With Nuño Sempere - Superforecasting and Global Risk
Nuño Sempere is a founding member of the legendary forecasting group ‘Samotsvety’. His track record puts him at the top of the world’s most accurate predictors of the future. I sat down with him last month to discuss forecasting, global risk, and his new non-profit ‘Sentinel’, a free early-warning system for large-scale catastrophes like pandemics, wars, and financial turmoil.
This episode is also available on Spotify and Apple Podcasts.
You can find a transcript of this discussion below [like Patrick, I’ve also added some notes, set out in this format].
This is the second in a series of discussions I recorded during my recent ‘Emergent Ventures’ trip to San Francisco. Over the coming weeks, you’ll see more talks with guests from a variety of backgrounds - public campaigners, scientists, institutional theorists, and cutting-edge AI researchers. Thanks to the Wallis Family for their unbridled enthusiasm and support, Sammy Cottrell for the superlative recording studio, and my editor, Lillo, for putting this together.
Timestamps
(00:00) Intro
(02:22) Modern Forecasting 101
(11:03) Types of Superforecaster
(18:09) Sentinel Global Risks Watch
(29:28) Institutional Uptake
(35:36) Prediction Markets
(42:09) New Zealand’s Billionaire Bunkers
(44:37) Specific Visions
(47:49) China
(56:20) Ukraine
(1:03:57) Salami Tactics
(1:15:54) New Zealand and Civil Defence
(1:21:41) Preparing for the unprecedented.
Transcript
Adam: In 2011, IARPA, an arm of the US government, found there was a select group of people that were so good at predicting the future that they even outperformed intelligence analysts with access to classified information. These are the ‘superforecasters’, a select group of a few hundred individuals with consistently higher accuracy than even subject matter experts, across a wide variety of domains.
My guest today is Nuño Sempere, a legendary forecaster whose track record puts him at the top of even the greatest forecasters in the world. Among his many credentials, he is a founding member of the 'Samotsvety' forecasting group, which Scott Alexander described as:
"...some of the best superforecasters in the world. They won the CSET-Foretell forecasting competition by an absolutely obscene margin, around twice as good as the next-best team in terms of the relative Brier score. If the point of forecasting tournaments is to figure out who you can trust, the science has spoken, and the answer is “these guys”.
Nuño works as an independent consultant and has recently founded the non-profit 'Sentinel', an early warning system for global risks like pandemics, wars, and financial turmoil. Nuño, welcome.
Nuño: Thanks Adam.
Modern Forecasting 101
Adam: I'd like to start by giving an audience that perhaps is unfamiliar with the world of formal modern forecasting a bit of a scaffold to help them understand how it might differ from a futurist that they might see at a conference or somebody that's hired to think about the future [e.g. an analyst]. Can you walk us through how your process differs?
Nuño: Right. What I'd say is that forecasting is a specific tradition of knowledge, started by Prof. Phil Tetlock, who created the superforecasting team you were talking about. It focuses on putting probabilities, specific numbers, on predictions about the world, and then checking them later. It allows you to check if you were wrong, if you were right, and by how much in a way that gives you a feedback loop that is different from the one you might see in academia or a futurism conference. Even then, experts and academics are inputs into the numbers that we put, and the probabilities we arrive at. It's just that the forecasters aggregate that uncertainty in a way that has really good feedback loops.
Adam: Sure. Just to really drill down here, I think a lot of people, even in positions of authority, with large [budgets, teams] - I'm thinking of institutions here - tend to think of forecasting as something like ‘will an event happen? - Yes or no?’ It's a binary outcome. I said yes, and I was right, or I said no, and I was wrong - or perhaps there's a date associated with it. What you're doing is quite a bit more sophisticated. You think of things in percentage terms, across let's say 100 forecasts that you make, some might be 99% (near certainty), some might be 10%, a one-in-ten chance of happening, with the idea being ultimately that across all these different forecasts and all these different events, when you analyse in retrospect the ones that you said would happen 10% of the time should happen roughly 10% of the time, right? And then you can assess who is the best at this, by essentially plotting a regression and coming up with what your error is.
Nuño: Right, here's a nice example. I was at a forecasting conference and they gave me this coin. If I throw the coin, it's either heads or tails. So then I look at it, and I see whether it's heads or tails. I assign 100% to the correct outcome. But then - you don't know which it is, so you assign 50%. This is the simplest, most cookie-cutter example of a probability. If you throw it twice, then you have a 25% chance of two heads or two tails, and so on. So you want to have a calibration. Then for events that are much more complex than these coins, and don't necessarily have that neat of a structure, like wars, presidential elections, many other things.
Adam: So, why don't you walk through from that very simple example of the coin, where the probabilities are essentially deterministic, through to something slightly more complex - let’s think about something that's a reasonably common event type (you mentioned a presidential election), let’s think about that, then maybe we could talk about something that’s never happened before, and how you might consider such an event.
Nuño: From a coin, you could go to something like a die. From a die, you could go to something like a poker game, where you don’t only have the probability of the cards, you also have think how people are bluffing or not - but even then you have a strong probability component. From a poker game, you might go to a presidential election in a system which has two parties. A good probability to start there is still 50%, in part because if either of the two parties are too far away from 50% they have more reason to compromise on their policies. So that’s a good starting point, and you can do polling, you can do statistics, you can ask people how they think other people will vote [Adam notes: a French trader famously made a $85m using this method to predict Trump’s 2024 victory - under the assumption that people are less likely to falsify their predictions of neighbours votes than their own]. You can look at people's popularity rankings and see whether incumbents have an advantage or not. So this would be in an election like the US.
Adam: Right, Nate Silver became quite famous doing this by disaggregating high-level popularity data from across the entire country, [and instead] looking at individual states (which is obviously the level at which presidents are elected). And he was doing a few more things as well, looking at quite a sophisticated model that looked at [how predictive various datasets were at predicting] previous elections.
Nuño: Yeah, he looked at previous elections. He added some factors for economics. He was modelling different demographic groups in some detail because you have imperfect polling, but you can adjust because you know which share of the population is youth, etc.
Adam: Right, some people aren’t on the phones or what-have-you, so the polling that’s published isn’t necessarily representative, but you can do some work to compensate.
Nuño: Eventually, the imperfections and limitations of that method became more apparent because reaching people that are representative of the population tends to be fairly hard.
