Professor of Geophysics; Vice Chair for Graduate Programs, UC Davis
By David Zierler, Director of the Caltech Heritage Project
July 25, 2022
DAVID ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Monday, July 25, 2022. I am delighted to be here with Professor Magali Billen. Magali, it's so nice to be with you. Thank you for joining me today.
MAGALI BILLEN: Thank you for having me and giving me this opportunity to share about my time at Caltech and what I'm doing now.
ZIERLER: First things first, you have a very unique first name. Tell me what it means and the origin story.
BILLEN: Yes. I'm from Belgium, and Magali is a name that probably comes from the name Magdalena, brought over into Western Europe by the Roman people in Hungary and that area. It was actually popularized in the 1960s with the repackaging of an old French ballad as a love song. And my mother, of course, was a teenager in the 60s, and that's how I ended up with my name. [Laugh] Not a very common one.
ZIERLER: On a more official level, what is your title and institutional affiliation?
BILLEN: I'm a Professor of Geophysics in the Earth and Planetary Science Department at UC Davis.
ZIERLER: Tell me about Earth and Planetary Science at UC Davis. Is it a big program? Is it in growth mode? What are some of the exciting things happening right now?
BILLEN: Right now, we have about 20 faculty. We're a little bit smaller than we were seven or eight years ago. We've had a lot of retirements, and the program's slowly figuring out how to build back up after that. We're a very diverse program. We cover everything from paleoclimate, paleobiology, paleontology, through the solid earth into planets and the formation of the solar system. With only 20 people, it means we're each very specialized, and we have to work to look for the overlaps in our research to create a vibrant place for our students to learn. We have a lot of people who work across the different disciplines.
ZIERLER: Just as a snapshot in time, what are you currently working on?
BILLEN: Right now, I'm working on deep earthquakes. It's a new direction for me and my research. I normally work on the long-term evolution of sinking tectonic plates, but now, I'm trying to link that deformation to how earthquakes happen on a shorter time scale in the sinking plates that go through the upper part of the mantle.
ZIERLER: Deep earthquakes, shorter time scale. How deep is deep, how short are the time scales?
BILLEN: Normally, I work on how plates move across the surface of the earth and sink through the upper mantle on the time scales of 10 to 100 million years. Really, how do we actually pull plates across the surface of the earth? Once they sink, they're denser than the surroundings, and that's actually what moves the plates across the surface. That's the long time scale. I'm interested in how the slabs sink through the upper mantle, which is about 600 kilometers from the surface of the earth. There's a big change in the material properties of the earth at about 660 kilometers, so everything sinks faster through the upper part of the mantle and slows way down in the lower mantle. It might only take seven million years to get 660 kilometers deep, but it might take 100 million years to get all the way to the bottom of the mantle after that. So there's a real difference in how fast things evolve in those two different parts of the mantle. But now, I'm shifting to really short time scales, like how earthquakes happen over a hundred-year time scale in the sinking lithosphere, and what that tells us about how that plate deforms as it journeys through the upper mantle.
ZIERLER: What are some of the observational or theoretical changes in the field that have prompted this shift for you?
BILLEN: A big part of it has been the development of open-source code that allows me to be part of a community of scientists developing mantle-convection codes. It's kind of interesting, it's not that there are new observations, it's more that the numerical methods have allowed us to ask different questions that we've had for a long time. There's been a paradigm in the field about deep earthquakes, that they happen with a particular mechanism. But if you look at the observations, it's never quite been a good fit. It has certain predictions about where there should be earthquakes in the plate and where they shouldn't be. And it doesn't quite match. Those observations inspired me to look at some of my longer-term models in a different way. I realized that where I see deformation in the long-term models actually matches the pattern in deep earthquakes.
I decided to actually bring in some of the material properties that we usually ignore at the longer time scale that allow us to actually store stress that could cause earthquakes. I'm really trying to basically take what we thought might've been a solved problem that wasn't quite working and figure out how we can look at it differently. It's a combination of the numerical code that's available now, looking at the observations and saying, "Why doesn't it fit better? It should fit better if we really understand what's going on," and trying to figure out how to do the observations and models that will allow us to predict where the earthquakes are happening now and why they're there.
ZIERLER: What were the misapprehensions that suggested this was a solved problem when, in fact, it wasn't, and best case scenario, what will it look like when it is a solved problem?
BILLEN: The accepted theory was that deep earthquakes happen due to a phase change within the slab not happening. You have olivine in the upper part of the mantle that has to transition at about 400 kilometers' depth to a phase that's stable at higher pressure. But it's so cold inside the sinking plate that actually, that transition might not happen where we expect it to happen. It may stay unstable and go deeper and deeper. But because it's unstable, when it finally transitions and goes into the deeper phase, that can happen very suddenly and can trigger an earthquake. That's been the predominant theory, and that's called transformational faulting from metastable olivine. The problem with that is, it makes predictions that you should have these deep earthquakes in places that are very cold, where we have slabs sinking very fast with very old plates. But what we see is that we actually have deep earthquakes everywhere that slabs are deep, whether they're young and warm or old and cold.
And that's the biggest issue. And the other thing is that we sometimes see them very deep, but not in the middle depths. We'll see them at 500 to 600 kilometers, and we'll see them down to 300 kilometers, but then there'll be a big gap of earthquakes in the middle. And that just doesn't fit with the predictions for metastable olivine. But what I found in the numerical models I did is, one thing that had been missed is that metastable olivine doesn't just depend on this temperature effect. You actually have to be deforming the slab fast enough to make that transformational faulting occur. It's an instability. If you deform too slowly, the transformation will occur, but it won't occur suddenly, so it won't cause an earthquake. What I see in the numerical models I do is that some parts of the slab really just don't deform very much. The deformation is localized. You'll have a bending region where there's lots of deformation, and below that, there's a part of the slab that really isn't deforming very much. Even if there's metastable olivine there, we might not have earthquakes.
If that's true, it could be that what we've been missing is, we haven't been accounting for the deformation of the slab. The other problem is, we haven't actually observed this metastable olivine wedge in the mantle except in one place in the world, only in Japan. There are other mechanisms that could be causing deep earthquakes. There's something called thermal shear instability, and that doesn't depend on there being this metastable phase. What I envision is a shift from having one mechanism that explains everything, this one-model-fits-all approach, to understanding that we may have multiple mechanisms that are causing deep earthquakes and that the deformation of the slab itself is controlling where the earthquakes are happening.
