Join us for our capstone episode on the Animal Intelligence season. We recap what we loved, what we learned, and things we wish we had gotten to spend more time on. This is a great episode to see how the podcast is produced. Now that the season is ending, our current co-host, Becky, is moving to emeritus status. In this last installment we got to spend a little more time getting to know Becky and where her work will take her after this. Did Data Skeptic inspire her to learn more about machine learning? Tune in and find out.
Data Skeptic
Animal Intelligence Final Exam
(upbeat music) - Welcome to another installment, or perhaps the last official installment of Data Skeptic Animal Intelligence. I guess we can always do cameos and revisit things, but we're gonna be switching topics soon, which is a bittersweet thing. We have to bid a little farewell to you, Becky. Thank you so much for all your contributions this season. - You're welcome. It's been really fun and I've learned a ton. And you have helped so much in ways I think listeners will know obviously from being on the show and doing an interview and help setting the stage, but behind the scenes you've been a tremendous help. This has been a really busy time for me personally and without you we wouldn't have had this season. - I really appreciate that. Yeah, it's been pretty fun. I mean, just leveraging skills as a graduate student at this point looking through the literature and finding good people to contact and good stories. That's been awesome. - I guess there's nothing too surprising about what it takes to make a podcast and I think listeners could guess, but has there been anything surprising for you from the time we started chatting to finally producing shows together? - Yeah, there's a couple of things that surprised me. So I'm used to reaching out to other scientists pretty frequently for work or in my past life working on museum exhibits. We would get subject matter experts that were just kind of assigned to us depending on the client. So just this year kind of ratio of contacts to guests that actually come on was a little surprising to me 'cause I don't know exactly what our ratio was but definitely we had to contact a lot of people. So that was really surprising. And then working through raw audio, I got a chance to do that. That was really interesting for me too 'cause we do these interviews online. So basically how things get compressed and how the recording software works so it's interesting to me to figure that out and how you avoid pitfalls and make everything look really nice. After doing that and revisiting episodes, I can really see where your editing is coming in to make things sound just really great. - There's a lot more editing than most listeners would think or at least hopefully 'cause editing should be seamless. It should never sound like an edit but one of the exciting parts about doing data skeptic is I'm talking to quote-unquote real people. You know, I don't want your PR person who has a couple of bullet points. I don't want a politician who is gonna give a stock answer. I want, you know, a person who's like, mostly in the weeds doing the real work, AKA not someone who is the most polished public speaker. And we cut ums, we cut coughing noises and every once in a while we cut a question if I ask something dumb or tighten up an answer a little bit. It's kind of like being an editor, I think, to a certain extent. I don't think I do many editorial choices but I will clip a long-winded hem and ha before someone gets to their real answer sometimes. - And that definitely shows through there's not a lot of ums. There's not even really loud breath sounds between pauses. Like there's a lot of work that goes into that. And I think that makes actually a huge difference for making the podcast friendly to non-native English speakers. They can trust you that they're gonna sound really good and you're very good at giving them the time they need to actually formulate answers to the questions. And I think that means you get to tell cooler scientific stories because the guest pool can be really broad. - Do you have any noteworthy guest topics or insights that you took away from our season? - Yeah, I was looking back through the season just to answer this question. And I had a lot that I really liked. One of my big favorites was advances in data loggers with Ryan Hanscom who's another grad student. That was using things like accelerometers to be able to make training data to look at animals that are really hard to observe 'cause they're nocturnal. We've underground all those sorts of things and actually get real behaviors from those data by essentially getting the behaviors you want in the lab and then comparing that to wild animals. That to me was like mind blown. Didn't know we could do that. Amazing. - One of the highlights for me has been, I guess, the ways in which advancements in computer vision have rolled out into the real hard sciences. It's easy to go online and see some demo that can recognize a chair or a beach ball or that kind of stuff. Everyone knows that solved. But recognize the difference between an eastern swallow and a western swallow is this customized thing. And we've heard from a lot of angles about how that integration has been seamless or difficult and tooling along the way. I hope we find that people will look back on that almost as a quaint part of this, that it will just become ubiquitous to researchers in the future as the tooling gets easier. - That would be exciting to revisit this in 20 years and see what we think. - Let's mark the calendar, yeah. Before we started chatting, I think maybe we touched on this earlier in different discussions. But could you remind listeners, what was your understanding or exposure to machine learning? - It was pretty minimal. So I'd had a couple of classes in systematics and then bioinformatics where we went over things like hidden Markov models to look for patterns in genetic sequence data. 