Chaos to Order: Doyne Farmer on Complexity Science and Economic Transformation

[MUSIC] You're listening to a podcast by the BCG Henderson Institute, DCG Stank Tank. In this series hosted by fellow Dave Young, we'll interview business leaders and explore how companies can build competitive advantage by creating a sustainable world. Now on tour episode. [MUSIC] Welcome, I'm your host, Dave Young. In this episode of Building Competitive Advantage in a Sustainable World, I'll be exploring how advances in complexity economics and agent-based modeling can help us better navigate the energy and sustainability transition for a healthier, competitive and just economy. I'm pleased to discuss all of this with a visionary pioneer in the field, Dome Farmer. J-Dome Farmer is an American complexity system scientist and entrepreneur with interests in chaos theory, complexity and econophysics. He's the director of the complexity economics program at the Institute for New Economic Thinking at the Oxford Martin School, a professor of complex systems science at the Smith School of Enterprise and the Environment, the University of Oxford. And an external professor at the Santa Fe Institute. He's also the chief scientist at macrocosm. Dwayne has just released his new book, Making Sense of Chaos, A Better Economics for a Better World. So I'm thrilled to welcome Dome 2, Building Competitive Advantage in a Sustainable World. This is terrific to have the chance for you to be with us and sort of share some of your latest thinking and research. I know everyone out there will be eager to hear more about your story and what you're working on. And so if you wouldn't mind, just introduce yourself and the journey you've been on professionally and then I'd love to go into some of your thinking and some of the research you're up to and the book that's coming out and lots of exciting stuff. So I'm Dwayne Farmer. I had a somewhat unusual career. Started as an astrophysicist. Then in graduate school, built the Curse World of Computer to beat roulette, which then changed my focus of attention and I got entranced by what's now called chaos theory. And so I did that for my PhD thesis. We developed methods for determining whether data was being generated by a chaotic attractors or not and used that for forecasting purposes. And I worked at Los Alamos National Lab for 10 years. While I was there, I started the complex systems group at Los Alamos and also was involved with the Santa Fe Institute and early days of setting that up. When I was talking about my forecasting work, somebody always asked me, have you tried applying this to the stock market? I got tired of saying no. And so then I started a company called Prediction Company that was based in Santa Fe. We were one of the earliest quant funds. We did proprietary trading for what eventually became UBS. And I left in '99 to go back to my passion for basic research. They were sold to UBS in 2005. So then I spent 12 years at the Santa Fe Institute, shifted my focus of attention to economics and finance. And 12 years ago, moved Oxford where I'm the director of complexity economics for the Institute for New Economic Thinking at the Oxford Martin School and a really different professor of complex system science in this school for enterprise and environment. You know, a lot of people may not know Santa Fe Institute. Would you say a few things? And what was sort of the founding idea behind it and why was it needed? Yeah. So the Santa Fe Institute was founded by some of the senior fellows at Los Alamos, including what George Cowan was the main one and with help from people like Murray-Gleman and Dan O'Lom initially. And so the idea was that we really need a place where we can understand complex systems better, which at the time nobody even knew what that was. But it was clearly very interdisciplinary thing. They thought it was better to not have it at Los Alamos because Los Alamos is in the background of weapons lab. So they decided to do it in Santa Fe. I think they also liked to be excused to spend a lot of time in Santa Fe. So we set up in 1984, really got going with a real campus in 1987. And has an interesting structure in that there's only like eight people on the residential faculty, but a much larger community that's now grown to about 100 on the external faculty and was a place where complex systems as a scientific movement was really born. And you mentioned chaos as well. So you've kind of been at the beginning of a number of sort of important new ideas, chaos among them. What's the relationship between the chaotic systems? And complexity. Chaos is a subset of complex systems. I think complex systems is about information and about structure and form. But chaos is a remarkable thing because you have a deterministic, dynamical system, which you would think should be predictable, but isn't because of its nonlinear geometric structure. And so chaos is a subset of the broader field of complex systems. It's about turning order into chaos. A lot of the rest of complex systems is about turning chaos into order. One of the things you understand when you even look at chaos is there's structure that can be preserved. My thesis was called order within chaos. And I studied chaotic systems that preserve certain components of order very strongly, the same time that they made other things. Chaos. But the big question that complex systems still hasn't solved is how does self-organization happen? How do you get life and intelligence and things like that out of primitive building blocks that don't have life or intelligence? Those are the ultimate emergent phenomena. There are lots of other emergent phenomena too. And we're not going to figure out where intelligence and life came from instantly. That's going to take my guess a centuries to resolve that. But meanwhile, there's lots of other problems we can solve. And so complex systems is really about understanding that. And you mentioned this intersection in your own life between the physics side of things, mathematics side of things, chaos, complexity, and economics. Yeah. I mean, economics is actually, in