COMPLEXITY: Physics of Life

R. Maria del-Rio Chanona on Modeling Labor Markets & Tech Unemployment

Episode Notes

Since the first Industrial Revolution, most people have responded in one of two ways to the threat of technological unemployment: either a general blanket fear that the machines are coming for us all, or an equally uncritical dismissal of the issue. But history shows otherwise: the labor market changes over time in adaptation to the complex and nonlinear ways automation eats economies. Some jobs are easier to lose but teach skills that translate to other more secure jobs; other kinds of work elude mechanization but are comparably easier for humans, and thus don’t provide the kind of job security one might suppose. By analyzing labor networks — studying the landscapes of how skillsets intersect with labor markets and these systems mutate under pressure from a changing technological milieu — researchers can make deeper and more practical quantitative models for how our world will shift along with evolutions in robotics and AI. Dispelling Chicken Little fears and challenging the sanguine techno-optimists, these models start to tell a story of a future not unlike the past: one in which Big Changes will disrupt the world we know, arrive unevenly, reshape terrains of privilege and hardship, and reward those who can dedicate themselves to lifelong learning.

This week’s guest is R. Maria del Rio-Chanona, a Mathematics PhD student supervised by SFI External Professor Doyne Farmer at the University of Oxford. Before starting her PhD, Maria did her BSc in Physics at Universidad Nacional Autónoma de México and was a research intern at the International Monetary Fund, where she studied global financial contagion in multilayer networks. We met at the 2019 New Complexity Economics Symposium to discuss the use of agent-based models in economics, how the labor market changes in response to technological disruption, and the future of work.

If you enjoy this podcast, please help us reach a wider audience by leaving a review at Apple Podcasts, or by sharing the show on social media. Thank you for listening!

Visit our website for more information or to support our science and communication efforts.

Join our Facebook discussion group to meet like minds and talk about each episode.

Maria’s Website & Links to Papers.

Maria’s Google Scholar Page.

Andrew McAfee & Erik Brynjolfsson on Technological Unemployment.

Carl Benedikt Frey & Michael A. Osborne on Technological Unemployment.

Podcast Theme Music by Mitch Mignano.

Follow us on social media:
TwitterYouTubeFacebookInstagramLinkedIn

Episode Transcription

Michael: Maria, it's a pleasure to be joining you amidst the complexity here.

Maria: Thank you very much for having me, Michael. Thank you very much to SFI for inviting me here.

Michael: Yeah. So, I'd like to... Before we dive into the details of your work, this is... I think you're the youngest person we've had on the show here, and this is a really great opportunity to ask you some questions, as a young scientist, about your early career work, and how you got into the field. I'd love to know what drew you into your study with Doyne Farmer, and into the discipline more generally.

Maria: Yeah. So I did my bachelor's in physics, but I have to say, when I was in high school and I was deciding what to study, I considered physics or psychology. I went for physics, because people told me it would be easier to switch afterwards. But I think the psychologist and the curiosity for human behavior never stopped at me. I loved physics. But towards the end, and even at the beginning, I sometimes felt the world needed a bit more economic solutions, that it needed physics solutions. So, I wanted to do a PhD, and I literally Googled “physicist economists.” I like the UK, so I Googled there. I found Doyne Farmer, and I started reading about him, and thought, “This guy's amazing.” I read about... and I actually saw the documentary of him playing roulette.

Michael: With the shoe computer?

Maria: Yes. It was like, that's amazing. I'm the type of person that could never tie an electric circuit. But seeing that, I was like, “Oh!” I was inspired in some way. The way he says we can predict the economy or we can understand it, and we have to get our models, make them empirical, go to the data and show it works… For me, it showed how the future of economics should be, and I wanted to be part of that. Yes, I guess I just wanted to contribute to the world, and I felt the only way was if our models, which I mean scientific models, agent-based modeling in particular, were backed empirically. So I wrote Doyne an email. He replied, he said, “I'm going to Mexico.” I met with him, had a great conversation, end of story. I applied for the PhD.

Michael: Right on. So you spoke at the annual symposium that we held this weekend on New Complexity Economics and made a very strong case, although to a very sympathetic audience, for the role of agent-based modeling in economics. I'd like to hear you unpack that a little bit for people that are not familiar, because the audience to this show. I know there're a lot of experts, but there's also a lot of people that are new to these ideas. Talk a little bit about the value of a data driven approach and agent based modeling in economics, and why it is that you've cleaved to these methodologies.