ZIERLER: Is what's unique about Japan the geology there or some observational advance that's only available there?
BILLEN: It's about the network of seismometers they have. They have such a high density of seismometers in Japan right on top of this deep subducting plate that they have the ability to see and make observations you can't make anywhere else in the world. We would love to look for this metastable olivine wedge in places like under South America or under the Tonga-Kermadec subduction zone, which is north of New Zealand and has a huge number of earthquakes, but we just don't have the density of seismometers necessary to make the kind of observations that are needed.
ZIERLER: So this is a resource issue more than anything else. Theoretically, these things could be all over the world, and we'd know much more than we do now.
BILLEN: Right. But the chances of having that resource created in another place on a short time scale is probably pretty low. In some ways, we're using the numerical models to ask if there's another way we can answer this question of whether we need to have metastable olivine in order to cause deep earthquakes or if another mechanism works just as well. I think first, you have to show where the holes are in the paradigm. You say, "We've been saying there's just this one solution. Why aren't we considering the other ones and what predictions do those make as well?" That's the first step. The second step is to use these other mechanisms to show whether we can actually reproduce the observations we have. For example, we can, in the models I'm running, pull out what would be a predicted rate of seismicity in a relative sense, places that should be higher or lower, and we can also pull out the orientation of the stresses, both of which we can also get from deep earthquakes themselves.
There are ways to bring in the observations to move beyond metastable olivine, which just gave us a depth-dependent pattern of seismicity, to say, "We actually have a lot more observations. Can we observe all of those?" What I'd ultimately love to do is actually link with people who are studying the rupture processes of deep earthquakes to see whether there's something in the observations of the deep earthquakes and how they rupture that we could link to the mechanism, if the time-moment release is different if it's a shear instability versus metastable olivine. There are definitely observed differences in the seismic waves that come from deep earthquakes and some other features of deep earthquakes. And people have kind of been at a loss to understand why. It's all those questions we don't know. "There are differences, but we don't know why." Then, going back and saying, "Those are questions we should answer. We should know why, we should be able to explain it."
ZIERLER: If I could zoom out, a very general question about deep earthquakes. Absent the dream scenario of unlimited drilling capability, how can we realistically know and understand these things about deep earthquakes without being physically close to them?
BILLEN: I always think about it as being really, really clever about how you use your observations, then using numerical models to check your intuition. I think our intuition leads us to see things in the data that make us ask really good questions. That, I think, is what leads us to discover things. The numerical models are really, once we have an idea, once we think we might know what's going on, how we test whether it's actually physically possible. We have a lot of observations, a ton of seismic data from these earthquakes. We just haven't had the numerical models to link them to, to say, "Why is it that a deep earthquake has a different frequency content than shallow earthquakes? Why do they have a larger range of stress drop than shallower earthquakes?" These are all things we've observed, and nobody knows the answer yet. I think we can build models that will allow us to answer these questions, but we don't have them yet. We don't have a model of transformational faulting of metastable olivine. What we have are lab experiments on that process on analog materials.
ZIERLER: What's missing to create these models? Data? People power? Computation?
BILLEN: I think it's computational. It's an instability, and you basically have to fracture the rock. It's developing that model that would be the next step. This is why I'm trying to test this other model about thermal shear instability, because we have models of thermal shear instability, so we can test that hypothesis because we have a numerical model that can actually calculate when a thermal shear instability can occur. Nobody's developed that framework for metastable olivine yet.
ZIERLER: I wonder if you can walk me through a scenario where, as you say, a numerical model can be tested against your best intuitions. What does that look like?
BILLEN: I'm doing this right now. We have this guess, basically, this idea that where the slab is deforming the most is controlling the pattern of seismicity. We're creating models for individual subduction zones, where we create a synthetic subduction zone within the numerical model, and we calculate forward in time what the buildup of stress is in that slab, so we're building up elastic stresses. Then, we sample a bunch of locations in the slab in terms of the strain rate, the stress, the pressure-temperature space, then we put that into a thermal shear instability model and ask, "Under these conditions, would we actually form a thermal shear instability?" Then, I can map out where in the slab the earthquakes can occur, whether it matches the pattern of seismicity we have now, whether I see things like the alignment of the stress rate being in the right orientation, the right relative variation in the rate of seismicity, whether it matches the differences in the rate of deformation we see in the slab.
It's really taking a synthetic subduction zone and trying to predict what the outcome would be. The models I use are just predicting where thermal shear instability would happen. The next step would be to actually model the dynamic process of a thermal shear instability and the seismic waves that would come out of that. That would allow us to directly compare to seismology. And I'm not a seismologist, so that would be something where I would need to collaborate with others. I've been working a lot in the last two years to learn more about how people are modeling shallow-earthquake rupture and what they observe from them to try to understand what we could observe from deep earthquakes to do the same kind of analysis.
ZIERLER: It's obvious there are so many exciting advances making this research possible. If you could give a historical sense of just how far the field has come. Imagine you're in a time machine, and you go back 50 or 60 years to some of the founders of the Seismo Lab. What aspects of this research would be totally alien to them, and what aspects would be intellectually deducible through extrapolation?
BILLEN: I think a lot of these ideas have been around for a while. Thermal shear instability was first proposed in the 1970s as well as metastable olivine. I think the big difference is that those were all done through analytical models and experimental results with no way to link them over the time scales of deformation, and that's really where the numerical models come in. Also, to really ask, "If put in the material properties for the earth that we think exist rather than an analog in a lab experiment, does it still hold together?" A lot of these ideas are based on lab experiments done in conditions that are not the same as the conditions in the mantle, then you have to extrapolate through very big changes in strain rate, rate of deformation, and also very big changes in the pressure and temperature conditions. Just to give you an example, most experiments of deformation in the lab are done at strain rates of 10 to the -6. Doesn't matter what the units are. But that has to be extrapolated down to 10 orders of magnitude slower processes in the mantle, and that leads to a large range in uncertainty in terms of whether what we see in the lab could actually be happening in the mantle itself.
ZIERLER: Two questions in sequence, and I won't know how to ask the second until I get your take on the first. The big question, earthquake prediction. Do you believe that earthquake prediction is possible at some point in the future, or are earthquakes inherently unpredictable?