a sense, where complex systems thinking was born. Adam Smith said, "How does economy really work? Economy works by specialization." So we all, you know, the butcher, the brewer, the baker, I think. We can all be better off if they do what they do well, and we profit from their expertise to paraphrase him. And that's the basic principle that the economy runs on. But to make that work, we have to put in place institutions that foster coordinating our behavior, like the stable monetary system, like ways of determining what kind of credit is okay and what kind of credit is destabilizing. We have to have the rule of law. We have to deal with monopolies. Their rules about monopolies go back to Mesopotamia. They even classified the kind of monopolies. So these problems have been around a long time, and we're slowly figuring out how to run our economy. So it works for everybody. But as the world progresses, we get more and more specialized. The things we can do as a community become more and more powerful, and it becomes even more challenging to understand ourselves and guide our activity to make these institutions work for us instead of against us. Now one of the things that's interesting in the history here on the Santa Fe Institute is that it's seen the multidisciplinary aspect of it. The system is not just physics. It's ecology, computation, economics. These various disciplines, that seems to be to your point about specialization. It seems to be an effort to try to look across these disciplines to try to understand a system, which is an easy from the standpoint of any one piece of it, right? >> Yeah, I mean, I'm a firm believer in interdisciplinary work. My students are drawn from math, geography, and environment, computer science, economics, and I'm about to get one from engineering. So I've seen over and over again, if you put together a team of people that have complementary expertise, but a common interest in making something work that can develop a vocabulary and language that allows them to communicate with each other, that takes advantage of their specialized knowledge and interest and skills, then you can really make a lot more happen than you can having isolated individuals attack something. And that's a lesson that mainstream economics has not learned very well, but is become widespread. Even in physics, you have papers with 4,000 authors. You discover an elementary particle. You have everybody from the engineer who built the detector to the leader that got the money to do the experiment. You have teams with diverse skills. Now I think where we've been lacking most as a society is in doing things that really require serious cross cutting interdisciplinary activity that our system just doesn't make happen very naturally right now because academia is so siloed and the chasms between those silos are vast and so we need people who can attack the problems that sit in those chasms and that's really what I've tried to do for my whole career. You know I've worked in math physics, theoretical biology, computer science and now economics. So you know I think we were talking about this a little bit earlier which is what makes this moment in time particularly interesting from being able to wrestle with these complex systems problems whether it's the economy as you said or you and I have a shared passion around the challenges of the economic transition within climate and sustainability. What is it about right now? Well right now we've been enabled. We have the computer power we need. We can model the world at one-to-one scale at the level of individual households or even people and we can do it in a way where we're not violating everybody's confidentiality. We can model you know millions of different businesses and we have increasingly greater amounts of data to ground such models so they make sense empirically and so we have the tools at our disposal to do something that you know Herbert Simon in the 1960s could really only dream about and so we could do economics in a completely different way than mainstream economists learn to do it because that whole way of doing things develop you know in the 50s and 60s when the computational power and data were simply not there and I would even add knowledge about individual human beings social science that's progressed a lot since the 1960s we understand much better the universal patterns than the people behave and the variety of behavior which we can faithfully represent with computers in a way that it's very hard to do with pure mathematics and this I mean you're almost arguing that our ability to get the granularity the deep insight on the granularity of the economic physical human social system is what has the potential to give us insights to the system level that we haven't been able to get. Getting the granularity right is just one of the things you have to do to necessary but not sufficient condition the harder part is really empirically grounding a model so that you understand that it actually works. It's easy to make up models that do stuff yeah it's much harder to make up models that do stuff that you know you can trust and that's where you really need the data so that you can take part of the data to build a model and hold back part of the data to test the model and where you can microfound the model by taking each one of the pieces and grounding them against data at the level the pieces like your model for household you can go look at census data and see how households behave look at commercial data for other behave look at lots of kinds of data to model the way the household to behave have another set of data for how to the firms behave and then you put the model together and you see what emerges and then you test the model at the level of aggregate quantities like unemployment and GDP and so on where in mainstream economics you try and model things up at that level from the beginning it'd be like if to understand whether we try to build a model for a global mean temperature and global CO2 levels we wouldn't get anywhere we got you know models of the weather and climate are limited by the