Maria: Because I think it's intuitive. Because I think agent based models relates to how we see every life. Every person is independent and does decisions. The thing is, you'd like to see that in an experiment sort of way, like us in physics. But you can't do that. For example, Rob Axtell said, well once you have a policy simulator, you can run hundreds of scenarios. Standard economics, they start with some actions, rational agents for example. Those actions lead to an equilibrium where supply equals demand. The way they think is... Well, many economists think, and even some physicists doing economics, how do we deviate from this equilibrium? But it's focusing on the end rather than on the trajectory. That's the point I believe mostly in, that we should focus on how the economy's evolving and not where it should be, because the reasoning behind that is, well it should be an equilibrium because, if it wasn't, there will be this force, this invisible hand that will put it back in there.

But, as many complex systems scientists say, if we're at equilibrium, we’re dad, and I think that's the one thing we know about society. We're not all dead. We're quite alive. That agent-based model is focusing on the dynamics, on the evolution, and it's the thing you see too in everyday life. Now, I think we're at the point where economics can start developing in a different way. Why? I believe the recent economics started with this action axioms and these theorems was because it was so hard to get data to experiment, to see that your ideas were true. Then, it might be a bit easier and even useful, when you can do it, to just lock yourself out and just think rationally, think logically. “If A and B happens, then do you get C?” And actually, math is beautiful. I love math. But then the thing with math is that, once you prove something, you know it's right, and that's beautiful, and that can be, in a way, addictive. You get addicted with playing the game of math, but you forget if your actions, if your assumptions, match the real world.

But we couldn't do much because it wasn't easy to observe the economy. We could see how we shop, et cetera, but you wouldn't record all those transactions. But now I think now the world is changing, and now we can say, I believe the world behaves in this way. I do my agents, I put them some behavior, and I can calibrate their behavior with real data. For example, we can see now, in general, if you have one occupation, what's the likelihood you get unemployed? What's your average wage? Those things let us know how agents might be able to behave. Your wage might define your consumption patterns. So, now we can actually observe the world and we can test if our ideas, if our models are correct. I think that's the new way to view the economy, where we have to be very strong in our models, but we have to bring them through the data. If they fail, we know we're wrong, and then we try again until we get it. Because I think we owe it to the people. I think the economy is something important.

One of the quotes I learned the IMF when I was at Christine Lagarde’s talk, and one of the things she said was, “people before numbers.” That stuck to me. That's the one thing I want to do when I do economics. I always doubt myself. I doubt my models, and I think, as a scientist, we should, and the only way to be a bit more sure about it is to bring it to the data.

Michael: So I'd like to talk about two of the papers that you coauthored, and I think that they're very intimately related, but the one I'd like to get to first is about looking at the labor realm through a network map, a network model. This is a paper that you wrote with Penny Mealy and Doyne Farmer: “What you do at work matters: new lenses on labor.” So, I think most people have general intuition that the skills that you learn in one job sort of change the possibility space for you if you're going to move to another job. But this paper really looks at this in a very granular way, and in a way that, like you just said, would not have been possible in an earlier period of scarcer data. So, could you unpack this a little bit? How did you actually set up this model, and what did you find?

Maria: Right. Yes. Great phrasing of it. I think the way it started is with the conversation with Penny. Penny was the driver of this paper. Penny's a great scientist, and she knows a bit about economic... Well, she's probably one of the people that knows most about economic complexity. She had this idea that you could see how similar two occupations were depending on what you do at work. So for example, a paramedic and a nurse might be similar because they do stitches, they care to wounds, in the same way that a taxi driver and a bus driver might be similar. Based on the idea of economic complexity, you can define a network of occupations where they are linked. These links are weighted by the similarity between them. We defined similarity as the amount of work activities they share, but some normalizing factor.

So, when I was talking to Penny and she mentioned this to me, it seemed like a great idea. I knew a bit of networks, so we started working. Then it comes this thing that always comes out when you work with Doyne. You have this idea, and he thinks, “Yes, I think that's true, but let's test it.” So how do you test if it's true that two occupations are similar? What we thought it would be is, “Well, let's look at data on job transitions. If it's true that two occupations are similar, then maybe you can see if they're similar enough for one person to switch from one job to another.” That's what this paper focuses on, and we managed to show that there is some relation. There is actually more relation to other networks. So there's, for example, Own It has this career changer network where they suggest to you other occupations, and we actually managed to beat this in a way.