Even then, in the US you have two political parties. You can move on [in terms of forecasting complexity] to something like Germany, which has complex coalitions where you not only have to predict the vote share for different parties, but also the possible coalitions. This still has some neat structure and historical frequencies you can draw on. You can then move on to something more subjective, like ‘would Trump declare the Insurrection Act?’ You can look at, well, it was invoked reasonably frequently after WWII, to the tune of ten times in eighty years, so it’s not that unlikely, and then you have some riots in LA, so you can update from that. You can go from there to, well, ‘what’s the probability of a really big solar storm?’, where you have seen past solar storms, and you can fit a distribution to a really big solar storm. Or, you have seen asteroids hit the Earth. They’re rare, but we have a historical track record of how large big asteroids are. You can move from there to more speculatively things like ‘will China invade Taiwan?’, will the US lose its…
Adam: Unprecedented, or somewhat unprecedented events. You and I have done a lot of work on AI forecasting, for example, where it’s very difficult to find what the base rate of what a particular thing might be.
Nuño: For AI, there still is some structure in past technological progress. You can look at a wide range of technologies, from say the Bessemer Process [which allowed the first mass-production of steel], to the evolution of trains, to the evolution of railway networks, to radio, semiconductors, and so on.
Adam: So you can look at things like adoption curves, what the production rate is like in these sorts of circumstances…
Nuño: Yeah, and you can ask yourself how many of these in the past have had a sharp jump, and how big has the jump been? Of course, that’s not going to extrapolate perfectly to AI. Technological progress has a neat feature that, indeed, it brings you new things you haven’t seen before. Maybe that’s a lens that you can look at.
Types of Superforecaster
Adam: For sure. So, one of the things I’ve noticed, and while I’m far less experienced in this field than you, I’ve been involved in a number of exercises including with Phil Tetlock… there seems to be roughly speaking three different kinds of approaches taken. One is a hyper-quantitative approach. The superforecasters I’ve talked to in this kind of way are essentially performing a role similar to what you might see in a quant trading firm, many of them are, in fact, working for a quant trading firm, either because they were already, or because they’ve been recruited by the likes of Jane Street. What I see them doing is essentially very large numbers of forecasts across a wide range of diversified fields, but where there’s relatively mechanistic interactions going on. And there are various websites where this can be done, in some cases with play money, in some cases for money, and they’re really looking at relatively small edges over a market, and winning in the aggregate.
On the other side [of forecasters], I tend to think of the archetype almost a European aristocrat, with very broad world-knowledge, and perhaps less sophisticated mathematics ability, but the ability to work through, for a particular forecast, what the world states might be like in order to bring about that event. Let’s say predicting the chance of 20% year-on-year growth in world GDP by 2050, this is a common AI-related question, and then thinking through - ok you’re probably going to need some kind of sharp discontinuity in terms of something like energy production, for example, but perhaps for some reason that’s held back, perhaps by a pandemic, and then you can go through the base rate process we talked about and multiply these probabilities together across a wide range of these potential worlds that lead to this particular outcome, and the probability is the weighted sum of these events.
Nuño: And there’s a third approach…
Adam: And then there’s a third approach, which I call (er, well, maybe there’s a fourth) I think of ‘meta-forecasting’. I think Ezra Karger, who works with Tetlock, has done a lot of work in this space. Every time Scott Alexander runs a forecasting tournament he [Ezra] is there, every year, in the top 5. As I understand it, his approach is largely to look at other people’s forecasts. You know, ‘what’s Nuño saying?’, ‘what the prediction markets on this saying?’, and we can talk a bit about prediction markets, which is another case of this where people are buying, selling - trading conditional on some future event, and the forecast the market is presenting as the aggregate of these [traders]. Ezra looks across as many of these forecasts as he can, and has quite a sophisticated model that depends on the accuracy of these different places, so it’s kind of like a wisdom of the crowds type approach, but in a really rigorous fashion. But is there another kind of methodology you’re thinking of? Does all that align with your view as well?
Nuño: Yeah. So I’ve thought about the spectrum between those two. You do have a spectrum arranging from how much data you have. If you have a lot of data you can throw some very powerful statistical machinery. You can do randomised trials across a large population. [Perhaps] Norway has some really good surveys. If you have even more data than that, you are maybe YouTube that has data from people’s clicks, you can maybe train machine learning models to make recommendations [Adam - which are ultimately just forecasts about what content to serve next to maximise user revenue]. If you have less data than that, you can still look at what the historical frequency of what something is, and make some adjustments based on judgment calls about how different things are. So that’s the spectrum that I was thinking about, but yeah, it does make sense to have a meta-approach of having different platforms that use different methods to incentivise probabilities and having some taste to apply more powerful methods that may overfit if you only have very little data - it makes sense.
Adam: Would Samotsvety, the group, be an early example of this? You’re obviously a group of you, you’re discussing different ideas, coming up independently with your own forecasts, debating them to ultimately come up with a better forecast.
Nuño: Yeah, for sure. The group is much stronger than the individuals alone, even though the individuals are fairly strong. But we will differ in terms of, say, sympathies to the Trump Administration…
Adam: Sorry, could you just unpack what you mean by that?
Nuño: Sure, if you’re making a forecast about a political event, it’s very useful to have somebody who actively dislikes the administration and is able to point out the flaws earlier, versus somebody who is sympathetic or amused, or ideologically closer to the administration who takes the other side. You also have somebody who just lives in another country.
Adam: Yeah, I’ve been thinking about this a lot. One comparative advantage I’ve had, in my forecasting perhaps, is just an extreme distance living in New Zealand from the day-to-day goings on [here in the US] and what some high-status expert is speculating about the future of some particular technology - ‘in three months it’ll be this’. Maybe he’s right, and this is back to Tetlock’s original 2011 study is that more often they’re not, or at least, you can outperform their predictions by taking a broader-based approach.
Sentinel Global Risks Watch
Hopefully, that gives people a sense of the state of the art. We could probably sit here and talk about it all day, but I’m keen to turn to your non-profit, Sentinel. I think it’s fascinating example - essentially you’re publishing these risk forecasts, a summary for decision-makers every week. There’s more in-depth work, but a lot of this is freely available online. You can Google ‘Sentinel Substack’, that’ll get you there. Why don’t you tell us a little more about it?
Nuño: So there are many topics you can forecast on. The niche that we’ve chosen is large catastrophes that could kill over a million people. In part because we think that focusing on them is valuable, and partly because other people are covering other parts of the space. So what we’re doing is processing large amounts of information. We’re processing millions of news [items], we’re trying to find the right questions to ask, we’re trying to detect what could lead to a large catastrophe. So we have some idea of what hot spots there are in the world, what geopolitical conflicts might arise. We also have a level of uncertainty and willingness to be surprised. We also expect that some of the worst catastrophes are things that we may not see coming that long beforehand, in part because, if we could, it’d have been easier to prepare.
Adam: Rumsfeld’s unknown-unknowns.
Nuño: Yeah, but you can do something Nassim Taleb is not gonna like, which is to say ‘what is the frequency of Black Swans that you have seen over the last 20 to 100 years?’
Adam: For a certain definition of black swan - at the state level for example.
Nuño: My inner Nassim Taleb is screaming because he’s saying, you can’t model outside the distribution. But you kind of can - you can think of how often society has been greatly surprised by events.