'Cause that to a human, that's just gonna look like a bunch of big nonsense and you're looking for specific kinds of things. In systematics, you're often doing modifications of some kind of random walks to simulate things. So that was really my exposure is just knowing the definition of a few tools. - Makes sense though, they aren't, I mean, they're present certainly in your field, but it's not the focal point per se. One can have a rich and rewarding career without learning any of these underlying algorithms and technologies. Or at least that was historically true. Do you think that's gonna continue to be true into the future? - I don't. Maybe for our age cohort, it'll be fine. But I think for folks generation after us, it's probably not gonna be fine. So what I noticed as we were inviting scientists on and looking through literature is in biology, you often have a bunch of specialists, whether that's physiology, ecology, whatever you want. And if they're doing anything with machine learning, they have like a machine learning person that's on the author list. So they have like their machine learning buddy. And I think as we move forward and machine learning tools get a little bit more user friendly. So I'm just sort of picking the kind of tool I want as opposed to designing it myself. I think it's gonna be a lot more central to what people are doing. - Open source is sort of an obvious way to go. Everything's done in public. People can see it, they can make improvements, they can fork it if they don't like the way it's being maintained. There's a lot of strengths there, but you also get a lot of rough edges with open source solutions. I guess do you have a sense of if people have perspectives on that? Is there an ick factor with using some commercial solution? Like maybe it's gonna go away or they're gonna jack up the price? How do you find people you talk to look at it? - So I've seen both. If you are at an institution, say like a zoo or a private company, it's easier to get like a out of the box solution that you're happy to pay these like enterprise renewal fees for the most part. And people don't seem to be really afraid to do that. Where I see more trouble is in adopting things that are open source with no GUI. And I see a lot of other grad students struggling with that. I was just talking with someone a few months ago at University of Missouri. We have a fancy compute cluster called Hellbender. It's named after salamander that's endangered in the state. The Hellbender cluster, you gotta know some Python and you really need to get trained to be able to use this. And it's a really great tool for us 'cause there's so much data that you can't really even run on a good gaming machine anymore that you'd have to leave it running like all weekend or week. And this student and I were both basically saying, cool, we have access to this cluster. We even loaded up some instances, but I can't get my Python packages to work for. I was using sleep, which one of our guests talked about, Talmo. So I think that adoption for folks is really hard. If you have something where the data and what you're using is just connecting to R and you're writing a little code and it's a black box, I think people can do that. But I think without Gooey's, a lot of grad students that are sort of at my level that don't have a lot of computational training are probably not gonna adopt very quickly. - Well, the hurdle and an opportunity in different ways. - Yeah, I think so. But I also think if we don't get our act together in some way, we're gonna be less productive researchers and in a super competitive job environment, that's kind of a make or break prospect. - And you've touched on this a little bit, but do you have thoughts on how machine learning will be part of your work personally in the future? - Yeah, so one of the biggest things, and this came through in like three or four episodes this season is I have to go through a lot of video data of animal behavior. And right now I train undergrads to do it and they spend hours and hours and hours going through video footage and manually label everything. And they're using an open source software to do that. They're not like by hand, putting things in Excel, most of the time. I think that is where machine learning could really help me is to train on pose estimation and then use that to go back in and label behaviors based on the pose estimation information. I think for me, that's gonna be the biggest tool. Also some of the, like we were talking about the data loggers, but then also talking about time series data with like GPS collars. I think those are also some avenues I could run into, but for now it's the videos of animal behavior. There's an area in, I guess it's both in code, but in data and in science in general of reproducibility. So we have some tools, things like a Jupyter Notebook, which is a pretty good documentation about what you did. I would call it, it has the spirit of reproducibility without necessarily guaranteeing reproducibility. It's a hard problem. There's other approaches with like Docker or shared data stores or GitHub. Where do you feel that you and your colleagues are at in the maturity lifecycle where I could kind of push button to reproduce somebody's experiment? - Not very close, I'd say at this point. We definitely can write down our methods, make all our data public, include really good metadata, keep really good lab notebooks so we can go back and reference things. But in terms of like a one click, I'm gonna load it up. I think right now that's mostly feasible for statistical analyses where I can say, here's my CSV file, here's my art code, have at it. But for other functions, we're just not there yet. - Yeah, and some of that I think is a technology problem, but an interesting one to be working on. - With everything that's getting retracted lately, I feel like it's more and more of a problem that's in the forefront of people's minds. You know, it'd be horrible to have to issue a retraction for a silly mistake. - This episode is brought to you by WorkOS. If