So, we showed that this network is in some wa, predictive of how people can transition between occupations. When we saw that, we saw... I think that's when you know that, okay, the ideas of how you see the world actually have some sense in reality.

Michael: So, there's two very interesting maps that we'll link to in the notes for this episode. One is the “job space network,” and then the other is the “work activity space network.” You already mentioned that a lot of this has to do with the energy, or time, or attention required in re-skilling the way that these cluster out. But, looking at this, it's beautiful how this sorts into... It seems so similar to the questions of evolutionary possibility based on the anatomy of the organism, and then whether a new niche that's opened up can actually be populated by the organisms in that ecosystem or not. I don't know. So I'd love if... Sorry, I have a way of branching into tangents here, but as part an effort to create bridging analogies and generalizations, what do you see as the sort of general insights of the job space and work activity space networks? At a very deep layer, what are we actually looking at here, energetically or informationally? What is governing these geometries?

Maria: Oh, that's a lovely way to put it. I wish I had thought about it more that way. So I might... I'll speak from here hypothesizing. So, these are ideas I haven't tested or we haven't tested, but I think it's beautiful the way you put it, to look at it as an evolutionary perspective. The way you might see it is as if occupations were sort of different... Well, actually “species” is not the correct term, but there were organisms, let's say, with different phenotypes. So they're the same species, let's say, but they have different colors or different features. You're able to jump from one to the other between generations. Here, the analogy would be between re-training. Then the thing that would be amazing to study is how this changes, because the way you could see it is, occupations are in an ecology. The ecology is to remind the demand affirms what do they want. So for example, a lot of people right now are asking for things related to data and statistical analysis. So, in a way, the environment is changing and the organisms have to adapt. So you'll probably see organisms or occupations... Well, the organisms in the occupations, which you can say the population of one occupation moving towards, moving to adapt to this new landscape.

When it comes to work activities, I see that more of the characteristics of the people, what things you can do, the skills. Then, if they're close together, it means... I would say it means it's like when you have one gene, you're more likely to have another gene. There's some correlation between genes. In the same way, maybe if you're good at painting, you're good at sculpture, while if you're good at solving math equations, you might be good at coding, things like that. So it's, let's say they're genes that might be correlated in this sort of ecosystem that I'm not even sure I defined the correct way, but I would venture and put it that way.

Michael: So, one of the useful insights that emerges from this network analysis is the demographic distribution of labor. I'd love to hear you speak about that. The way that looking at this helps to explain certain persistent issues in the labor market with the availability of certain kinds of work to certain genders or education levels, or these kinds of things. So what do you think precipitates out of this? How is this generative of useful insights to policy makers or employers or educators? How can we use this to shed light on the ways that we are failing to provide opportunities for people, or what other insights spilled out of this for you in that regard?

Maria: Yes, when you see the pictures, you see this network, and then upon this network of occupations, we plotted things like education, and you see they're clustered, differently. You see things like wage. We also plotted the gender, and it's also split, but it's split in different ways. So, in a way, you can see a clear cluster of, let's say, high wage and low wage. If you put a gender on top of it, you sort of split it in four, which would be high wage female, high wage male, low wage female, low wage male. So it tells us that... Well, first thing, wage is driven by many things, but one of them is demand and supply. So in a way, when you see this plot and you see low wage in one place and high wage in the other, if you wanted to make it more homogeneous, what you'd say is the people, the occupations with high wage means they're difficult to reach. It means it's difficult for people to have those skills. So, the policy there... and this might be over-simplistic, but what first comes to mind is teach the skills in those occupations, because you're saying it's scarce. That's what drives wage. People cannot do that as easily.

Well, the little skills you'd say, well maybe I don't have to focus that much on it. When it comes to gender, for example, that's a bit more complicated, let's say, because is it good for people to split in genders that split occupations? I would say no, but economically I'm not sure if there's a reason behind it. I think that one is more interesting to understand what causes it, and it's one of the things that I'd like to build a model to explain this pattern. I think the big question is, is it genetics or is it cultural? My personal opinion is its cultural. I don't think women and men are born totally different and that determines if you want one occupation or the other. I think it's more cultural. But I think this is a way of thinking back and saying, well what makes her culture tell women to thrive in some occupations, in men in others. So I think those are the questions I might be interested in.