granularity with which you can build a model though the granularity isn't sufficient but it's a necessary step and then you have to get the physics right for a weather model you have to get the economics right meaning both the accounting and the sociological interactions with economic activities that allows you to predict what the economy is going to do depending on what happens through time now we had a very unfortunate right totally shared experience with COVID yeah and it threw everything up and it created an environment where I don't think anybody could actually predict what was going to happen we weren't sure what the we knew there were going to be bad impacts on economies but I don't know if anybody could really do it however your team actually did what I think is a fascinating piece of work could you talk more about that because I think to your point there's some balance here between getting granular getting things at work and then having an impact on the system if you will yeah no COVID from our point of view was a great testing case for what we could do because when COVID started to happen I mean it was clear even before any lockdowns happened that this was something that was going to sweep over the world and was going to shut the economy down as we tried to cope with it and so I put together a team of my best people and we did a crash program where for a couple of months we just were literally around the clock because we had somebody in Australia so that we could hand work to her and she could hand back to us the end of the next day you know everybody was locked down but we'd have a video call every day and we realized that this is a case where the kind of models we build should be able to shine because the shock was going to be huge was going to be really sudden and so traditional equilibrium models were just not too good it was also clear that it was very heterogeneous it was going to hit different sectors in the economy very differently and I had a couple of people on my team Maria Del Rio Chinan and Penny Mealy who were real experts on occupational labor and its relationship to industry so that we could predict ahead of time which industries were going to be shut down because the workers would not be able to go to work so we put together a prediction for remote labor index that turned out to be really good so we actually predicted the shocks ahead of time because we could predict which industries would be afforded because they didn't have enough labor and then we built a model that tracked the way the shocks were going to reverberate around the economy and how they were going to interact because we knew that companies were going to be shut down if they didn't have the labor they didn't have the raw materials they didn't have the demand so we just put those things into the model we were lucky because iHS market pro bono did a survey for us in a week where they asked all their industry analysts if you have to go without which inputs can you go without for two months without having a large effect on industrial output so we knew what the essential inputs were for each industry so and that turned out to be critical because we had an accurate production function that was accurate industry by entity so each of these industries in some sense was an agent in your model yeah we modeled it because for convenience purposes here the sweet spot was the model at a sectoral level so we had 52 sectors with a representative firm in each sector and then we could model what it's output was going to be day by day model ran on a daily time scale we got data on the inventories goes so we could watch how the inventories were getting chewed up and then we could see when you hit a critical place where the industry either ran out of inventories because stuff that needed wasn't getting produced or it didn't have the labor to produce it or nobody wanted their product so they had to stop producing and we could look at how all the 52 industries were interacting through time so we made a model that successfully predicted the second quarter of 2020 GP hit for the UK we predicted 21.5% when the dust settled the right answer was 22.5% 1% that bad it was pretty good and we predicted more or less right industry by industry in order to post mortem we could see you know where do we get things right and for example having the right production function made a huge difference if we've been wrong in either direction our predictions would have been a lot worse so we think it's a nice proof of principle that these complex economics model yeah let me interrupt you just a second because you were talking about kind of the aggregate outcome yeah yeah of a GEP contraction at the 21.5% but given that you modeled these individual industries and the way they would evolve how many of those if you will individual industries were you able to get pretty close to in terms of the way they were involved giving you described as type connection between labor various inputs for production and the connection between them well we got about an 80% correlation between the industries that we predicted would you know go up or down most of them either going down or holding flat but the discrepancy was huge some industries went up stayed flat weren't affected much others you know throughout but went down by more than 50% so it was a very heterogeneous system or set of outcomes and if you just make the scatter plot of predicted versus actual we got about 80% correlation we got a few things wrong you know we predicted health care would go that would go up that actually went down a bit because of routine procedures being delayed but by and large we got stuff right and by the way we got the time course right too the profile of the recovery we got pretty much right and I shall also say our predictions were conditional on the policies the government would enact and so the paper we sent to the UK government said if you guys do this this is what we think the economy will do if you do this instead this is what the economy will do and we pointed out that There was a sweet spot where if you kept all the upstream