BILLEN: I'm not sure I'm the best person to answer that. I would say that for deep earthquakes, it's not important to be able to predict them as much as to understand what's controlling where and why they happen. Because they're so far away that they very rarely cause hazards. I would like to think it is possible you might eventually be able to predict earthquakes, but I still don't know enough about what the inherent processes are to say whether they're instabilities that just cannot be predicted in a case-by-case basis. My guess is it'll be statistical. It's not going to be, "We monitored this one place, and we know when the next one will come."
ZIERLER: Is that to say that there are, however infrequent, some deep earthquakes that are capable of causing damage at the surface?
BILLEN: There are. First of all, people define deep earthquakes as anything from 50 to 660 kilometers' depth. I call things that are above this transition to where metastable olivine can form intermediate-depth earthquakes because there are actually other processes happening at shallow depth that have to be taken into account. Earthquakes that happen 100 kilometers beneath the surface, especially if you have a slab that sinks almost horizontally, and there's a city sitting on top of it, seismic waves can come straight up, and that can cause damage. In places like Mexico City and Tokyo, earthquakes in the slab can actually cause damage and seismic hazard in the plate right above it. But deeper than that, it's very unusual for deep earthquakes to cause damage. There are a couple of examples. There was one in Madrid at around 500 kilometers' depth that actually caused brick buildings to topple, and there were people killed. It was quite a while ago. But most of the time, deep earthquakes are not damaging in any way. They really, I think, can be an amazing probe into how plates actually deform as they go through the upper mantle. Using this observation of the earthquakes is, in some ways, another way to probe what's happening at depth and link that to experiments we have in the lab.
ZIERLER: That gets me to the second question. If it's possible, and the jury is obviously out on whether it is, there's a tremendous public policy interest in getting to a place of predicting earthquakes there are near the surface. My question for you, in your area of expertise, it's one planet at a level we don't currently understand, and all of these things are fundamentally connected, so how might some of the advances that you're making in deep earthquake give us broader understanding that might get us closer to earthquake prediction for those that are damaging at the surface?
BILLEN: There is a direct connection there. The models I'm developing are going into this time scale where we're looking at the buildup of elastic stresses, but within a larger framework of the overall deformation of the subduction zone. And that actually allows us to do the same kind of analysis I'm doing at depth at the surface. We can actually start looking at the buildup of elastic stress over time scales of thousands of years and how that might build up to the pattern of stress along a subduction zone. For example, along a thrust fault. We know from rupture models that heterogeneities in that pattern of stress can be really important for how large a rupture is, how much energy is radiated. The models we are doing can basically create a background stress state for people who are doing the rupture modeling to actually take into account. By linking that, we might be able to do much better predictions or forecasting. Like, "Do we expect a magnitude-9 earthquake along this boundary or not because the stress state is more heterogeneous or there's other ability to release that stress over the long term?" I think, going forward, linking the kinds of models I'm doing where we can actually track buildup in elastic stress will really help with the shorter-term rupture dynamic models, having a much better starting state that is more linked to the real stress state within a subduction zone.
ZIERLER: Just to get a sense of the scale of this project, are these the kinds of questions you envision occupying yourself for the next 5 or 10 years? Or is this the rest of your career, your graduate students, and their graduate students?
BILLEN: Actually, this comes into a much larger framework of, like, 10 to 15 years. You talked to Emily Brodsky before, so you know about SZ4D. There's this big initiative within the US earth science community for studying how subduction zones evolve over space and time directed at trying to understand hazard. And that's going to be, like, a 10-year project of gathering data, linking models and observations together so that we can better understand the observations we have. What I've worked on with the deep earthquakes, through my interaction with SZ4D, I've see that there's this potential where the kinds of models I'm doing could be a very important source of understanding the buildup of stress in different subduction zones, which could be used in the SZ4D framework. There aren't a lot of people doing that right now. I think it's a new direction for people to try to link long-term subduction dynamics to the hazard question, which just hasn't been a priority before, mainly because of the numerical part. We just didn't really have the resolution with the larger-scale models, the computing power to actually calculate using realistic material parameters that would allow us to link to that shorter time scale at shallow depth where there is seismic hazards.
ZIERLER: To really zoom out, at Earth and Planetary Science at UC Davis, is your focus strictly terrestrial? Are there observations on other planets that might be relevant for the questions you're after?
BILLEN: There's only one seismometer on Mars. [Laugh] And we don't know if there's any subduction. It's possible. That hasn't been a direction that my work has gone. I tend to collaborate more with the geologists in the department who are making observations, trying to understand mountain-building processes, which happen above subduction zones through earthquake after earthquake. But on the planetary side, it's interesting to think about this. It's possible. For understanding earthquakes on the moon, you basically have a thermal cooling of the moon, but you still have active earthquakes. Trying to model that with the elastic release of energy could be a direction. I haven't thought about it, actually. Most of the people in my department are destroying planets by running them into each other or working on icy planets. [Laugh] We haven't made that connection in terms of earthquakes yet.
ZIERLER: This has been a great overview of your research. Let's go back and develop some personal history now. Coming to the United States from Belgium, what were the circumstance? Was this part of a family move? Did you come here for college?
BILLEN: I moved to the US when I was only 2 months old. My mom came here with my brother and I. She was following her mother, who got remarried to an American. My mom saw that there might be more opportunities for us in the US. I wasn't really aware of being different, you might say, than other Americans. [Laugh] Except when I realized the food we ate was really different from other people. We cut using our fork and knife differently than other people. Then, there was a very funny episode when I was in high school, and a friend called, and my mom answered the phone. This was when we used to have phones on the wall in our house. My mom gave me the phone, and my friend asked if I had a French maid. [Laugh] I was like, "I don't think my mom would be happy with that reference." That was the first time I realized my mom had an accent. Moving here was really just an opportunity for my mom to give us a different life and different opportunities. Which it did. It gave us opportunities to go to college and to end up where I am now.
ZIERLER: What were your first languages growing up?
BILLEN: We spoke English. Only a little bit of French. My mom was very focused on learning English so she could work, so we didn't actually speak French in the house very much. It's one of the things I wish had been different. But I mostly learned to speak French in college, writing to my grandfather in Belgium. Then, I married a German. Now, all my French is gone, and I can speak a little bit of German. [Laugh]
ZIERLER: Did you grow up in the Northwest? Was Puget Sound close by?