Michael: What surprised you about this research?  Looking at these correlations, did anything pop out to you as counterintuitive?

Maria: Yes. So I would say there wasn't anything too surprising from this work. Some of the things were intuitive. The fact that there's a clustering and there's high wages and low wages is intuitive to daily life. But there were some red alerts, and some of the red alerts include, for example, that occupations with high wage have low problem of being automated, and also do not tend to be very pollution intensive. Low wage tends to be automate-able and tends to be pollution intensive. So if you think, this ecosystem is evolving, and it's evolving towards adapting to climate change and will reduce the low polluting occupations, and will automate some occupations… This is a warning sign that says, well, if the jobs are going to remove our already low wage, that just gives a bit of a warning. I think that was still something that struck us, and when we saw out we were like, yeah, I think we need to go deeper.

Michael: It should come as a surprise to no one that we're talking about the low wage jobs that have the highest exposure to the risk of automation are the majority of this map. So it gets into this question that comes up on the show a lot, which is, in situations where this seems like it might be the symptom of increasing returns in the network, positive feedback, at some point, the whole thing empties itself out. You get into these economic questions of, there's the anecdote of the labor union coordinator and the car manufacturer walking through the plant, and he's saying, “Look at all these great new robots. They don't sleep, they don't complain, they don't strike.” Then the union leader says, “Well yeah, but they don't buy cars.”

So this tails into this other paper on which you're the lead author about automation and occupational mobility. So building on this understanding, looking at the network as a way to understand how people can transition through periods of change in the labor market. I'd like to hear, if you want to lay out a little bit about the thinking behind this piece and how you decided to dig into the data, and what data formats... How this piece methodologically differs from the piece that we just discussed.

Maria: Yeah. So as you say, in a way, the previous paper led into this one with those red alerts. We need to dive into those. We need to understand it? When we started with Penny seeing these red alerts, we wanted to see if we could do something more. The thing is we thought... Well, we think agents adopt, and hey move. As you say, we're not static. So could we include that in the model? So, do you think... Well, we think labor is important because it has had an impact in history. So you can see from... If you go to the medieval cities or cities that have medieval backgrounds in Europe for example, in Brussels, underground plaza, you'll see guilds, and in the guilds it, there were these institutions where, if you wanted to be a fisherman, you had to be affiliated to the guild.

Now, this was a way for the guild to keep track of how many fishermen they were and not assign too many. Because, if you have too many fishermen, you have a lot of supply of labor, and then wage goes down. By the way, this was also a way to control how many women or how much women could work. So it's funny how labor goes back all the way. You can think also of Jim Crow laws, when there was all of a sudden an extra supply of labor. People started to think, well let's make a law that will separate people. If you go to today's politics, the ideas of the president of the United States, the ideas behind Brexit, tend to be about securing the scarcity of jobs. So making our skills scares. If everyone can do what we do, we're not going to get a high wage. So when robots can do what we do, we might be into trouble. That's how this started leading here.

Now, what we did is we basically took previous research on some predictions of which occupations would be automated. Then we thought, well, people are not going to say, “Oh yeah I got automated. That's it. I'm unemployed, and I'm going to remain unemployed.” Yeah, I think they would fight back. How do they fight back? Well, we have data on job transitions. We have data that tells us where people move. So we thought, “Okay, let's build a model.” And we constructed... So the first thing is, we saw we cannot solve this only with data. As much as we like these maps that overlap, it wasn't going to give us the answer. We had to do a model, and we had to do a model that was based on what we thought was reality. So we thought, “What are the things that drive this?” Well, first there's some spontaneous process. People leave their job and new vacancies get open. That just happens just random, let's say. But then there's a force driving the economy. There's this landscape change. This is how the economy evolves that says, “Well, I'm going to push a bit and I'm going to make less vacancies, job vacancies, for taxi drivers and make more job vacancies in the healthcare system.”

So there's was this other force that would push it towards a new prediction of demand. The prediction of demand, we got it with the automated ability of occupations. I have to say here, there's a major assumption. We assume, in a way, that the labor and demand stays constant. Why do we do this? Well, historically, there’s the Industrial Revolution. Automation is not a new thing. If anything, it's the only thing human has always done. We've always been through technology, but what happens is it's not that we get massive unemployment. It's more like some jobs are destroyed and new are created. So our hypothesis was, let's take the predictions of the automated ability and say the ones with a low automation probability are going to increase their demand. The ones with high automation probability are going to decrease it. On average, we're going to stay constant. The model can be flexible and can assume different levels, but that was the starting point.