Adam: This is quite interesting. Do you think these are uncorrelated? What would a state, let’s say, do with this information? Is it purely academic? ‘we haven’t had a black swan for four years, so we need to be preparing, or what?
Nuño: Right. Let me give you a historical example of what you could do with this information. Before nuclear weapons arose, it was very difficult that they might exist - only if your physicists were able to conceptualise them. Even then, the same physicists that created the weapons were saying they might be impossible only a few years before. Similarly, with let’s say, radar, a bunch of improvements in semiconductors, flight, that kind of thing. So for me the takeaway is your threat models should contain some weight on the specific threats you may be worried about, say AI, Bio [e.g. pandemics, engineered viruses], great power conflict and so on, but they should also have a chunk on response to things we can’t currently think about. So, for example, the US recently renamed its ‘Centre for AI Safety’ within the government to the ‘Centre for AI Security’, and as part of that, they said we’re going to look at demonstrable threats - we’re going to look at cybersecurity, bio, and a couple others.
Adam: This is mainly in response to the idea of China stealing tech and so on, or making use of their own tech for drone warfare? We can get onto that later, sorry.
Nuño: Yes, so to wrap up that point. You can create a response, or you can be worried about specific threats, but if you look at the number of black swans we’ve seen over the last hundred years, there have been a fair number. So that influences our approach, and we want to be able to make forecasts on large-scale risks sooner.
Adam: This idea of black swans - let’s say we look at this frequency and one could argue it’s almost constant throughout the 20th Century, or every few years there’s some completely game-changing technology or event. Would the primary takeaway for a state ultimately be something like ‘maintain flexibility and slack in order to be able to have capacity to respond quickly to something that’s unexpected’?
Nuño: Yep.
Adam: Nothing much else one can do in that space, do you think?
Nuño: One can look at common points of failure. Like, say, food supply. What does your food supply look like? Europe is massively subsidising its agricultural sector, which, most years, is pretty wasteful, but what it does get is an independent source…
Adam: It provides some resilience against..
Nuño: No, well, Ukraine — which produces a lot of wheat, right?
Adam: Sure. One of the things that I’m… so you’ve got Sentinel, it’s published weekly, with more in-depth work published alongside. There’s a risk summary, which is an executive summary really - it’s a page or two, breaking down: this event has occurred, here is our assessed likelihood of potential follow-up events. It might be, you know, trade tariffs have come into place, and here are our expectations for how that might go, or the risk of global recession - those kinds of highlights. And this is done across global health, depending on what’s been happening, there’s an update there from economics, from global conflict, from these key risk areas.
Nuño: Mhmm. More specifically, we’re looking at risks and things that could become large-scale risks. So the niche that we have is if something becomes a really large catastrophe in a couple days to a couple months range, here’s what that could be.
Adam: Right, and you’re doing some quite interesting things, as I understand it, with data scraping of the news in order to help populate and help give you [potential risk vectors]. It’s not just a few guys in a room that happen to be really great at forecasting, you’re really setting the scene, and you’ve got some models to make sure you’re not missing important events that don’t happen to be covered by the New York Times, or your standard [social media] feed.
Nuño: Yes, so we are processing millions of news [items]. I think our setup is quite promising. I was talking with some hedge fund guys who were doing next-level stuff that they were willing to somewhat share with me at this Manifest conference, so I will be improving it, but it’s already quite potent. In particular, one fun thing that happened to me recently is that I have these general-level scrappers on the internet, and they recently found themselves. So we put out this newsletter, and my scrappers went through the internet said ‘oh, this research is highlighting…’
Adam: And it’s your own thing!?
Nuño: And it’s my own thing!
Adam: That’s good, there’s some degree of increasing salience of your work?
Nuño: Yeah, it’s that sense of an increasing salience of what we’re doing, but also that our general purpose passes are finding what we want them to find.
Adam: Sure, yeah, perfect! Like I said, if there’s one takeaway go look this up, but it seems to me that this is just like - you've got a team of the world's best forecaster (and we've established now what that means), publishing for free this information, and it's in a quick executive summary. It seems to me that like any institution, that has kind of global reach or importance, or state-level institution, whether it's a civil defense or the president, or the prime minister's office, somebody should be subscribing to this and just spending five minutes on a Monday or whenever it is just reading through and going ‘okay, yep, no, no, hmmm that’s interesting - a 10% chance? Well, I should probably start, you know, I’ll direct somebody to just go have a little look into that and see what our exposure might be. Or, conversely, there’s a social media contagion, panic spreads and so on, everybody’s suddenly talking about something. You see this time and again, and the tendency perhaps would be, otherwise, for an innordinate amount of resource potentially spend on something that is, for a few days, considered to be almost certain, but you talk to people with more of this rigorous approach, or read this newsletter, maybe you only see it's like a 20% chance or something. So you can just calibrate how much attention should be paid to these different things.
As well as, you know, black swans aren't black swans for everybody, right? They're very common in New Zealand. I don't know if you know the original story, but for Europeans, it's like, well, all swans are white until they're not. So there are information asymmetries. Chances are, with all the work and the infrastructure that you're putting in and continuing to improve, there's a good chance that there'll be times where you're aware of something, like a category of threat that someone might just otherwise completely pass by and never be considered at all [until it’s too late].
Nuño: It can be quite exciting to have a sense of all the bad things that are going on.
Institutional Uptake
Adam: You wouldn’t necessarily know from your subscriber list, which might just be an email address, but are you seeing an uptake in institutional use in this fashion, or is it mostly nerds like myself?
Nuño: So, one fun thing I can do is I can look at the domain name of an email. And so if it's a Gmail, then it could be anybody, but then if it has a domain, then I can look a bit more into it. So we have some nice people listening to it. So we have some nice people listening to it. We have some, yeah. We have some adoption across the halls of power, but not that much.
Adam: Right, right. We were talking about this before, so maybe this [podcast] will improve it on the margins, but an interesting kind of related idea to this is that access to power doesn't necessarily corrupt, but can. You know, we could see a situation where Nuño sells out because he's really passionate about some particular thing, and that affects his judgment a little bit. This forecast goes on in this way, with the expectation in your mind that that might affect a decision maker in a way that you want. And I think this is a reasonably common trend for intellectuals in some respects after access to power. It's very difficult to maintain independence, like true independence, [compared to] when you're just some guy without thinking consequentially about ‘if I say X, then Y will happen’, or anything like that. If it's just dispassionate, you're putting it out there on the internet. But I suppose the nice thing about this particular case, and it's just the degree of rigour that is brought through the modern forecasting process that we've talked about, is that you can judge in three years' time, or a year's time, is Nuño still putting out good forecasts? Is this group still actually really accurate, or is this other group of complete nobodies that have been publishing recently - actually switch to let’s them? There's a robustness to it that I don't think you don't normally expect from the introduction of power, which I think is lovely.