industries open, but you closed the downstream customer-facing industries, then you only had incrementally more deaths than you did from a complete lockdown, but the economy remained much more healthy than it would be from a complete lockdown. I mean, it strikes me that sort of insight at the kind of a granular level is a whole different way to inform policy-making. I don't even know how you get at the levers at the macro level that would suggest, "Hey, here's where that sweet spot really is." Maybe retrospectively, everybody could look and say, "Well, that makes a lot of sense." I mean, if we think about value chains and we think about human risk, but you need some sort of confidence as a policy-maker that there's a basis to this, you know, beyond the politics, there's a basis to take particular action. You know, what you're describing sounds quite kind of revolutionary in terms of the insight that one could have at the economic system level. Yeah. I mean, so first of all, I think it's important to go down to a more granular level for several reasons. One is, you can make predictions about a lot of the things that you couldn't predict otherwise, because, yeah, you want to have an aggregate prediction, but if you're in a given sector, you care about what your sector is going to do. Right. And your sector may behave completely differently than some other sector. So we can make predictions that you can't make with an aggregate model. But secondly, you have a lot more data to test on. So we're in the process of writing a more theoretical paper about this right now, but if you think about what happens in a typical aggregate macro model, you take, you know, a few indicators, GDP, unemployment, inflation, interest rates. So you have a few time series that you don't measure that accurately that are highly correlated from quarter to quarter. And you only have 10 or 12 business cycles since World War II that are remotely relevant to what's happening. Now, you take all that data and you compress it with an algorithm. My guess is you might have 500 bytes of information, maybe a thousand bytes. And yet you're trying to model the macro economy, which is a complicated beast. It's telling you that you have more than two or three free parameters. You're not going to be able to estimate them very well from that data. Now, if you start doing micro data, you have a lot more data to fit your model. And so instead of ranting the world through this information bottleneck and then trying to predict what's going on, you take the richness of that micro scale and you try and model that. The key is to do it in a parsimonious way so you don't overfit what's happening. So there's two big advantages to going down to the micro scale. And then you have a lot more ways to test whether what you're doing is working right. And if you see that, oh, this sector, we're not getting that right. You know, that's your bottleneck. You need a zero and figure out what's wrong with that part of the modeling. Fix it. So, you know, I'm a firm believer that if my model can't make predictions that are correct through time, why should I believe any of it's, if I do a counterfactual, what if we implemented this policy, why should I believe it's true? If it's not already predicting things under, you know, the non-counterfactual, the normal world day and day out, I think, you know, the great example again to go back to weather and climate, weather models are really well tested because we've been testing them every day for decades day and a day out around the world in a microwave. We know exactly how well they work. And we can have some confidence if we do a counterfactual that we're getting things right. We don't know, but it's a lot better to start with a baseline that's solid than to start with a baseline that already bullshits. Yeah. And unfortunately, I think that's the way it's been in economic modeling. Now, this is a revolution in economics because in mainstream economics, models are derived by assigning everybody utility function and deriving the set of decisions that maximizes everybody utility with everybody taking everybody else into account. Whereas we're saying, let's do things completely differently. Let's take behavioral models that are built by observing how people behave. Let's put them into a simulation of economic activity that's inherently dynamic where everyday agents make some decisions as best they can using heuristics and limited reasoning. They're bound to be rational, like real people. We try and mimic the key mechanisms of the economy as closely as we can with what I call verisimilitude. So it ought to smell right, you ought to feel like, yeah, that's about what people do. So we don't have to use as-if arguments. And then we model the dynamics of the economy. People make decisions. They interact with each other. That influences the decisions people make. We go around the dynamical loop again and again. We might converge to an equilibrium or we might not. If you sort of play this forward and you say, there are more predictable times than others. A lot of time is based on our intuition and our histories and our ability to reason through given the situations we've learned over time. And there are periods that are huge disruptions. And one could argue that in disruptive periods, there's probably more nonlinear effects at work. Those things are seldom intuitive to many people. And so you risk in terms of being able to reason effectively through times of disruption or times of rapid change or times where many, many things are changing at the same time. You should be skeptical of your ability to get that right. But it's as I was saying is if we can actually apply the technology in the science that we've accumulated over time and built over time in complexity economics, complexity science, and apply our computational capacity at more and more highly resolved levels that we actually can deal with these more sudden changes. Is that right? Yeah. And one of the big advantages of agent-based modeling, which is really just the term for simulation and the social sciences, called agent-based