BILLEN: No, I grew up in San Diego. It was during earlier droughts in California, and I really wanted to go someplace greener. I was only 15 when I finished high school. I skipped several grades. And coming from a family who didn't grow up here, we really didn't know how the college system worked. I went to a college fair, and I was lucky enough to talk with a nice recruiter from the University of Puget Sound in Tacoma, and it just seemed like it would be a good place. A small school where I could be comfortable, since I was younger, and not be overwhelmed by a big university. I actually remember sitting in my room with a stack of university packets, trying to figure out what in the world to do. Just the connection with the recruiter is what got me to apply to UPS.
ZIERLER: Were you on the math and science tracks from the beginning in college?
BILLEN: I was. Actually, I thought I'd be an engineer. Growing up in Southern California, I'd been interested in designing bridges. I was interested in kind of the earthquake part of it, how to build bridges that were stable in different kinds of earthquake forcing. But when I got to UPS, I learned there really wasn't an engineering program. I was really naive and didn't know what I was doing. But they had what was called a 3-2 engineering program, where you majored in math, physics, or chemistry, and you'd transfer to an engineering program. Basically, I went and met with my advisor, and he told me this. I slept on it, and the next morning, I was like, "All right, I guess I'll be a physics major." That was it. [Laugh]
ZIERLER: Did you have any exposure to geophysics or earth science as an undergraduate?
BILLEN: Not really. I was a physics major, and the thing that really got me into geophysics was, there was an NSF-funded workshop that was held at the University of Oregon for female undergraduates at small liberal arts colleges. Very niche. But it basically brought women undergraduate science majors from a lot of the small Northwest liberal arts colleges to this workshop, and they introduced us to what they call the hyphenated sciences. I had never really heard of geophysics before, or physical chemistry, or biophysics, any of those things. They took us to labs, they showed us what it meant to be at a big university, and that's really where I learned about geophysics. Then, I started talking to people, asking what it was about. There was a women in physics ListServe in the early days of email, and a few people basically said, "Here are some good schools," and that's where I ended up applying. The funny thing is, I think Caltech is the only place I got in. [Laugh]
ZIERLER: It's usually, "It's the only place I didn't get in."
BILLEN: I know, and I think it's because people just didn't know what to do with me. I graduated when I was 15, I was a physics major at a small college, I had done some research during my senior year, but I never took geology as an undergrad. I think Caltech was like, "No problem. You'll learn geology." They wanted people who had that strong math and physics background. When I got to Caltech, I remember Don Helmberger used to walk into our office all the time because his students were in there, and I think I asked him, "I don't really know any geology. What should I do?" He was like, "Well, here are two books." I just taught myself intro geology during the summer. It was like, "Okay, now I'm a geophysicist." [Laugh]
ZIERLER: Was the plan the Seismo Lab immediately? Or generally GPS, then you sort of migrated over to the Seismo Lab?
BILLEN: I was definitely interested in seismology and earthquakes. I thought I would work with Tom Heaton. Actually, I worked with him for my first year. I wanted to work on something that was going to have strong societal relevance.
ZIERLER: Earthquake engineering, there you go.
BILLEN: Exactly. That connection with him and engineering seemed like the right place. I did some modeling of earthquake statistics in the first year. But Tom was hard to find, he was in different places at different times, and we just didn't quite get that connection. I did another project with Don Anderson and Mark Simons, who was a post-doc at that time, on a method of comparing geoid and topography. And Mike Gurnis, who became my advisor, was my academic advisor at the time. It was actually kind of a nice thing. My first year, we worked on various different projects and were given a lot of latitude to just explore and learn about things. I remember spending hours and hours at the Seismo Lab coffee, just listening to what people were working on. Then, I got into subduction really because Mike asked me. He was like, "I have this grant. Do you want to work on this?" I actually remember telling him, "I'm not sure I should do this because I don't know anything about subduction, and I've never done finite element modeling before." He kind of looked at me and was like, "Aren't you here to learn that?" and I thought, "All right, yeah, I'm not supposed to already know this stuff."
ZIERLER: Was Mike working on subduction? Was this related to his research?
BILLEN: Yeah, it was his research. I think he had recently gotten an NSF grant and decided I might be a good fit for that research.
ZIERLER: What modeling were you going to learn?
BILLEN: Finite element modeling. It's a numerical method for modeling–basically, you take a partial differential equation, which describes the physics of your conservation of mass and momentum, and you discretize that. You basically take your whole space you want to model and break it up into a bunch of different chunks. The finite element method is a way to do that mapping from a continuous solution to this discrete solution, where you only have to solve it at a bunch of points on a mesh. Finite elements is one of the ways you do that.
ZIERLER: What were some of the big questions Mike was after at this point?
BILLEN: One of the really big questions at the time was how to explain the gravity signal above subduction zones. Basically, the prediction from long-wavelength studies was that you should have a gravity low over subduction zones. But most subduction zones have a gravity high. That was a bit unexpected. You would also expect, because the slab is sinking and pulls down on the surface of the earth, that it would have a low topography, and it doesn't. We were trying to understand why that was. The first paper I did with Mike was focused on the hypothesis that if you had a low-viscosity region in the mantle right between the sinking and overriding plates, that would decouple the sinking motion of the slab from the overriding plate, you would no longer have a low topography or low geoid, and that hypothesis was correct. And it was kind of interesting, partway through my PhD, Dan McKenzie was visiting Caltech, and I went in to explain to him what I was doing. And he was basically like, "Well, of course that's what's going to happen. We showed that in such-and-such paper."
And he did. He showed that if you decrease the viscosity, you'll decrease the size of the topography. But the key was, we were actually figuring out how you could make the viscosity lower in a physically consistent way. I showed that it if you took into account the fact that you were melting the mantle above the slab through hydration, water coming off of the slab, you melt it to higher degrees than you would otherwise, and that would allow you to have both a lower-viscosity region and a more buoyant mantle than you would otherwise. And when you put that process into the numerical model, you can actually match the observations of both the gravity and the topography over the subduction zones we model. It was taking a concept that I think was already understood from Dan McKenzie's analytical solutions and linking it to the physical processes that could actually make it possible.
ZIERLER: To clarify, when Mike initially gave you this problem, was he already your thesis advisor? Or was this what clicked to make him your thesis advisor?
BILLEN: That's what made him my thesis advisor. I did my qualifying exam the beginning of my second year. He was on my committee, but at the time, I wasn't working with him.
ZIERLER: Once you had the thesis topic set, was there a lot of field work, or was it all data coming in?