I have to say here, a caveat of this work, when we talk about automation probabilities, when Frey and Osborne and Erik Brynjolfsson, and [Mitchell and Raague?] put their predictions out there, it's the probability that it can be automated. It's not a probability that it will be automated. The difference is we might be able to make a robot for a waiter, but it might be just cheaper to pay someone. So, in that way, it won't be automated. But anyway, that was an assumption in the model. So, sorry. Recapping is some random fluctuations. This force that drives the economy towards new jobs, and then the adaptation of workers, and they transitioning between occupations.

That's how the model works. Three simple rules. That's it. We put the shock, we let it run, and what we observe is how much unemployment changes in each occupation, and how much longterm unemployment changes. Because, in a way, it's not very bad if you're unemployed. We could take a break. Just kidding. It sucks, but it's not as bad as if you're longterm unemployed. If you've been unemployed for a year, that's when it really, really starts getting bad. So, that's the other measurement we took into account.

What we saw in this model was that, for example, we take the example of childcare workers and statistical assistance. I like to say, if your nephew asks you, “I either want to become a childcare worker or a statistical assistant,”  you look at these automation probabilities. Childcare worker, not likely to be automated. Statistical assistant, likely to be automated. You might think, “Well, of course you should be a childcare worker. Then you're not going to be automated.” Your job is not going to be automated.

Our results tend to hint otherwise. Why? The thing is, people are not only their occupations, they're their skills and, with their skills, they can transition. They can transition in this network. The thing is the statistical assistance might be automated, but they have skills that allow them to transition to other occupations that are growing in demand. Well, childcare workers might not be automated, but they're easy to reach. In a way, childcare work is something intrinsic to humans. So other people that may be automated might go for that occupation. Again, talking about demand and supply, that's why. And this is something that network reveals. It's not straightforward to say automate-able occupations are bad, sorry, are going to suffer. Non automate-able occupations are going to grow. But it's somewhere in between. And to see that concretely, we have to go to a network.

Michael: So, it strikes me that this might be a useful lens to look back, like you kind of suggested a moment ago, historically, and understand why childcare has been basically invisible to the economy this whole time, that there is an abundance of... There's just so much supply on that end of the labor market that it's driven the economic value in the very rudimentary, coarse way that we have historically been able to evaluate that to a very low point. But then there's also... And I guess there's two things that you could address here. There's also the issue of, again, this map. When you talk about a disturbance to the network, changing the relative incentives that exist for someone within a given set of skills. What we're talking about here in the real world is not a kind of random bump to the table. It's endogenously generated. We were talking about this with Brian Arthur just recently, about the way that technology propagates new niches.

So the ways in which things are being disrupted are definitely slanted in a particular direction towards the new affordances that are created by these spaces. So I'm curious: the obvious question is, how do you see the changes that we're seeing in the technological landscape now? What is actually opening up in terms of the possibility  space, and why? How do you think that that's going to change the weights that we assign to various skillsets?

Maria: Interesting. So there's many things in that question. One is this system is... There’re feedback loops in the system, and it actually goes back to what you said about the beginning, about the person with cars and the labor unionists, and the labor union is saying, “Well, robots don't buy.” The thing is the economy has feedback groups. So what people buy and what they demand feeds into the network of firms, and those firms hire the workers. What I'm saying is, it's not a straightforward question because automation itself is going to displace workers that might change the demand for goods that will change the demand for labor. So it's an ongoing process. This work didn't focus that much... Well, doesn't focus on that. But for example, Matthew Jackson has a good paper on how the input-output network, which is this network of firms buying between each other, how it rewires, how it might rewire because of automation. So that's one thing, considering there's feedback loops and it's not that straight forward.

Maria: The other thing is... I think your question was, “Where are we headed?” So stopping the detour, where are we headed? This is just hypothesizing. I do think we're headed... There's going to be an increase in health, in the health sector. If there's one thing people care about, it's health for their family, for their friends, and living longer. The longer we live, the more we demand on health. So, that's one thing that I think is going to grow.