modeling because you have to deal with the fact that there are people making decisions and they have agency. So it's not like an atom. An atom has to do what laws of physics tell it to do. What was that? What was that quote that your founder had once Murray-Gelman was, what was that? The particles had their own minds. Yeah. Yeah. Yeah. The atoms had their own minds. It would be much, much harder. So we have to deal with that. Yeah. And that's always the most difficult part. But by using a simulation, we can deal with all those nonlinearities because there's nonlinearities that make mathematics hard when you have lots of nonlinearities solving for these optimum becomes, you get a rugged landscape, solving for the optimum becomes particularly difficult. But in a simulation, this stuff is relatively easier. And you might say, "Oh, why don't you just use machine learning?" Yeah. What's why not? I mean, we're all talking about it. But two reasons, not human machine learning. First is the economy is slow. So not something where you have data sets with millions of data points. You have data sets with often only hundreds of data points. So if you're just going to fit a nonlinear model to the raw data, you just don't have enough data, and you need more data to model down. But the second one is that any statistical approach, whether it be classical quantum metrics for statistical modeling or machine learning, they're both just different flavors of the same thing. You rely on the world, the future being like following the same patterns that it followed in the past. And if those patterns change, all bets are off because the function you're fitting in the underlying machine learning model can become wrong. Whereas an agent-based model, okay, you can still have problems with those kind of non-stationaries. But because you're trying to get at the causal mechanisms that are underlying what's going on, you can first of all build a model where you have a lot more product knowledge built in. So you can get that model to work with a lot less data. And secondly, you can do what if experiments were because you've understood the causal mechanisms, you have a much better chance of understanding what will happen when you take a course that years on, of course you've been on. Now, of course you have to make sure, and COVID would be another example, where I don't think we understood the way people were going to interact, react socially, the whole anti-vax thing. Those would have been tough for us to do. So it's not like this is a panacea we can do everything. And one of the things I learned about science is the art is to figure out what you can predict and what you can't predict. So the reason our COVID model work, one of the reasons it pretty mechanical effect, you know, the shocks hit the economy because of who wasn't going to be able to go to work. And those shocks propagated around the economy in a very mechanical way. Understanding the anti-vaxers would require a much more sophisticated understanding of social mechanisms. I'd love to understand that someday. But part of what you want to do is start with the things that are much more mechanical. Like I think the energy transition is pretty darn mechanical. Maybe to put an ass risk by that. Well, only after you understand technological change, you can't predict this is going to be the new innovation. What you can predict is this new innovation is probably going to perform about here that is probably going to be an improvement about this side. Let's go there for a little bit. I think in this period of time we have this massive change. ahead of us in terms of the energy transition and sustainability transition, food system, transition, material transitions, right? And this notion you just mentioned of sort of deep insight into the, if you will, physics and the behavior of technologies within systems as they mature as they scale, say more about your research in this area because I think you've done some fascinating both lookbacks and forward forecasts that pretty much resulted in an opinion that you develop about how expensive this energy transition was going to be. Yeah. Well Gordon Moore was you know a great visionary when he said this is the rate at which integrated circuits are going to shrink and of course integrated circuit shrinking drives the cost, it drives the energy usage, it drives the speed. So you shrink it, you get all three for free once you've done that and you know he made a prediction in 1965 that he amended a little bit later and it's held true until now remarkably well. An even earlier pioneer was Theodore Wright who predicted that there's a relationship between decreasing cost and increasing experiences measured by cumulative production that Boston Consulting Group made a lot of use of in its early days. So these are things that went before that made it clear that look these rules about predicting the future and people didn't realize for a long time that Moore's law doesn't just apply to semiconductors there are versions of that for lots of other things. Other technologies, some technologies basically over a century hardly improve at all. Other technologies like transistors improved dramatically by a factor of a billion as in contrast by factors of less than five. And so it's remarkably simple to just measure that and make these forecasts to do it right though you need to understand how good the forecasts are. That's the contribution we made is coming up with a good algorithm for predicting not just what we thought was likely to happen but what are the error bars on that? What's the probabilistic forecast where we were able to collect data you know make six thousand different forecasts and derive how accurately we thought those forecasts were and show that our method actually predicted how accurate the forecasts. Across those across family of technology different technologies going through time. And so that's been a key driver of the rest of our work because just as Moore's law was an extremely useful thing because it forecast if you know where the industry is going to be in five years then you can plan