Nuño: Yeah, maybe my expiry date is just a couple of years, but before then hopefully I'll give some signal.
Adam: So beyond the immediate [application] - clearly people involved in national civil defence, just put it on there. Do you see that there being more applications for this project in particular or the integration of the kind of forecasting that we're talking about into institutional decision-making more generally? We've talked specifically about this kind of high-level catastrophic risk forecast. But I'm just curious to see where you see the whole field going as it becomes more salient and more integrated with institutions.
Nuño: Yeah, good question. I think many of the institutions in the space don't really have a business model. A bunch of the platforms can try to charge fees on trade. So if you are a prediction market and you're incentivising people [to share their] beliefs through financial gain, you can try to get some fees. But that's tricky because people can move to a different platform. You can try to get donations. Yeah, but that's in some sense maybe a different problem than forecasting, and so building institutions there is somewhat tricky. You can try to be a hedge fund that makes money but doesn't make their judgments public. So to some extent this is something to figure out. So this is on the, where will the field go…
Adam: Those internal to the field, what different directions are they going?
Nuño: Yeah, how are they trying to produce value, but also to capture a fraction of it, and that loop is quite interesting. And then, the other implicit half of the question would be, who gets that value? And you have a range from individuals to large institutions. So if I say that the chance of Russia using a tactical nuke has risen a bit, then if you're nearby, or if you're in the region, then maybe you want to move to the countryside for a bit. That'd be a tenable but very small impact. You can influence the smaller institutions. Maybe you have a company that is geopolitically exposed to China-US trade. You can think of large insurance companies that might often want to have models of exposure, and we're trying to do some work there. And then there's a lot, you can think of larger and larger companies, and then states. So, but to some extent, figuring out how to integrate the forecast with existing institutions and how to build institutions that can integrate themselves with people making decisions is a different flavour of problem than just making the forecasts.
Prediction Markets
Adam: So in most cases, I think at the moment, either the information is public, and in the sense of it's published by yourselves, or you can go on to a market, like Polymarket, and there’ll be some question there, a bunch of people are trading, you can get a general sense of, okay, this is the likelihood of a particular event happening. And I think the recent US presidential election was quite a good example of that. You know, the Polymarket had like a significantly higher win percentage for Trump than the general punditry.
Nuño: In part, quite interesting because a large trader made a very high volume conviction trade, which is quite interesting because it brings this subjective factor that we were talking about. It's not just the coin, it's the conviction, the willingness of somebody to put the money behind that coin.
Adam: Yes, and in that case, as I understand, they'd done an enormous amount of quite innovative work with the idea that many of the polls contained a systematic bias because people were unwilling to say to a stranger that they were going to vote Trump. And then, quite a clever approach was to do some selective polling in key states and ask people what they thought their neighbours were going to [vote]. So I suppose there's a few things here. One is, you know, it's interesting that you've got this mechanism now that these markets are sufficiently sizable and mature to incentivize that kind of, not only thinking and investment into getting an edge in that sense, but also that, you know, I suppose it's probably difficult to overstate how valuable this is to the general public, right? Whether it's companies, previously you've only got access to whatever the Gallup poll was saying, and then suddenly you've got something which is objectively and verifiably more accurate, and you've got these various incentives for people to spend quite a bit of money, but only in the cases that they feel that there's a significant mispricing, because you've got a situation where most questions are probably pretty well covered by general consensus or what have you and as a society or as an institution, whether it's a business or a government doesn't want to be putting a whole heap of resources into asking detailed questions and getting some really smart people on every single question. Instead, we've got this emerging incentive structure that is really only creating those incentives to do that sort of really innovative stuff when they feel that the market is wrong. And they get paid out if they win, and if they don't, it's really only them losing, rather than the public.
Nuño: Yeah, so I'm generally a fan of these platforms. I think that the property that ‘people who are right more often’ get a larger bankroll is really interesting. Saying that, they do have limitations. Maybe the most germane to me is that sometimes they don't have the question. So if you have a liquid market with the possibility of winning a million if you do the best research, Polymarket is going to be very good at that. But sometimes Polymarket will not have the market, and so that's where I think my method has an edge on trying to identify the risks, and putting a thoughtful probability that is explained. In markets, you have the incentive to correct the probability, but also not to give the reasoning, so that people bet against you and you can get more of an edge, whereas we just make the reasoning public. Which is something we could also improve, though.
Adam: But there are other kinds of failures in these markets, as well, right? They have like a systematic problem when the probabilities get either very low, or, very high. You end up with, let's say, some people just have like some mad conviction that aliens are going to land, for example, or this particular thing. And so in actuality, it should be 99%, not 98. But the incentive for a market participant to close that gap [is limited]. If they're buying at 98, they only stand to you know make a 0.2% return on that [actually, a 1% expected return], which, depending on how long it takes for the market to close, it might be an event that's happening in six months' time, they're better off just buying bonds from the government or something. Related to this is the fact that they really struggle with longer time horizons too, for similar reasons.
Nuño: Yeah, for sure.
Adam: So extreme events, long time horizons, which is exactly the niche you help fill.
Nuño: Well, not that long a horizon, but on something like the two-month scale, it's now considered long term by hedge funds. Like - “we're working a long time play”, and it’s two months.
(laughing)
Adam: Sure, but I guess the other thing would be just very short time horizons as well, right? Like if something's happened, you can make an assessment quite quickly about such and such, whether something some follow-up event will happen this week, which might be critical for a decision-maker, but a market by the time, you know, it's being created, people become aware of it, a trading in it, and the price reaches some equilibrium. That might take weeks.
Nuño: Yeah, possibly.
New Zealand’s Billionaire Bunkers
Adam: All right. So to turn now, besides your nonprofit, which, you know, people should donate to [or become a paid substack subscriber], you also do some consulting work as well. People can hire you for, I think it's like, currently, a hundred and fifty dollars now. But maybe that's going to rise rapidly.
Nuño: Yeah, it has been rising. At the same time, the government of New Zealand is very welcome to reach out. It's geopolitically in a good position to survive some of the ‘out there’ catastrophes. It is closer to China, but at the same time...
Adam: We're very far away actually. We're not closer to China. It's basically equidistant [to China and] the US. The Pacific is absolutely enormous, but it [China] is our largest trading partner. They buy a lot of our milk and dairy products. Sorry, to interrupt.
Nuño: No, it's in a good position to weather wars. So some of the long-tail events I'm worried about. Peter Thiel famously has a couple bunkers there, right? A bunch of billionaires do.
Adam: Yeah, Sam Altman, I believe.
Nuño: Yeah, my one business model I was thinking about was, I would love for these people to pay my forecasting community to run continuously, to just tell them “you should go to the bunker”. If you're paying for a bunker, I'd love to get that service.