BILLEN: This is where it's kind of confusing because we pursued two totally different things. I wrote a paper with Joann Stock based on kind of a mixed project that she and Mike did. The ship we went out on was the Palmer, and every now and then, they would move the Palmer from one port to another. It wasn't being used, just moved. They came up with this idea that they could send a couple students on the ship during that time to collect data, mainly magnetic and bathymetric data, that would help answer some key questions about plate tectonic reconstructions that Joann was working on at the time. But Mike was also really interested in how subduction zones are deforming where the plate actually bends. We ended up taking the opportunity, while these ships were leaving, to take data in a particular place. I was sent out with a couple other graduate students to just collect data with nobody else on the ship except the people running it, which was kind of interesting.
And that led to several projects. I did this paper with Joann, where we constrained the age of the plate north of New Zealand, which wasn't known before, and we modeled the magnetic anomalies and showed it was actually younger than people thought it was. And that's kind of cool because every time I see an age map of the plates, my data's in there. There's a dead ridge I figured out the age of. That was my first-ever paper as a graduate student. But then, we also collected data of bathymetry and gravity along the trench that we were interested in trying to ask, "Can we actually figure out what the elastic response of the plate is?" There's a way of evaluating the gravity and topography to try to back out the material properties of the plate. And that actually was a failed chapter of my thesis. We just couldn't get it to work. But after I finished, I finally figured out how to make the method work, even though we didn't have ideal data, so we ended up writing a paper on that as well.
Then, I also was doing these longer-term subduction models, trying to understand the role of this low-viscosity wedge in explaining the gravity and topography signal over subduction zones. I had this multicomponent thesis that just grew out of these special opportunities. And it was great because it gave me–I'm a numerical modeler, but I had this huge appreciation for firsthand data. I understand where all of this marine geophysical data comes from, what the uncertainties are, and how to use it. If you just do the numerical modeling, you're very limited in the questions you can ask because everything we do comes from seeing the observations and recognizing thata particular observation would give you a constraint on what's happening somewhere deep in the mantle. One problem is there's non-uniqueness. The gravity and topography problem, there are all these tradeoffs. I can put a layer at 660, at 410, or 900, and the surface gravity and topography might not know or care that a jump in material properties exists. But in the numerical models, we can make it much more complicated than the original models were and add in more realistic material properties that then allows us to understand what we're actually sensitive to in a much more realistic model of subduction zones.
ZIERLER: Looking back, what were some of the computational advances that made this research possible? And given the access to the tools you have now, what was labor-intensive then relative to what could be accomplished today?
BILLEN: The main numerical advances have really just been computer power. Being able to run models at much smaller resolution to allow us to actually resolve the slab. The slab is only 100 kilometers across, and there are variations in the material properties that mean I need to have one-kilometer spacing between my elements when I'm calculating that. At the time I did my thesis, I don't think I could go below 15 kilometers. We have models now there are 10 times higher resolution just in one dimension, so they're, like, 1,000 times bigger than they were at the time. Not only that, because I have more computing power, I can put in more realistic material properties. At the time, I was using viscosity that was only temperature-dependent, basically, which meant where it was cold, it was stronger, and vice versa. But it turns out that the rocks are actually also stress-dependent.
The more I stress them, the weaker they should get. And that leads to much more heterogeneous structure, which is harder to solve numerically, which means it takes more iterations to solve it. My research has really been kind of a long progression in how to add more and more realism into these models. Because I know the models we did in the past aren't right. And we know there's even more stuff rocks do that we still haven't included in our models. I'm trying to build up numerical models that actually take into account the real deformation of rocks. The stress-dependence is important, the yielding is important, the fact that their grain size is evolving by many orders of magnitude, the fact that rocks deforming create a fabric, and that fabric has heterogeneous properties, if you deform it in the direction of the fabric, it's weaker than if you try to deform across it, all of these things are real properties of rocks that we're just beginning to put into our numerical models.
And we don't fully know what the effect will be yet. People will look at these models and say, "Look what we can show. We can match this, this, and this." I look at it and say, "Well, that's good, but we still can't match these other observations. And it might be that we have what seems like a right answer that's just completely wrong because we haven't put in all the complexity that exists." I'm very skeptical of every model, whether produced by myself or others, because I look at it and see what's missing from the real world.
ZIERLER: To clarify, you're experiencing this skepticism in real-time as you're developing the thesis. This is not retrospective, looking back.
BILLEN: No. It's really that we know, "Here are the numerical limitations we have now. Let's try to move forward as much as we can within those numerical limitations, knowing that we're making simplifications along the way."
ZIERLER: On that basis, between either you or Mike, how did you know you had enough to defend, given that this was all very much a work in progress?
BILLEN: Because it was more realistic than anything anybody had done at the time. And we used more different kinds of observations than anybody had used before to basically say, "We're not just looking at gravity and topography, but also the stress state and the flow of the mantle." We really tried to use as many observations as we could to constrain the things that we don't know. And that's actually one of the biggest things I learned from Mike, that if you're going to put something in your model, what is the observation that tells you it's a reasonable choice? Because we're always making choices. It turns out, even with all the rock-mechanics experiments that have been made, if I go to a paper and ask, "Here's how olivine deforms," the error bars on their data will translate into orders of magnitude error bars in the material properties in my models. I always have to bring in other constraints to decide, "Actually, I need to use this value, not this other value." That means it's reasonable, but I always wonder with error bars that big, "What am I missing? What could I be missing that I'm blind to?" That's always in the back of my mind.
ZIERLER: A social and cultural question. I've been very privileged to talk to all sorts of the pioneer women graduate students of the Seismo Lab from the 70s and 80s, some of whom experienced troubles, just as a result of being a woman. Did you experience any of that personally? Was your sense that those troubled times were behind the Seismo Lab at that point?
BILLEN: I didn't experience those. I was very lucky that I don't feel like I experienced that directly. There were a couple of incidents that still stick in my mind that were unfortunate, but they never felt to me to be a barrier to my success. And I definitely felt like in the Seismo Lab, the mentors I had were fully supportive of my success, and I felt very, very fortunate for that. I also felt fortunate for the group I came in with. The year I came in, there were quite a few geophysics students who came in together, and several of us were women. I think that just gave me a sense that I belonged. It really felt different. I remember that because I had to take continuum mechanics in the engineering department, and there were only two women in a class of 30. And there, it felt weird. It felt like maybe I wasn't supposed to be there. [Laugh] Not having that feeling in my own department was really important. But I do feel like the students I came in with who really gave a lot of support were key to my success there.
