and objective of a program before it starts than after

01:50:41.600 X dozen or hundred people are working for it and are deep into their objectives. So now feels like a really

01:50:47.140 important time to get these arguments out into the world, which is obviously why we're doing this. If anybody

01:50:55.880 who's listening to this is concerned, what might they do to try to influence folks, to try to spread the word,

01:51:02.260 etc. Well, I don't want to be irritating and say everyone should do something that would really bury someone who is

01:51:08.240 not even directly involved in this with a mess. But I would suggest, you know, despite social media being

01:51:14.100 consistently identified as being one of the high candidates for a net negative, USAID has a Twitter account at

01:51:20.120 USAID. You could tweet at them and say, this program could do a lot of good in some ways, but the security risks

01:51:27.580 inherent in pandemic virus identification seem pretty considerable. I think you should reconsider that

01:51:33.240 and perhaps move all those funds into the other aspects of the program that could really contain a

01:51:38.300 pandemic before it starts.

01:51:40.620 Yeah. And I totally agree that it is probably counterproductive to bury any particular individual

01:51:46.020 with messages on the subject. But in addition to tweeting at USAID, which is a great idea because I'm sure

01:51:52.200 that account is monitored by the folks inside, people can go to USAID.gov slash contact hyphen us

01:52:00.180 where you will find the following message. General inquiries and messages to USAID may be submitted

01:52:06.880 using the form below. They also have a phone number. And again, we don't know this factually,

01:52:13.020 but it stands to reason that this is an email account or a submission process that is monitored.

01:52:18.980 And USAID isn't like the IRS where they get literally tens of millions of consumer requests

01:52:25.160 in a very short period of time each year. So I think that's another mechanism. And perhaps

01:52:30.720 a more reliable on-ramp government is through elected representatives. If you reach out to a

01:52:37.060 representative of whom you're a constituent, they do have staff to field all of those inbound

01:52:42.600 messages. And I know this because I had quite a few friends in college whose job it was during their

01:52:47.120 summers in Washington to deal with these things. I'll give the specifics in the outro. But just for

01:52:52.040 now, if you live in Maryland, Virginia, Hawaii, Connecticut, Massachusetts, Tennessee, Kentucky,

01:52:58.640 Texas, Wisconsin, Florida, New Jersey, or Idaho, it's a lot of states, you have a Senate representative

01:53:06.320 on the United States Senate Foreign Relations Committee, on the State Department, and USAID management.

01:53:12.880 So those would be good people to alert as well. And then, of course, there's social media, blogging,

01:53:19.400 whatever megaphones you happen to have. If you feel like spreading this word, there's lots of ways

01:53:23.940 to do it. And please do so. And we thank you. Kevin, is there anything we haven't hit on that you think

01:53:30.100 is important that we should touch?

01:53:31.800 I think just the precedent that this would set. USAID has already done the right thing by stopping virus

01:53:37.040 enhancement research. If they recognize that this is a problem and decide they're not going to do it,

01:53:41.860 then that is one more step towards the US as a whole, moving away from identifying pandemic-capable

01:53:48.420 viruses and sharing the blueprints online, and thus being able to credibly lead the international

01:53:53.660 community towards something like a virological test ban treaty for pandemic non-proliferation.

01:54:00.760 Which would just be so powerful. Important as it is to do what we can to not allow this work to

01:54:08.320 happen that's being currently contemplated. If it's the start of a series of dominoes that precludes

01:54:15.340 an enormous amount of this work happening on a go-forward basis, that's profoundly powerful and

01:54:20.320 potentially profoundly curative. So please, listeners, don't despair. There's a lot of

01:54:26.040 concerning information here. But this horse is not out of the barn at all. And we may actually

01:54:32.400 be in a extremely propitious historic moment to dramatically slow and perhaps even put a stop to

01:54:40.740 the most threatening activity that we've talked about today. And that's why both Sam and I think

01:54:46.460 this is a particularly important conversation with what is perhaps extraordinarily significant timing.

01:54:54.420 So thank you, Kevin, very much for joining me today.

01:54:57.180 Well, thank you for the invitation and for, again, highlighting this potential issue of

01:55:02.400 inadvertent proliferation and what we really can do to stop it.

01:55:06.500 And listeners, please stick around for a brief moment of a couple outro thoughts and more detail

01:55:12.040 on those 12 states and who your representative is if you are moved to reach out to that person.

01:55:16.920 Okay, so that's a lot to process. But I hope you collected enough background information as well

01:55:27.760 as a rich enough sense for both sides of the debate to make your own informed judgment about

01:55:32.440 whether you share Kevin's concerns. If you do and would like to help the situation, I have a couple

01:55:37.780 more suggestions before I list those senators. First, USAID has designated Washington State University

01:55:43.680 to coordinate most of the scientific work that Deep Vision is funding. So if you have a WSU tie,

01:55:50.540 then your school or employer or alma mater is Deep Vision Central. And if you know any heavy hitters

01:55:57.080 over there, you may want to share your perspective with them. Next, after Kevin and I wrapped up,

01:56:02.460 it occurred to me that at least someone who's hearing this, and maybe quite a few someones,

01:56:06.460 probably knows Samantha Power, the head of USAID herself, or other heavy hitters inside the agency.

01:56:12.400 If you are that someone, and are deeply worried about this, then please pass on your feelings.

01:56:18.700 Or just a link to this episode to Samantha or one of her senior lieutenants. Finally,

01:56:24.960 the members of the Senate Foreign Relations Subcommittee on State Department and USAID Management.

01:56:31.440 Quite a mouthful. If you're from one of the 12 states I mentioned, here are your representatives on

01:56:36.600 that subcommittee. Maryland, it's Ben Cardin, and he is the chair of the committee.

01:56:41.800 Tennessee, it's Bill Hagerty, and he's the ranking member, which means he's the most senior member

01:56:48.360 of the opposition party, currently the Republicans. For Virginia, Tim Kaine. Kentucky, Rand Paul.

01:56:56.700 For Hawaii, Brian Schatz. For Texas, Ted Cruz. For Connecticut, Chris Murphy. For Wisconsin,

01:57:04.700 Ron Johnson. Which does rhyme. For Massachusetts, Ed Markey. For Florida, Marco Rubio. For New Jersey,

01:57:14.520 Bob Menendez. And for Idaho, Jim Risch. And that's all I've got. So thank you so much for listening to

01:57:22.520 all of this with an open mind.

Making Sense - Sam Harris - February 28, 2022

Special Episode： Recipes for Future Plagues

Episode Stats

Length

Words per Minute

Word Count

Sentence Count

Misogynist Sentences

Hate Speech Sentences

Summary

Transcript