Adam: Yeah. But you've been doing small-scale stuff for people as well, just like kind of advice about what they should do.
Nuño: Yeah, so I think this is good for the market discipline, just ‘am I providing value to people?’. At the same time, what I really want to be doing is building this institution up. I want to be building Sentinel up. I think it's a good niche. I think we have some good core forecasting. I'm juggling the funding and the distribution, the business of readers, the people at institutions that could use it, the forecasters, and the people we employ. Yeah.
Specific visions
Adam: So to sort of change tack a little here, what's one thing you disagree with like a current broad public consensus on? Like what's a really salient one where you think everybody else is wrong?
Nuño: But no, right? If there is a strong public consensus, I am going to give something. It would be hard for me to give less than 10% to it. That said, I tend to think that people's pictures of the future are sometimes too specific. So there was this recent AI 2027 exercise, which I thought was great, but because it was conditioning on a lot of stuff, I think that the probability of getting that particular scenario (without any crucial considerations that don't arise) is lower than some of the people in my cluster, which isn't necessarily the greater intellectual sphere.
Adam: I think this idea of people being way too crystallised about [a specific idea of] the future is extraordinarily common. Both in general society, there's some degree of probabilistic thinking that's present in almost everybody of like, ‘oh, I think this is gonna happen - but I'm not entirely sure’. I think they understand that there's some range of possible outcomes, but I think that maybe it's particularly in this day and age where given just how much change we're seeing, that really people should be a lot more fuzzy about the exact shape of what that ends up looking like rather than specifically pointing to this and this [highly specific prediction].
There’s a very classic futurist conference sort of thing, just as an example, which people love, you know, because often these people are quite charismatic and paint a really vivid picture of the future, but it's always like: we're going to see autonomous vehicles in every city and people won't be allowed/able to drive normally because they all have, you know… you and you can construct this scenario which is like eight or nine different layers of what's gonna happen with the street lights and regulation and then so on and I think it's probably true that pieces of this puzzle are quite likely, but that the combined probability of all of them together is actually extraordinarily low.
Nuño: But in some sense, I do have sympathy for that. Painting and conceptualising different futures is still fairly valuable in the way it will fall into my models. But if I have to think about the actual uncertainty that there is, to some extent, a different question, right?
China
Adam: All right. I was keen to talk about China. We raised it earlier, I think in the context of Taiwan, you know, the forecasting conference that we've just been at, [it was] one of the perhaps primary topics of discussion - there seems to be quite a consensus here… maybe that's one answer to the question I asked earlier [about where the public view is wrong] - I think people here in over the last three days seem to be much more bullish on China's near to medium-term prospects than maybe the world as a whole, which I think has maybe still caught up in what was true perhaps 20 years ago. Friends of mine still have a model of Chinese manufacturing as being extraordinarily labour-intensive, and that's the source of their competitive advantage. Whereas, in fact, you're talking about a situation here where it's the world's centre for manufacturing engineering and engineering excellence. They've got an extraordinary number of extraordinarily talented engineers and modern factories that are (I think this is perhaps still somewhat of a meme in general, but they’re getting there) dark factories, where raw materials come in and finished products come out. But I just wonder, you've obviously had a whole host of conversations on the topic and made a series of forecasts, including with Sentinel on China. Can you give us a bit of a sense of, across economics, and military, and so on, where you see China going in the next three to five years?
Nuño: Right. So to firm the discussion, the probabilities that I've seen across various forecasters and at this conference are, very roughly, in the 10-15% to 60% range that China will invade Taiwan by 2030. Yeah, so it's not 0.5%. It's not 95%. Yeah, it's in that range. So I came here in the lower range of that, and even then, part of my probability was that if China has the capability of invading Taiwan, it may not have to. So even then, I'm not saying that China is [weak], I'm talking about the specific operationalisation of boots on the ground.
Adam: Right, right. Because one of the very threatening modes of war for Taiwan, I think, is just blockade. You know, they shut down their nuclear plants. They're almost entirely reliant on imports for energy and food and so on. Obviously they export an awful lot of manufactured goods, but as far as daily living is concerned, incredibly vulnerable to blockade. And with the changes that we've been seeing in the last, even this year, such as in the Red Sea, the vulnerability of, up until very recently, almost kind of unquestionably supreme US naval power. For people who aren't aware, there's a series of cases where an F16 was shot down by Houthis…
Nuño: I don’t think it was shot down, it fell from the aircraft carrier [into the sea] because the ship was under attack and they had to take the maneuvers so that the missile that doesn't hit.
Adam: Sorry - yeah, and this happened twice, right? And then that led to the carrier group essentially withdrawing from that particular theatre.
Nuño: It's a bit more complicated. The US did a pretty large bombing campaign. And it spent hundreds of millions - a few hundreds of millions maybe. I did the math at some point, but a lot of money bombing Yemen. It deployed a whole lot of its reserves of munitions. You did have this incident. I wouldn't characterize it as, you know, now that the US name is weak.
Adam: No, no, no, but it seems to me almost certain that the era of *unquestioned* US naval superiority is over.
Nuño: I mean, I think that will be the case when somebody sinks an aircraft carrier. But even then--
Adam: But this is like the Houthis in Yemen we’re talking about here, we're not, I just can't see that the US would have the same approach to parking a couple aircraft carriers near Taiwan and just expecting that that would be enough of a deterrent anymore. I kind of struggle to see it given the capabilities of China are significantly greater than that of the Houthis. And then we also saw the incident in the recent conflict between India and Pakistan where the Chinese jet shot down the, you know, the Rafael, the French-derived ones, which do share quite a number of similarities with the US equipment that is currently in service.
Nuño: I think you do want to go for a big picture of like… I think some other salient factors are - I think China has been taking costly steps to increase its readiness. It has been building new technology to be able to land in different parts of Taiwan and as invasion barges, it has the technology to cut underwater cables more efficiently. It continues to integrate the military and civilian infrastructure. Overall, it's building the capability.
Adam: Whether or not it uses it, or it's just a stick to threaten with, yeah.
Nuño: Yeah, for me another salient aspect of the Houthi-American conflict wasn’t so much the F16 falling over as: having more hotspots that distract the US. So if the US is contributing to the Ukraine conflict, and it has to support Israel against Iran, and maybe there's some third hotspot, maybe Venezuela invades Guyana or maybe..
Adam: Is this purely [hypothetical], or have you done some work in that particular area? I hadn’t heard anything about that.
Nuño: I don’t think Venezuela invading Guyana is particularly likely.
Adam: [You’re saying] it's across the set of all of those possible things, it's actually reasonably likely that some kind of third hotspot arises of interest to America.
Nuño: I’m tracking the possibility. Guyana has some oil..
Adam: It’s seeing enormous growth, and US companies are deeply involved in it.