ZIERLER: One aspect I talked with Emily about was that this was among the first, if not the first, academic generations in seismology to really have the full power of the internet at their fingertips. I wonder if you can speak to that.
BILLEN: That's interesting. I'm not sure I really took advantage of the internet so much as just the computer. We used the internet a bit to find papers and things, but at the time, we were still getting data and using references and things from paper. Read a paper, follow up the references. The data I got was usually because I knew who to contact to get it. There weren't websites where we could just go and download data. NOAA was just working on creating these data repositories for all of the bathymetry gravity data we were gathering. I think, for me, the internet wasn't quite as big of a factor, although I did learn how to program HTML and make my own web page. [Laugh] But I don't remember that being a big factor.
ZIERLER: Where was the data coming from if not the internet, at that point?
BILLEN: You would get CDs with the data on it. [Laugh] That was also a big change. We cleaned up our office once, and we cleaned up this top shelf that had tapes and card decks. And I'd never seen a card deck before. When we'd make an input file for our programs, we'd call them input decks because there used to be a deck of cards that was fed into a machine. I came after that generation, but there were still these archives from the group before us. But we used CDs. I have a pile of CDs in the office, which is my thesis. [Laugh] There's no way that would even work now. The models are so big, they're, like, terabytes of data for a model. It wouldn't work that way anymore.
ZIERLER: Besides Mike, who else was on your committee?
BILLEN: Joann Stock and Mark Simons. Don Anderson was one of my advisors, but I think Mark Simons ended up being the other advisor on my committee in the end.
ZIERLER: Any excitement or interesting questions from the defense?
BILLEN: There was one question, but I can't remember the person's name. He was a visiting geologist from Turkey, and he asked me about this other subduction zone. I don't remember exactly what I said, but it was more that he was sharing his excitement for subduction and my results than anything. Whatever I said, everybody ended up laughing afterwards. That was the main thing I remember from my defense, just excitement for my research that people were happy to hear.
ZIERLER: Knowing that the modeling was incomplete, is it fair to say that the research was successful because to date, it was the least bad of the available models?
BILLEN: Yes. It's funny because doing geophysics is a hard process. We have a lot of questions, a lot of understanding from the field, but we're limited because different sub-disciplines are advancing at different rates, and we're all trying to move forward and answer questions when we don't have all of the data we really need. I would love to have all of the experimental data on the minerals in the mantle that I need to build my model. I know what I should put in my model, but I don't have the data. But it seems ridiculous to just say, "We're just not going to ask how plates move until we get all the data." You have to move forward and try to answer those big questions with the data you have in hand, knowing they might be imperfect, and knowing that sometimes you might have the answer wrong. But I think most of the time, we learn fundamental things that we build on. We still understand that there is a low-viscosity region in the mantle wedge that's very important for decoupling the subducting plate from the overriding plate. It's probably not due to water, which was the hypothesis in that thesis. It's actually probably due to stress-dependent rheology, which is much more important than the water effect. But the hypothesis that the low-viscosity wedge was key was not wrong, and that still persists.
ZIERLER: When it was time to start thinking about post-docs, was part of the decision-making where you could go to improve upon the models you knew needed improving?
BILLEN: No, actually, it was quite different. At the time, everybody told me I should go do a post-doc in something different than what I was already doing. I was trying to find something else that would expand my research and my understanding in a different direction. I looked at a couple different opportunities, and I ended up in England on what seemed like an interesting question at the time. We were looking at lithosphere instabilities, and it was a big question of interest at the time in terms of how they form and where they occur. I ended up going to England and modeling a process that was happening underneath the San Gabriel Mountains in Southern California. [Laugh] It was kind of bizarre. But I realized when I was there that I had a lot of unanswered questions about subduction that I was really interested in. When I was at Caltech, Greg Hirth had visited, he is a rock mechanicist, and I took a class from him and was very excited about the connection between the material processes of deformation, how rocks deform, and how that determines how plate tectonics plays out on the earth. I was lucky enough to get a second post-doc to work with him, and that brought me back down the path I've basically been on ever since. It's really how we link how rocks actually form, what we see in the field in terms of how rocks deform, to larger-scale processes through these numerical models.
ZIERLER: When you were at Leeds, did you spend time in Belgium? Did you explore your family roots while you were there at all?
BILLEN: I did a little bit. It was an opportunity to visit some of my family. I don't have a lot of family left in Belgium, but I did visit there. My mom actually came to Belgium so I could visit with her in Belgium. It's kind of a hard story. My father died before I was able to go back, so I never met him. But we were able to go on a trip there where we visited where he was buried and visited the family history in that way. But my French is not very good, so it's hard to get along and expand more than that.
ZIERLER: At Woods Hole, did that naturally get you more involved in oceanography and those kinds of issues?
BILLEN: I think it could have, but the direction that really got me excited was the rock mechanics side of things, trying to figure out how to build in the complexity that Greg was seeing in the lab in order to understand how subduction was working. That was really what we made our focus. That was a lot to do for the time I had there. I was already hired at UC Davis before I started the post-doc at Woods Hole. I was only there, I think, from October through July, then I had to come start my position.
ZIERLER: Institutionally, what is Woods Hole like?
BILLEN: I don't know what it's like now because I haven't been back in a while, but when I was there, it was just a very dynamic, collaborative institution to work at. I worked with Peter Kelemen and Greg Hirth, and several of their students, and one of the things I remember the most there is the geochemistry seminars that would go on for a couple hours. Somebody would come in and give a talk, and there were so many questions that they wouldn't finish their talk until two hours later. Just to see that kind of excitement to dig into a question and the willingness of the participants to answer and engage in that discussion was really motivational. It made me want to answer more questions.
ZIERLER: Given how fast you were hired, were you on the market? Were you being recruited already at that point?
BILLEN: Yeah, I interviewed before I finished at Caltech. I interviewed a few places my last year before I finished, then the first year when I was at Leeds, I interviewed at three other places, and that's when I was hired at UC Davis. I think it was a time in which there were quite a few positions open, and there was a big demand for geodynamicists. People were starting to really want to answer the question of how plate tectonics worked, how the inner-workings of mantle convection linked up to the surface observations we see. There were a lot of places looking at geodynamics at the time.
ZIERLER: Again, as you mentioned, these are foundational questions that were not really new at the time, they were just unanswered.
BILLEN: Unanswered, yeah. And people saw that the numerical capabilities were really growing, and that that was going to allow there to be a lot of questions answered that we hadn't done before.