to meet targets that are going to be there in five years. One of my favorite stories is from my friend Alvi Rae Smith who was you know one of the founders of Pixar a fellow New Mexican and so Alvi told me that basically five years before they made toy story they pretty much had all the software ready but they just didn't have the computer power so they could say to John let's just five years get it five years we're going to be all ready to make real movies with this so they perfected it but they really needed to just wait for Moore's law to catch up to them and so that happened in many different spheres. Similarly for the tech transition if we can provide road maps to where the technologies are going to be at different points in time with the likelihoods they're going to be there then we can plan the energy transition in a sensible and effective manner and businesses will be able to understand where things are going to be and arrange their goals to be in the right place. Is this sort of behind the reason why consistently forecasts about adoptions and cost and technology performance of solar and when it been wrong? We're in a year. There's some reasons in the idea of those those forecasts are then wrong. The conventional integrated assessment models and their cousins like the I.A. model are built around the idea that they ask the question what is the optimal route in terms of maximizing discounted consumption for if we adopt a given target for how much emissions we're going to release what is the optimal plan to get there. So right away you've got something that's not really empirically verifiable because if you say well look have you predicted what's been happening they'll say well we don't know what plan we're on what scenario we're following and we're just predicting the optimum do we know that the world is following the optimum. As we say well let's look at historical trends let's take empirical things we understand and half can we predict what's likely to happen and then policy-wise can we nudge it in a good direction to make it a little better or maybe a lot better but let's start by understanding the underlying pattern so that we have something we can test and that's key you've always got to have something you can test so you know whether it's any good and so we really just asked a different question and we attacked it with a very different set of methods and we were able to test those methods incrementally and show that they were the right ones and that gives us some confidence that going forward we have a lot more reason to believe what we've said and what they've said and by the way you know they missed by an order of magnitude at least consistently underestimating how quickly renewable technologies were going to be deployed and how cheap they were going to get. We could sort of dive a little bit deeper on this because if I recall correctly I think you were interviewed a couple times on the BBC and I think you wrote a little bit about this as well and it was a view that because of this foresight that these technology maturing curves offered and in terms of their consequences on cost and everything you could have a view about well you think the energy transition is going to cost you this much but if we play this forward over a reasonable period of time there's a lot of improvement ahead which means this is actually a pretty smart positive economic move to make could you talk more about that? Did I get that right? No you got it right. You know it's interesting because the integrated assessment modelers knew at least the more sophisticated ones and it took them a while to get that in there but they were aware that technologies tend to improve though initially I have to say they made poor assumptions or remember talking to some people at you know Pox Demis II for climate change when the leading models remind and their assumption that they were predicting all the technologies are going to have proven about the same way they said wait a minute that's not what history says fossil fuel technologies aren't going to improve solar energy PV solar energy is going to improve and so okay they fixed that but then they they say well but if we run our model what does it want to do? It wants to do nothing but build solar farms I can't be the right enter so okay we'll put a floor costs in so once the prices hit here they'll hit the floor cost and they won't improve anymore we'll put in the deployment constraints so they can't deploy faster than this rate and so that handicaps them and lets the you know CCS and other catatologies keep in the running and you know we have a plot in our paper showing how those floor costs just got punched through again and again and again they would lay out a floor cost and then the technology would knock right through it and they've done that many many times now and mechanized sort of skew our whole worldview of this energy transition can happen yeah so the fact that they made these auxiliary assumptions in the end meant that their models were wrong and this notion that hey wait government you don't understand you put the right time frame on this yeah and actually it more than pays for itself yeah so yeah let me come back and answer your question properly so whereas these guys are predicting all these roadblock we said well it's just looking what the data is telling us and the data tells you you know we looked at 50 technologies they have exponential trends for decreasing prices once those trends are set in place things fluctuate around them you know solar has been a decade right more or less flat but there's a long range trend that things tend to revert back to whereas others are saying well okay solar is now as cheap as these other technologies our prediction is it's going to keep getting cheaper and so even when you put a strong discount rate on the future the numbers are such that now we actually come out ahead economically by transitioning to renewables that is even if there were no greenhouse gases it would be a great benefit to society to make this transition just to make energy I'm up cheap purely just to make energy