Nuño: For sure, and in a section that Venezuela either wants or disputes. I don't think that it’s going to be there, but it's another [potential] hotspot. And so if the US attention is divided across different places, and you get maybe some instability in the US, maybe that would be a good moment to make. Yeah, I'm not saying anything hugely revolutionary here. All of the Russian, Iranian, North Korean, and the Chinese economies are all getting more integrated. So maybe a good, maybe a high-level trend to give you mind is whether there will be a bloc that gives some healthy competition to the US or whether they will integrate themselves with the Chinese, and will there be a hard conflict there?
Ukraine
Adam: You mentioned Ukraine. I imagine you'll have seen the images of fields that look like they're covered in silk or something, of just all the fiber optic cables, just absolutely covering the ground almost as far as the eye can see. You know, this is one photo I've seen, which I assume is representative of a lot of the front, but might not be. But let's just take a quick step back. That's a conflict that started in 2022, in a sort of almost vaguely World War II style, in terms of military affairs. We're talking about artillery. Trenches came a little later, but you're're talking about this bewegungskrieg [maneuveur warfare] kind of approach of the Russians early on - we're just going to absolutely rip into there, come into Kiev, but then that fails because of US [assistance] and because of some heroics at the airport outside [Kiev] to stop the landings.
In any case, it devolves into a quasi-World War I-style artillery and HIMARS [rocket artillery] conflict for a period. And then maybe around last year, you're talking about this sort of exponential growth in the usage of drones. And then we've seen this sort of like measure and counter-measure occurring where you know, you use electronic warfare, so, you know, to block the use of first person drones wirelessly, so they fall out of the sky, and in the response to that being utilizing fiber optic cables in a big spool, so that you basically, you're not wireless, you're just flying this thing and it's got a very thin little thread [trailing behind]. And now, like I said, you're in this position where it's just like - that's the war. The nature of the war is primarily a drone war [70-80% of casualties suffered by both sides are now caused by drones] with both sides rapidly ramping up production. You're talking about a complete revolution [in military affairs]. So there's a few questions that come from this. Do you see that as being the end state of that conflict? Is there another revolution in the nature of that war, or war in general, coming? Autonomy, for example, is something that's been mentioned in these [forecasting] circles. Are there some other technologies that you think possibly likely to change again, the nature of the war? And then, of course, quite dramatically recently, there was a sort of mother truck deep into-- behind Russian territory, destroying a number of strategic bombers. It strikes me that, or it concerns me, that if this war continues, you may end up with a degree of total war, unprecedented in human society, where it doesn't matter how far behind the lines you are, there's gonna be potentially drones flying around taking out bits of infrastructure and so on. Up until now it's been, with the exception of strategic bombing, relatively contained on the front whereas I don't know what it does to people psychologically. But anyway, I'm curious how do you see this conflict evolving?
Nuño: I'm not sure. One aspect that's salient to me is the similarity with the Spanish Civil War, where you have different sides testing their own technology, which then diffuses after the conflict is over. So I'm not going to be able to predict what the next turn is going to be, but I can make the meta-level prediction of: you are seeing rapid innovation in war, as you saw with radar, with submarines, with… as you see in wars, because people have a real need to cut through the red tape and Ukraine sees this as existential. So you can make the meta-level of a prediction of, OK, what happens with the diffusion of these technologies, what lessons are different players taking from the field?
Adam: So diffusion of technology. Right now, most of that expertise is highly concentrated in the conflict. Yeah. China is producing a lot of drones, but they won't necessarily have the knowledge of that kind of in-depth understanding of how to use them strategically and so on.
Nuño: Well, I mean, it's close to… But Russia, will, right?
Adam: Right, right, right. So that's what you mean by a diffusion of both the production capabilities specific to war drones (as opposed to surveying), and also how to use them at the operational level, and the strategic level.
Nuño: But then also if the conflict ends, the production capability will remain and there will be a profit motive to sell to other countries. Maybe Ukraine to the West. China maybe will use their own.
Adam: Aren’t most of the Russian ones produced in Iran?
Nuño: In Iran, yeah. I'm actually not sure what proportion is produced in Iran versus China. China is supplying to both sides, which is very amusing to me. Something I'm also worried about in the background, is what will happen with conflicts in Africa that are just going on constantly. Russia has used its Wagner group, which has now, I think, [been] called somewhere else, to intervene systematically in the region.
Adam: Particularly in the North.
Nuño: Yeah, Mali, then separately, also is now [trying] to prop up its presence in Libya in order to maintain access to the Mediterranean after potentially losing the Syrian base. So it's interesting how these conflicts can globalise themselves, right? France is also saying, Macron is also saying that if North Korea keeps supporting Russia and Ukraine, maybe France will support South Korea, which also makes total sense, but...
Adam: Well, how would that look?
Nuño: Technology transfer maybe...
Adam: Yeah, it's interesting. As I understand it, South Korean military technology is broadly comparable with anything that the West is producing. I know Poland purchased their tanks from there recently, perhaps somewhat controversially, instead of from Rheinmetall in Germany. Is it just rhetoric from Macron? I'm not sure, I’m not following that, I'd have to look into it.
Salami Tactics
Adam: So I suppose, you know, we've talked about Taiwan, you're talking about a 15 to 60% chance…
Nuño: In some sense, it's a wide range. You wouldn't see that spread in the prediction market, but then it's also a longer-term market. So there's something to say about how to synthesise the existing information, but at the same time, it is in some sense reasonably close together.
Adam: So, but whatever the percentage is, it’s very clear that the likelihood is significant, and that, given the operationalization you were talking about earlier, is only proportion of the total risk - 'cause you know, you're talking about boots on the ground, you'd have to add at the very least the blockade to that, which would also be an act of war. I've heard quite strong arguments that, strategically, that [approach] would seem to make a lot of sense.
Nuño: With blockades, it’s quite interesting because one thing we have seen in Russia and China is that they're good at the salamis slicing or this boiling the frog strategy of not giving the other party a ‘Pearl Harbour’ that it can easily coordinate around.
Adam: So this is like the cutting of the cables that we're seeing, the increasing incursion into Taiwanese airspace. While I was there, there was a missile which went over the top of the country and there was a warning out to everybody on their phones. It's all normalising.
Nuño: Yeah, normalising it, but then also forcing Taiwan to be on higher alert more of time, which is causing fatigue for the sailors and so on. It's interesting because you could see a straight up blockade, but then you could see things that are ambiguously a blockade. And then the US, maybe does a freedom of navigation exercise, and then the semi-blockade there ends, or doesn't end. You can see inspections [of commercial vessels, thereby disrupting trade]. You can see minimally possible deniable cutting of cables.
Adam: And this is happening, not just in Taiwan, but also particularly with the Philippines, right? Like in those various, not far from the Filipino mainland, these islands, and it's like use of water cannons and all kinds of things, like ramming the Filipino coast guard ships.