ZIERLER: Did you build up your research group right way? Or were you more focused on teaching at that point?
BILLEN: I tried to do both. I ended up only ever having one or two students until very recently, and mainly, that was because the process of making models was so difficult. [Laugh] It would take a lot of time to get a new model to do what we wanted it to do. If you look back at my publications, I tend to answer a question and move onto a different question, which means that every model I make, I have to make from scratch. It's not like I take a model, tweak it, and do another paper. Each student I had, we were doing something I'd never done before. [Laugh] We were making a different kind of model, trying to build something we hadn't before. It was always a process of getting it to work, really. There's sort of an art of numerical modeling, trying to figure out exactly why the iterations or convergence weren't working. Tiny things in how you set up your model would make a big difference. That was when I was using CITCom, and it just took a lot of time. I usually had one or two students at a time with teaching, and I've also been very active in managing our graduate program since I've been here. But now that I work with ASPECT, I have three students, a post-doc, and two undergrads. It's just a code that's easier for people to use, and it will work. There's less art involved in getting it to be stable. [Laugh] And there's a lot more infrastructure. There are cookbooks, manuals, a community; a lot of ways for students to get started at different levels. Now, I have a bigger group, and it's the first time I've had that.
ZIERLER: When it was time to come up for tenure, and it was an opportunity to think about your contributions, what was the case you were making at that point?
BILLEN: At that time, there was a research project we'd been doing with Margarete Jadamec, who's now a professor at the University of Buffalo, in which we'd worked incredibly hard to make a very realistic three-dimensional model of Alaska that coupled the mantle deformation to the surface deformation and was really trying to understand how the subduction at the boundary of Southern Alaska translated into the deformation across the Denali Fault and uplift of Mount McKinley 500 kilometers from the boundary. Basically, the argument I was making was, as you add in more realism to these models, you can actually start to very clearly link to the surface processes we see. We could explain why the uplift was happening, where it's happening in Denali, we could explain why you actually have deformation across the Denali where you have it, and that it's all being driven by the subducting plate underneath. Then, in the same model, we were able to make the connection between the flow around the edge of the slab and seismic observations, which actually allowed us to constrain the shape of the slab, which wasn't constrained from seismicity. It was the first time we really could say from the base of the upper mantle all the way through the surface deformation, we had models that could self-consistently predict how the plates are deforming and why they're deforming that way to a scale of individual faults, which hadn't really been done before.
ZIERLER: Tell me about the UC Davis Advanced Mentorship program, the networking initiative.
BILLEN: Well, there's more than one piece there. The Advance mentoring was for the leadership for new faculty that we started. That was through an NSF Advance grant I became a participant on, basically a group of people who wanted to help faculty be more successful in their first year on campus. The idea was that you needed to connect new faculty not only to mentors within their department but mentors outside of the department, who would help them in getting their labs established, give them advice on how to mentor students, how to balance the different demands of teaching, service and research in their first year. I spent quite a bit of time sitting on these committees and linking other faculty. I sat on, like, 10 or 15 committees in one year, just in math and physical sciences. It was actually really fascinating because you learn a lot about how each of the departments work, and you also learn that people come to assume there's a certain way of doing things. And each department's doing it differently. When you have people coming in from different departments, you get lots of different solutions to problems. It was very eye-opening to see how entrenched we get with thinking the solution we have is the right way of doing things. Even things like how to mentor, how to teach, how to run a lab. When you have people coming from different disciplines and different approaches, you get a lot more ideas on the right way to solve the problem.
ZIERLER: Bringing our story closer to the present, as your research group has grown, what about your peers in the field? How has the community in this subfield grown over the years as well?
BILLEN: It's grown a lot. It's grown in different ways. The open-source software I referred to several times really created a new community for doing subduction dynamics and numerical modeling. There's a very strong group of scientists in Europe that also focuses on subduction modeling. I think what's really changed is how much we actually inter-collaborate between those two groups. Before, each person had their own code and knew how it worked, and it was a lot to manage, whereas now, we're constantly developing together, building on each person's advances, which allows us to do a lot more science than I think we were able to do before. For example, I'm building up these different models that use this visco-plastic elastic rheology. This is a module that was built up in ASPECT by five or six different people and is still being developed. We keep just adding things we want that other people realize they can use, too.
I think the biggest difference is, by people collaborating and sharing the methods and tools inside a piece of software like this, it gives us more time to actually answer scientific questions rather than just developing the numerical methods, then redeveloping it. Because each person had to do it themselves before. It's been a real change in the approach. Before, in my generation, we had to know the numerical modeling, we had to be able to develop ourselves, code ourselves, then also be an expert in the geophysics, whereas now, we're building on libraries that are developed by the mathematical community. That allows us to spend much more of our time focused on the geophysical questions and less on the numerics. Which for me, I love. I would much rather be in the geophysics side. [Laugh] There are other people whose strengths are in the numerics, and I love working with them. It makes for a nice synergy between people's individual passions in their science.
ZIERLER: Now that you've built your research group up, what are some of the things your graduate students and post-docs are doing that might suggest where the field is headed long term?
BILLEN: One of my students is working on this deep earthquakes project that I spoke about before. Linking long-term tectonics to hazard and seismicity is definitely a direction the field is going to be going in. Another is, my post-doc, Menno Fraters, developed a code called the Geodynamic World Builder, which is this amazing piece of software to build realistic subduction zones using datasets like the Slab 2.0 dataset for locations of slabs, the age data, fault data, and being able to automatically generate a three-dimensional fault. It may not seem like a huge thing, but Margarete Jadamec probably spent three out of the six years of her PhD trying to make an Alaskan subduction zone that was good. And now, he's developed a code that automates a lot of this process, and he's made it open source so that other people can contribute to it. I've contributed to it, one of my other students has contributed to it.
That, I think, is really key because we have to move from generic examples to specific examples, and having a tool that allows anybody in the community to build a realistic model of the Japanese subduction zone or the Cascadia subduction zone will allow people who have interesting questions to answer those questions and not be limited by whether they can build that model from scratch or not. I think that's really a key difference. It's kind of interesting because it means we have to change what we expect from our students. It's not as important anymore that they know how to solve a finite element problem. What they need to know is how to bring in the data to these codes, how to do the comparison of the numerics to large datasets. That's really where the difference is now.
ZIERLER: Is this more a matter of computational skills, analytical skills, or some combination?