The other thing is analytical skills, being able to interpret the world. That's one driver. I think that's the positive scenario. It's the positive scenario because, and that might be my opinion, but I think those are work that we tend to value highly, because what we do as intrinsic to our value, and it's important for people to have jobs they have dignity with and they are proud of. It's important for people to be proud of their work, because it's eight hours a day. Okay? So we need people to be proud of it.

So I think that's the good scenario: where we educate people on everyone does interesting jobs, does health, does science, and understands the world, ultimately. The bad scenario is where it actually, we have so many unemployed people. Well, as I said, it's not going to hit massive levels, but we have a lot of people whose skills is not compatible with the health and science and all of that. Then we say, well, it costs energy to pay to have a robot, but there's people willing to do it. So let's just pay a low wage to those people. That is actually a bit dystopic, because that that means we're going to split people into the ones that can do high skill and the ones that can do low skill. This is actually a bit related to what Harari says. So Harari puts this dystopia, right, that maybe some people are the ones that are going to be able to buy goods that will amplify our capabilities. So we're going to expand their memory. And he, for example, talks about blood transfusions that might enable you to live longer. So there's people that are going to be able to afford it and the people that cannot. That's going to create two types of humans in a way, and that's scary.

I really hope we don't go that way, and I hope the way to ensure that is through retraining, because I think everyone can do a high school jobs if they receive the right education and the right motivation. I think that's the core of what this work is about. It's about job transitions. It's about what retraining do we need to get into, let's say, the steady state we want, the attractor we want. Well, in a way, we’re in the same attractor, but into the fixed point, or around the fixed point we want instead of the other. We have a choice. Well, I think we have ... I'd like to think we have a choice. Where are we going to push this system? Is that the one where everyone can go to where every job, because we have education or... Well, not every job, but to a job with dignity that is important to them. Or is it the other? I think that's where we need to really push for the science.

Michael: Yeah. It seems that, when I zoom out and I take this absurdly large macro-evolutionary view of the history of the planet, and that there's this ratcheting through major transitions toward agents that are capable of better, more adaptable models, broader perspectives, very generally speaking, that the kinds of intelligence required by someone living today, much more abstract than the kinds of intelligence required perhaps 500,000 years ago. To zoom in again, there's an enormous spike that I'm observing just online in conversations around sense-making, and that it seems like this is where your work connects to the work of people like Jessica Flack, looking at society as a collective computation and where it connects to the work of people like Andreas Wagner, looking at shocking bacterial cultures with intense selection pressure, and how that accelerates the mutational search algorithm that you see that culture going through. It sounds to me like where your work links to these other people is in suggesting that there's a sort of...

Just as economists say, if you really want to put a meta-investment down on something that's going to provide a diverse array of dividends, it's on education, it's on helping people learn to explore and adapt to a rapidly-changing world that, in important ways, cannot be anticipated. And so there's... I think I would like to remain hopeful that sense-making and creativity, that these are the things that we're going to move into naturally as a species. But I don't know. What do you think about that?

Maria: Well, I'd say I'd like to be optimistic too. I love how you talked about organisms and computing as a whole, because I think that's vital. We're in an age where it seems to be okay to divide people into us and them. And that's not how we're working. We're actually working as one organism. Maybe it's a “bacterium,” and there's some external shocks, and we have to deal with it, but we also have the advantage that we communicate. Well, bacteria communicate in one way, but I think our communication is a lot more complex, though some people might not use the whole complexity. But yes, I think we need to remain hopeful. I think we need to think about the big picture, think about how the whole landscape is changing, and think about how we're going to adapt. And planning different strategies. Seeing, “Do we want to go this way, the other way?” Maybe we can understand that with bacteria. Maybe we can do some micro simulations. Maybe we can understand it in coding. Just doing an agent-based model and seeing what incentives you need to give.

In a way, this is like control theory. You have one policy, or several, and you can play around with it. We need to hope that we do it the right way. Honestly, there may be many right ways. We just have to get… And that's a one thing I'm optimistic about. I don't think it's an Avengers-type of thing where Doctor Strange says, “There’s so many multiverses and we're only successful in one.” I think we actually can be successful in many ways. There's a lot of proposals. There is universal basic income, there is... Many countries have free healthcare. Many countries have free education. One way or another, all of those might lead to it. Another thing I'd like to mention is, when I say we need to start thinking of an organism as a whole, I mean the whole world. When I mean we need to make our models valid with data, I also mean the whole world, which means we need to push to have data for developing countries.