Nuño: Also cyber attacks. So the US is not going to declare war with China over a cyber attack, but at the same time…
Adam: Yeah. You draw this together, and it's going to be a significant risk there of an increasing… an escalation of what we're seeing. How should we think about how this how would [develop]? There's a lot of talk about essentially a domino-type effect.
Nuño: Yeah, so I’m focused on the domino-type effect, because I’m focused on the large-scale risks. At each step, I want to, the question that I'm asking is: ‘how could this scale up into a global conflict?’ In the case of geopolitics, this looks like me looking at the way in which you could get, in which conflicts could become global and you have World War Three. You could also get things that are much more local. Indeed, you have the...
Adam: Already ongoing with border disputes in the Himalayas with India and those kinds of things. And that just escalates a little bit, but it's always this kind of Cold War. Like, it doesn't ever… yeah, some lines aren't crossed. One of the arguments is that any significant change with Taiwan is going to leave Japan and South Korea, Philippines obviously, and Vietnam in a really precarious position. There are islands claimed by each of those nations that China also claims and were Taiwan to kind of… I suppose the peaceful reintegration scenario, where maybe the population in some respects maybe is just compelled to capitulate peacefully - those sorts of scenarios don't occur, but to the extent that it's some kind of more obviously coercive quasi-war scenario, you'd be thinking that those nations that I mentioned and India would be significantly raising their own readiness and almost belligerence, attempting to recruit, hard lines in the sand from the US about if such and such occurs.
Nuño: Yeah, you could see that the scenario, you could also think about the scenario where these countries want to be more integrated with China as its manufacturing capability rises. And they become more dependent. The US tries to set up red lines but fails.
Adam: Sure. Yeah, we saw quite a bit of this, in Syria, for example. Interesting. Coming back, we've talked a little about the military situation, we talked about how there are maybe some misconceptions about the nature of Chinese manufacturing and so on. But I think, in general, there's a lot of clinging to particular ideas around like local government debt, and an aging [Chinese] population, and a housing kind of bubble that's popped recently, and so on.
Like, you know, China is very obviously not some kind of Titan bestriding the world on the one hand. It's got its problems. You know, it has a range of internal tensions. You know, there’s some extent to which President Xi is a single point of failure that creates this kind of… really, really dumb things happen at times because of the nature of the bureaucracy. Like with the Shanghai attempts to be zero-covid long after, you know, that ship had long since sailed, but people were bolted into their apartments.
But the flip side of that is, because I think those sorts of things can often be quite widely disseminated on social media and so on, because they're quite dramatic. The flip side of it is that actually, economically, China's in a very healthy position and is producing all kinds of value that you're saying that the surrounding nations and Australia, New Zealand, obviously are keen to be a part of and continue to have access to Chinese markets and so on. And if the, depending on the current trade situation develops with US tariffs, a lot of that could really end up pushing those much closer with much closer economic ties to China, right?
Nuño: Let me give you some pushback on, yeah, I think you have in mind the picture of a listener that thinks that China’s gonna fall tomorrow because they’re a paper tiger.
Adam: Or how many communist regimes are very strong on paper, and maybe not nearly so much in practice.
Nuño: I have sympathy for that. Like, it’s hard to see regimes falling before they do. So maybe the forecasting perspective will be, okay fine, so you give a 1 to 10% chance of it falling over the next however-many period, and then the communist party just falls. Fine, this doesn't seem like, maybe you disagree with that, but don’t I think it's crazy. So, yeah, maybe somebody disagrees with you on this fundamentally - the forecasting perspective would say, yes, you know, like, assign a 10% chance to China falling in the next five years in the next decade right maybe because of dynamics were not seeing, because we just don't have the knowledge, maybe because the transition of Xi is botched right.
Adam: There was an interesting fact there that I learned recently. The two presidents before Xi were hand-picked by Deng Xiaoping. So it was kind of a ‘great man of history’ type character, almost like a second revolution within the revolution. And so there hasn't really been any succession before Xi. The system wasn't particularly well catered to do so before Xi, and Xi has dramatically reduced its capability. It's just one example anyway.
Nuño: Maybe you give a 10% chance to, the Communist Party kind of collapsing. It could also just change its name, right? Or sort of leave formalized while its system is…
Adam: Like a sort of a Stalin-to-Khrushchev succession?
Nuño: You could have also a Khrushchev to…
Adam: Brezhnev?
Nuño: Who was the person before Yeltsin?
Adam: Uh, there was a few [Ivan Silayev immediately preceded Yeltsin, but held office for only four months in late 1991. Prior to that, Valentin Pavlov was prime minister for roughly eight months following the dissolution of the Soviet Union].
Nuño: Yeah, a few of the old ones, right? You could get a few of them.
Adam: Right, right. There's this classic Pope scenario where everybody elects… they select a guy that's seen as being the least likely to be elected, and he's really old anyway and so we'll just bide our time and then we’ll [have our moment], and this happens in a series of elections [e.g. the compromise election of Pope John XXIII, who was expected to be a mere caretaker Pope, but ended up delivering one of the most significant reforms of the Catholic Church in the last 400 years].
Nuño: I haven’t thought a whole lot [about this], but I would have some sympathy to something in the 1-10% range, but then I would also have some sympathy to something in the 10 to 60% range of China invading, and in a big chunk of that it pulls it off, and then another maybe like 5% in which China blows everything out of the water.
Adam: Right, right. But I suppose to my mind, leaving aside succession problems or some internal black swan type event that means that the support for the party undergoes some sort of preference cascade where suddenly everybody's like, ah, actually, the hell with them - it seems that the country as leaving aside the political situation, the country economically, socially, even demographically, is very unlikely to be undergoing some sort of sudden USSR type collapse anytime soon. There's empty shelves and massive [shortages], these sorts of scenarios, perhaps are more popular than they should be. Would that be fair?
Nuño: I'm not sure. Yeah, I think to an extent you want to attack somebody else's big picture, but you propose a big picture of your own that I don't find convincing either. I’m sympathetic, but I would inject more uncertainty.
Adam: Sure, well, I'm just trying to paint for listeners an alternative viewpoint. I'm not… yeah, okay. Fair enough.
(laughs)
Look, I mean it was there anything else on China that, you know, you've been thinking about that you think, or more generally in that geopolitical space that you think we haven't talked about?
New Zealand and Civil Defence
Nuño: I'm curious about New Zealand, right? So you're more integrated with the institutions there.
Adam: It helps that we’re so small. It's kind of like Singapore, but if it were spread out across the entire U.S. Eastern seaboard. So what can I tell you?
Nuño: You tell me that it has already some infrastructure to deal with earthquakes, natural disasters. How do you think that, how does it look like for more ‘out-there’ possibilities?