BILLEN: It's still a lot of computing, but it's really data analysis and extraction. My students don't spend their time developing the numerical code anymore, but they end up with a huge amount of data, and they have to figure out the best way to compare it to observations, which observations they compare to that will allow them to learn something new, how to bring in the seismology, earthquake rupture, or geology part. It really, I think, allows us to spend more time on the geophysics questions rather than the numerical questions. It's still computing because you have to deal with a lot of data, how to analyze it, how to figure out the right comparison to observations, how to quantify whether it's a good fit or not, all of those things.
ZIERLER: Extracting gems from an enormous amount of data makes me think, where is AI in this, now or at some point in the future?
BILLEN: It probably will be some time in the future. I haven't quite seen where that is yet. I actually just recently took a first class on machine learning, and I think one of the main points they made in that is that machine learning in some ways is a giant interpolator. You give it lots and lots of examples, it learns from those examples, and then you let it operate on a dataset. Then, it can interpolate very, very quickly within the space that you have taught it. In terms of mining from the data, I'm not sure it's going to be on my end as a geodynamicist to be doing this as much as the seismologist who uses AI to mine out the gems of observations, which then I can point to and say, "A-ha. Let me model that. Let me actually see what the specific physical conditions are necessary to cause that." I'm not sure I'll be doing the AI on my end, but I'll definitely benefit from people extracting that information from the large seismic datasets.
ZIERLER: Now that we've worked up to the present, for the last part of our talk, a few retrospective questions, then we'll end looking to the future. To go back to the very beginning of our talk, you contextualized your current research sort of as a break from what you did earlier, yet what I heard as you narrated your thesis research was some continuity there. I wonder if I could get you to focus a little more specifically on where you see the intellectual continuity from those original numerical models and where, scientifically and conceptually, this really is a new branch for you relative to what you were doing at Caltech.
BILLEN: It's maybe a subtle difference. It's true, I've been interested in deep earthquakes since I was at Caltech. They're this enigmatic thing that fascinates geophysicists because there's so much about them we don't understand. It's really a question of time scale. That's where the break is. My focus has always been on how to understand the material properties that drive subduction and how that factors into how plate tectonics works at the surface of the earth. That's the question I've asked, and I've gone over it in subduction zones from the point where they sink into the mantle to the transition zone. And really, the break is that I was kind of getting bored with the questions I was asking because I felt like I had explored all the observations that were available to me at the time scale I was looking at. There were these uncertainties that couldn't quite be worked out within that time scale.
For me, the transition was realizing that if I added in elasticity, I could come forward in terms of time to these much shorter time scales and perhaps actually make an observation that'd allow me to understand what was happening at the longer time scales. That was really the change in thinking for me, which all of a sudden, opened up a lot of new ideas. I could actually look at how these large-scale subduction processes lead to buildup of stresses across the subduction interface at the surface or in the deep slab. I'm starting with the deep earthquakes because that's what's really exciting right now, but the more I learn about it, I realize I can actually look at this even in the shallow subduction zone as well. There is this continuity where I've really spent my career fascinated by how subduction works at the detail level, and now, I'm just bringing it to a shorter and shorter time scale.
ZIERLER: In this project researching the history of the Seismo Lab, I've come to appreciate a real theme of the history of the field is that there are orthodoxies that everybody believes are certain to be true, then, of course, they are demolished a generation or two later. Have you seen, in your own research career, that cycle of things that you learned initially that were definitely true that have now been proven not true or at least are on a much shakier foundation than you were made to understand initially?
BILLEN: Yeah. I think for some of those, I was right on the cusp. When I started, we were still arguing about whether there was whole-mantle convection or layered-mantle convection. That was just becoming cleared up as I came in. I think in my own field, one of the big questions is, what controls the motion of a trench at the surface? The place where a slab sinks is called the trench. There are people who really believe that they always go backwards, that's the direction they want to be going in. And there's a big uncertainty in this because it turns out we don't really know in what reference frame to measure plate motion: whether they're going forwards or backwards depends on the reference frame. What I've found in my models is that this is really strongly dependent on how the slabs actually deform. People will still make assumptions that slabs are these very weak, taffy-like things that just sort of flop into the mantle. The problem with that has always been, if they're taffy-like, they shouldn't have earthquakes in them. They have to be strong if they're going to store energy and have an earthquake. I think the thing that's really coming apart is this idea that slabs are weak. We're moving to a place where slabs are strong, but they're weak in certain places, like where they're actively deforming. And that means they behave in a very different way. They won't always roll backwards. You won't always have trench rollback. You can actually have a much more dynamic and interesting subduction history that you may not have recognized in the geologic record before. That's a big change. Strong slabs, not weak slabs.
ZIERLER: Finally, last question, looking to the future. You mentioned a research agenda you can envision out 10 to 15 years. Best case scenario, what will we know then that we don't know now?
BILLEN: Best case scenario is we know how deep earthquakes happen, what the mechanisms are, would be my hope. That would be fantastic. Then, also that we can actually calculate what the heterogeneous stress state is on active subduction zones and couple that to earthquake rupture and hazard predictions. I think both of those things are possible.
ZIERLER: You're aiming big.
BILLEN: Yeah. But we have so many pieces right now that are moving forward together that I can already see how some of this would be possible in the next couple years.
ZIERLER: To go for such fundamental questions, was there an approach to science that you learned at Caltech at the Seismo Lab that encouraged you to think along those lines to aim so high?
BILLEN: There are examples everywhere. [Laugh] Maybe it wasn't explicitly said, but it's just how you saw people approaching problems. I remember Mike Gurnis telling me when we first got some of our results out, "Once this is out there, other people will want to do the same thing." But then, he would move on. I think that's just kind of this ethos of asking the big questions, getting that first-order result, and that's where the excitement is. Sometimes there's some cleaning up you have to do, "Did that really work or not?" But the exciting science is asking those hard questions, figuring out if you're right or not. And I'm definitely somebody who gets bored. [Laugh] I don't like to redo things. As soon as I can kind of think my way through and know what the answer to a model is going to be, I don't want to do it anymore. It's not worth it to me to run the model. [Laugh] I tell this to my students, also. They'll come to me and be like, "This model didn't do what we expected at all," and I'm like, "Great. Now, we have an opportunity to learn something." If it does what we expect, there's nothing for me to learn. That's what keeps me excited about my science.
ZIERLER: On that note, this has been a phenomenal conversation. I'm so glad we connected to capture your recollections and history of your career. Thank you so much.
BILLEN: Thank you.