cheap yeah not to mention we get better energy security we get rid of pollution we get all kinds of auxiliary co-benefit and of course the big one is dramatically reducing climate change yeah and this and this ability to if you will continuously advance if you will the science of prediction through both deep inspection of the past as well as smart modeling of the future granularity these strike me in combination is incredibly important right now given what we're trying to do if I'm an investor pretty money to work and I want to recognize these are these rare opportunities to take advantage of a massive economic transition or I'm a company in an industry and I'm recognizing my industry the competitive order is likely to get re-stacked here and the ability of my factories in one part of the world versus the other likely to get reshaped by the capacity to make these transitions differently in different countries. Or if I'm a policy maker gracious, what really should I be incentivizing and how much should I do it? And am I really getting the sequence right? It's something what you're talking about ought to be something we are incredibly excited about, grinding a hole of it, in some ways it's a technology that we should be applying to the energy transition. Yeah, well, you're pretty the car. I know it's just exciting. You know, it's exciting to think about because I don't know if we think about that kind of computational and analytical technology is actually being part of an integral too, our ability to guide this transition. The thing that's always puzzling me, I love you're thinking on this, the market will take care of it, the free market will sort this all out. And yet we sort of have ticking clock here. There are tipping points, right? Which would strike me as we need to perhaps guide this thing a little more tightly than the market will sort of that. How do you kind of think about that, you know, sort of well, you don't understand this magic of the market, the market will allocate capital correctly. I think we have a lot of examples of where maybe that doesn't. No, I mean on one hand, I think we're really lucky that just when we really need it, we have the new technologies that we need to deal with fossil fuel emissions and climate change. But it's not like the market just magically takes care of it all. We've benefited from, you know, price supports from lots of governments, research efforts funded by governments, the technology to really get long range, deep research, you really need governments to get behind it, need policymakers to back it up. We're lucky that we've reached a point where we are at that tipping point where renewables can more or less fly on their own, though we need to make sure that we get things like hydrogen storage cheap so that we can deal with the whole problem and not just part of the problem. So we need to keep pushing them, but we have the advantage that the economy is now really starting to work for us. We have the wind at our back dealing with climate changes no longer the seeming burden that used to be. The thinking used to be, oh, climate change is a really big burden. We've got to figure out who's going to bear that huge cost or it's a hot potato that everybody passed around. And that was the whole philosophy behind these earlier integrated assessment model. Whereas now we go, well, actually since renewables are taking off, since they're going to make energy cheaper than it's ever been, it's actually a huge economic opportunity. And now it's a race for businesses and for nation states to position themselves to take advantage of this change that's going to make things better for everybody in many dimensions. Could I talk to you about a huge profits? There we go. There we go. Yeah. Could I ask you about this concept of SIPS or strategic intervention points or however you like to think about it? But if I have true visibility or at least testable hypotheses about the system, how do I use this notion of nudges and SIPS and kicks in order to accelerate what you're talking about? Yeah. Is that possible? Yeah. Well, so there's a great picture, I don't know if they'd ever made it in our final paper, but it was a great example where now I had a satellite that was in orbit around the earth and they decided I'm forgetting now they wanted to send it off to Jupiter somewhere. So they waited till it's just in the right point of its orbit and they gave it a little kick a little night time and it made this amazingly complicated orbit that went and landed it right where they wanted to go. In other words, they got a huge change by applying a very small stimulus at just the right point. And this is something you often see going back to my chaos back then. Yeah, exactly. You often see in chaotic systems because of the non-linearities, if you want to control chaos, you can take advantage of the places where things are actually very sensitive to what's going on to give them little nudges to get them to go to a very different place. And so the idea of a sensitive intervention point is to understand the system well enough that you know how to nudge it to get it to do what you want it to do instead of all the end to some tipping point that takes you to a state you really don't want to be in. And right now we're at a place like that with a climate where if we just passively sit there, we're going to end up with who knows what two and a half, three degrees, worse. But if we do the right things and it won't actually cost that much, in fact, we can make a profit while we do it. But we still need some policy makers making some key nudges. If we make those right nudges, we can steer the system to be in a much better place much faster. And so we should be using every possible tool to reason this through. Yeah, model this through. Yeah, we should dilute ourselves. We have to keep humility. We should dilute ourselves in the thinking our predictions are better than they are. Scientists throughout history have always aired on the side of being overconfident about how much predictability they have. But on the other hand, we should make use of the