Adam: So, New Zealand is tremendously prone to civil defence disasters. We live on the Ring of Fire. Earthquakes happen. It’s tremendously geologically active. There are volcanoes. We're in the midst of an enormous ocean and cyclones are not uncommon. But more than cyclones, I think, we don't tend to get the sort of category five issues in New Zealand, of like really high winds and so on, but rather just enormous dumps of rain that cause widespread flooding. So these are all reasonably common, which gives us in a funny sort of way, quite an advantage in that civil defence disasters in general are almost never black swan, right? It's only been a few years, perhaps, maybe 10, since the last time that the local or the regional group has had to be spun up and we've had to rediscover how the systems work and so on.
Nuño: We had been in Spain with flooding in Valencia. We hadn't had that, and so the systems were just rusty.
Adam: Right. So I think we're ‘lucky’ in a sense that this is not the case in New Zealand. You don't tend to have... I think the real difficulties with institutions often are generational in nature where you can get.. as long as people within an institution learn from people that had been there, you know, it's not ideal but you know you're always going to get some degree of retirements and and people moving on, and so on. Whereas it's very difficult for the subsequent generation to learn from the people that have never been through ourselves either, and so you get a general atrophy of capabilities.
I think that the response to COVID, for example, is quite an interesting one. We were one of two nations, along with Taiwan, that became COVID-free. And I was initially… this surprised me quite a lot, to be honest. I had been banging the drum, I've been paying attention to forecasters in February 2020, and attempting, I was a much more junior at that stage, but attempting to sort of raise awareness “where we need to deal with this” and I was literally laughed at. ‘COVID, ha, I've got more important things to worry about’. So I was quite skeptical about our capacity to respond as we [ultimately] would and markets are crashing internationally. Italy was under this enormous lockdown and we're sort of merrily going about our day as news comes out that there's been some super spreader event at some wedding and people look like… And meanwhile, literally, the civil defence [is mostly idle]. I suspect national people in the national centres where we're beginning to, by that point, say we should be thinking about this, but at the regional and local level they weren't activated. Nobody was preparing, but so yeah, like I said, I was skeptical but once the decision was made actually we were able to like spin up within a few days.
Yeah, there were problems. Technically, well, not just technically, but the initial decision by the Prime Minister Jacinda Ardern - the stay at home order was actually later ruled to be illegal. But, in general, the nation kind of came together. We had not only the institutional capability, but the nation as a whole is sort of through it often enough that, well, ‘okay, what are we doing now?' okay, this is happening.’ Many people would have had stories prior to then, of being in an evacuation due to some event, like a storm event or something, or had friends or family that had been through it. So the idea of it all was pretty straightforward for us, even though the actual idea of locking down in the pandemic and what does that mean operationally in terms of food and all was completely unprecedented. We were able to do it well enough that COVID, that initial strain disappeared, And we were having big New Year's parties while the rest of the world stayed inside, which, you know, a certain degree of civic pride there, albeit many people in our position often are, there was a degree of overconfidence there. These new strains came out that were significantly more contagious, and popular will to like, ‘okay, we're just going to do this forever?’ begins to erode. There was a series of lockdowns that were reimposed, particularly in Auckland, and government (which was earlier reelected with a sweeping majority in [November] 2020) popularity began to tank quite quickly.
Preparing for the unprecedented.
Nuño: Yeah. So here's my reaction to this, which is seems like you do have experience with disasters on the five to ten year cadence. But then that seems like a good foundation for considering also worrying about scenarios that are bit less likely than that because you’ve access to that infrastructure. So then there is there is a long-road down, to unlikely, almost peculiar, stuff. You could look at solar flares, you have the Carrington Event - this was a very big solar flare!
Adam: As I understood it, the aurora was visible in New York during the day at the time.
Nuño: Yeah, it fried electrical cables. I think an event like that probably wouldn't destroy the electrical grids - disrupt them, but probably not destroy it. But it's not clear to me and, you know, having backup transformers just seems very possibly good. Though, at the same time, they are expensive and hard to get, but if you prepare for that in advance, that's something you can do.
Adam: The people that I've talked to in New Zealand around this particular scenario have actually suggested we're really well prepared for it. Which when I learned, it seems so esoteric.. ‘oh solar flares’.
Nuño: I'll follow up, because I am uncertain about this, and in particular, are you prepared for the largest event that has happened so far, or are you prepared for an event that's a bit larger? And if so, how much is that range? That's an interesting question.
You have COVID, but then from COVID to a Black Death, there is still some way. The Spanish Flu was interesting in part because it was selected for virality because, let me see if I get this right - oh yeah, no, no it was selected for virulence and didn’t care much about the death rate because it was in military hospitals. Because if you kill people where…
Adam: If you kill people before they can pass it on?
Nuño: No, no, if you kill people in a dense military hospital in a way that allows you to spread the disease more, the origins are quite packed together. That might select for both virility and deadliness, whereas normally, if you have a strong disease, maybe don't go out to a party, so it selects for mildness.
Adam: And specifically military hospitals, we were a lot of these people are on the edge of death anyway, and it's also viewed by authorities as a co-morbidity, to the extent that's noticed at all, rather than oh gosh a whole heap of heap of people are kidding, really second dying, we should take action.
Nuño: Yeah. Yeah. And so, this is not [typical] natural selection, it’s sped up through World War II.
Adam: One.
Nuño: Oops. Yeah. Oh, wow. I keep… Yeah, you can imagine versions of this [viruses] that are more out there. You can imagine AI creating a more ‘Black Death’ kind of pathogen…
Adam: Or enabling some particularly moustache-twirling villain to do so.
Nuño: Yeah, so you can think of complete black swans - how do you prepare for a complete black swan? It’s a tricky question. I was talking with a fiction writer here at this conference, who was saying, okay, I have a nice scenario that is really out there and extremely low-probability, but are you sure that your audience hasn’t thought about? He was proposing running a war game on invasion from another dimension. The probability of that physically is negligible. But it's good as a scenario you haven’t thought about. Food for thought. Like, I’m not worried about invasion from another dimension, right, but…
Adam: Like, as a way of considering preparedness, you might consider something absolutely absurd. And just as a way of working through the idea of something equally absurd to us now or outlandish, but that actually happens.
Nuño: How do you prepare for the invention of nukes, you know? Maybe you prepare by, you know…
Adam: Yeah, that’s really interesting. Well, look, I think we're coming towards the end now, but is there anything else you… like, last burning thoughts, on forecasting, on risk, that you wanted to cover off?
Nuño: Yeah, I think the asking what to make is: if you have something really good infrastructure for dealing with natural disasters, considering expanding it to scenarios that seem a bit more of a landish or speculative, but that it might still be good to perform.
Adam: All right. Nuño Sempere, thanks very much.
Nuño: I will give you a fist bump. Cheers.
Adam: Cheers.