I know SFI is a driver of those. It has groups of archaeologists going to different countries and collecting data. I think we need to push for that, because this is something that happens a lot in research. We want to publish in top journals. To do so, we need top data quality. Which countries have top data qualities? They're developed countries. Ironically, I'm a Mexican studying the labor market for the US. I would actually like to do it for Mexico, for Latin America, for a lot of other countries, but it is true that the US has the best data, or one of the top data [sets]. So I also think, as scientists, we have a moral obligation to look for that data. It's not only moral. It gives us more about science because it gives us different environments. We're one organism in different environments. How do we develop? Overall, I think science will win with this.

So yes, I think my big picture is, we need to start seeing how do we get to the state that we want. How do we get together on it? We cannot forget about one part of the world because we're in a globalized world, so there's no way that's going to happen. And I am optimistic. As I said, I think there's many paths that can lead to it. I think we have the right intuition, but we have to... Paul Romer said it at some point, we have to decide. That's the one thing, we sort of know which are the policies we need for climate change. It's not totally clear, but we need some retraining. Well, let's decide. I think this... I would cite Romer, and say, at some point we have to decide, and that's going to be it.

Michael: So, to bring this back down to the human scale and wrap it here, I'm curious: you're young, you're talented and intelligent, and relatively well-positioned in your career, and yet you spend every day thinking about the turbulence and the change, and the disruption that is facing us and will be facing us for the foreseeable future. So what insights have you taken from this and into your own life and how you imagine your lifelong strategy for navigating this? You are the node. You're sitting on one of those little islands in the network. So what does this mean for you, Maria, and how is it shaping the way that you plan and prepare for the rest of your life?

Maria: Wow, that's a big question. I have to say SFI has played an important role. Doyne Farmer, Francois Lafond, a lot of people have .. Penny, have played an important role in this. They're sort of my guidance. Someone once asked me, because I'm one of those people that has always been like, “Oh, I think academia is what I'm going to do if I can.” They asked me, “How can you be so sure? I kind of don't believe people can be sure.” I said, “I'm not sure at all, but I know the things I can't do.” It's more about, I just know there's some things I cannot do and then there's only one path. Yes, in a way, every day I'm like, okay, I see the news and I'm like, “Oh, this might be going down,” but then I can't just go and solve the world. Maybe I should, but I don't see it in my adjacent possible. There's all this framework. What's adjacent and what's possible. And the one thing I can do is keep on working on these models, trying to make them better, hoping that...

I do think sciences is brought up in a collective. So I think I just want to contribute little by little. I enjoy it. I like it. It has ups and downs. I'm not going to lie, I'm not a person that wakes up every day and says, “Oh yay, I want to go to my computer and started this coding and start doing the math.” No, there's days I'm like, “Oh God, I wish I could stay in bed.” Or “Oh God, I wish I could just go for a longer run or stay and chill out with friends.” It happens, but it's not everyday that I want to do it. But I'd say maybe one in five days, I wake and I say, “Yes, today I'm going to write these equations, and I'm going to see if they work.”

Sometimes they happen consecutively, sometimes it's weeks that I'm like, “I have to do this.” Sometimes there's weeks that I don't want to do it. I still go to the office. In a way, it's still work, but I do love it, and I don't see another path. That's my personal view. I think different people find inspiration different ways. For me, SFI has been one way, and I'm not sure where I'm headed. It might be academia, it might not be, but somehow I'm optimistic that it's going to work, because it has to. If I don't think that, then I don't know how I'm going to wake up every day. But so far, so far so good. I'm enjoying it.

I don't know. To all the PhD students out there, I think we all struggle, and it's part of the path. We also have to learn to enjoy the little victories, because there's going to be a lot of defeats, but there's enough good in this world to just go for it.

Michael: I remember my friend Mark Nelson, who was one of the eight people that was locked inside of Biosphere II for two years here in the American Southwest. He said that hope is a form of yoga, and he said it was about remaining humble to the possibility that your models are wrong. So I like this, that you keep bringing up optimism because, in a way, I would make a case, I think, that optimism is actually humility to the possibility that your despair is mistaken. Anyway, Maria, this has been a pleasure, and I'm glad that you're doing this work. I imagine it will be very illuminating and helpful to a great many people, and situate you well for the numerous disruptions to come.

Maria: Well, thank you very much. I also hope it helps some people, and yeah, thank you. Thank you to SFI. It's a great journey to travel so far.