predictability we know we have in order to try and steer the system and steer society. The analogy, I like to think that science is giving us the opportunity to make a conscious civilization. So we can do things like look out at the sky and scan for asteroids and be ready. So if we see that asteroid, we can come up and perturb its orbit enough that it doesn't kill us. Climate change is a great example. By understanding the Earth's weather, then we reach a point we can understand the climate where we can see something that's really more of a problem 50 years out than it is now and anticipate it. Now, then we have to deal with the social systems to cope with what's happening. We have to deal with the economics of that. That is, we have to be able to model the way we ourselves affect the world that we live in. Consciousness is all about having a model of yourself. So we have to become a conscious civilization that understands itself and our collective behavior well enough to anticipate what's going to happen if we take different choices and so that we can guide ourselves into better choices. And I believe you have colleagues who've really advanced the modeling of social dynamics, social systems. These are some of these same techniques, but with the deep lens on collective behavior, emergent behavior is the sum of the way societies work. So we can bring these worlds together. They have sort of the physical long supply chains around the globe, the technology changes and the social systems now with technology that helps us at least get closer into our ability to at least reason more clearly about the future. Yeah, you know, our understanding of these things is really advancing right now because we can now collect data about behavior on a scale that social scientists could even imagine 20 years ago. We have big data sets that involve records of how people behave and we have the computational power to analyze them and we have emerging theories for how these things behave. I mean, it's striking that something as basic as the ising model in physics where you have little magnetic spins that can flip up and down can be turned into a model for opinion dynamics that helps you understand why polarization occurs and gives you some hints about how to keep it from occurring. And so these are all nascent things. We still have a long way to go, but I think there's great potential for them to really provide powerful tools going forward. Could I go into one of the other areas? You're just not thinking about this stuff. I mean, you're literally committed to actually, as you did with the prediction company, creating capability here that can help us. Can you say more about that or would you prefer not it? Well, no, I mean, since I beat Relat and were interested in, where are we all computer under each arm? Well, computer under one arm and 12 AA batteries with three kilobyte program. Machine coded, three kilobyte program that I hand coded myself. But I love predicting. It's just fun. And so with prediction company was a similar challenge. Okay, those were fun, but this is something where the stakes are much bigger. I was influenced by reading the foundation trilogy. I've always wanted to be a mini version of Harry Seldon ever since I read that when I was 16. So yeah, I think it's a great adventure, but it's something we really need. And something where, I'm older now, but I've had the luxury of having remarkable students at Oxford who can do all these things and really care everything forward with great energy and confidence and been able to assemble a team of people that I know are capable of doing this that work well together. And we want to do it to, you know, we'd like to make a little money, but we really want to make the world better because we think this is needed. Yeah. Well, I think we're all going to be watching with enthusiasm about your ability to move this forward and in a much needed technology to help us navigate all of this and certainly to get more insight into the real economy and the choices it can be made during this transition. I guess if people wanted to learn more. I'm not mistaken. You have a book coming out. You can say a little bit about the book or I know it's not out yet, but maybe just what's it going to cover and the date that Penguin has given me is April 25th, 2024. Here we go. So it's coming out soon. I mean, I've worked on this book for 10 years, so I'm very glad to see it coming out. It's, the book is really about what we've just been talking about. This interview, pretty much this interview is almost a summary of what's in the book. The book maybe talks a bit more about the financial system than we have here. In addition to talking about technology and macroeconomics, I also talk a lot about why is the financial system so unstable? Short answer is leverage. You know, leverage we need it, but it's a very dangerous thing because it can be a big destabilizer and you know, why does the market change all the time in the way? So I think we have some limmers about market turbulence and so on. But yeah, if you're very pleased if anybody appreciates my book. Terrific. So that's April, April 25th. The title of the book is, because it's the hardest thing for an author. I keep that title in mind. Yeah. Making sense out of chaos of better economics for a better world. Super. So we will make sure all of you have the opportunity to, when that book comes out, to be reminded of it and certainly to know where to go to get it. And again, I think for all of us who are passionate about the energy transition economy and sustainability and how this is fundamentally can fundamentally reshape the trajectory of the world. We're happy to hear there are new tools and computational resources that we can apply in new ways to give us a better chance than just banging along. This podcast was part of our series on building competitive advantage in the sustainable world. For more information about this and other research topics, follow the PCG Henderson Institute's research online at bcghendersoninstitute.com and follow our podcast series on Spotify and Apple Podcasts.