you're building a B2B SaaS app, at some point your customers will start asking for enterprise features. Like single sign-on, skim provisioning, role-based access control, and audit trails. That's where WorkOS comes in. With easy-to-use and flexible APIs that help you ship enterprise features on day one without slowing down for your core product development. Today, some of the hottest startups in the world are already powered by WorkOS, including ones you probably know, like Perplexity, Vercel, Jasper, and Webflow. WorkOS also provides a generous free tier of up to one million monthly active users for its user management solution, making it the perfect authentication and authorization solution for growing companies. It comes standard with rich features like social logins, bot protection, MFA, roles and permissions, and more. If you're currently looking to build SSO for your first enterprise customer, you should consider using WorkOS. Integrate in minutes and start shipping enterprise plans today. Check it out at WorkOS.com. That's WorkOS.com. Continuing education is, I think, a big part of why a lot of people come the day to skeptic. Maybe it's one of my top three reasons for why I do the show. I do have thoughts on how your own continuing education will go into the future. I guess particularly around literature in the field. - Yeah, so in terms of literature, I'll definitely keep reading. Like, being on the show had me branch out of my comfort zone looking at literature I don't normally look at. And I want to keep doing that because that gave me lots of good ideas for my work. In terms of immediate continuing education, I was very lucky to get a small scholarship from the Jackson Laboratory and then funding at University of Missouri, St. Louis, from the Whitney Harris World Ecology Center to go actually take a workshop in a couple of weeks for machine learning, for automated quantification of behavior. And that's in Maine. So I'm really excited about that and I'm hoping that'll help kickstart me to integrate more machine learning into my work. - Along similar lines, I mean, you've been doing a lot of science communication through your role here and I think it's some other things as well. What do you see as the role of science communication in your career going forward? - I think it'll always be part of it. I mean, that's how I started my first job after my masters was I was a museum educator and then I worked at an exhibit design firm and we worked with zoos, museums, aquariums, private companies, everything to make multimedia based exhibits for the most part. And that was a lot of fun 'cause we were often solving problems from scratch. But I think science communication is always gonna be part of what I do. And for folks that heard me talk much earlier in the season, I work a lot with invertebrates, specifically bumblebees and tarantulas. And I think there's some space there for a lot more communication about invertebrate behavior and conservation. I don't know, maybe I'll start a podcast. But I think it'll always be a part of what I do, even if that's just local outreach and volunteering. But I also like multimedia and web and all that kind of stuff. So I'm not sure, but it'll definitely be a part of what I'm doing. - I'm curious if you have an opinion about anything we missed in the season in our coverage or any areas we're weak in our emphasis. There's always kind of a bias to who I find, who will agree to come on the show and all that kind of stuff. But being that it's more your field than mine, what did we emphasize too much, too little or miss? - Oh gosh, in terms of a critique, I don't, well, I have a bias too 'cause I like helped make it, right? - Sure. - I think we actually did a pretty good job between us of getting pretty good coverage. It seems like one thing you and I talked about a little bit is we really wanted some more genetics, algorithm type stuff. And when you and I got to talking about it, it's like, oh yeah, I know what you're talking about from an evolutionary standpoint, but not a computer science standpoint. And it seemed like we had a little trouble finding just the right guess to talk about that, which really surprised me. - So me too, but I have a new theory or a new belief. First, ant colony optimization is not a real thing. No one actually uses it. It's only studied in classrooms. Secondly, I think that might be true of genetics algorithm as well. There are some limited use cases I've seen in industry, but for the most part, it's like gradient tree boosting algorithms like XG Boost or it's deep learning techniques. SVMs a little bit, but even those are getting exotic and have some computational scalability challenges. So I think my theory as to why we couldn't find genetics algorithms people is because it isn't really a contemporary subject that people are publishing on or doing novel things with. I'm sure there's students doing projects or whatever, but I haven't seen it cause a breakthrough, but I'd love to be proven wrong. - Yeah, that makes sense to me. And that was the feeling I got like you were talking about ant colonies is there's lots of algorithms being used to solve problems that are novel, that are inspired by biological systems, but they're not really used on those systems just like you were saying. So I think you're right about that. In terms of other things we could have covered, like I said, I think we did a pretty good job. Like there's a smattering of episodes that are just about specific species in their cognition, there's conservation stuff, there's evolution stuff, there's how do we model a nervous system or a whole organism? It was really broad. So there's something in there for everybody. One thing I was worried about sort of mid-season, it was important to me. I was like, well, I'm an invertebrate person and I don't want this to be all invertebrates, but we still ended up with a lot of invertebrates, especially bumblebees, bees in general, honeybees dude kept coming up. So I think we got real bug heavy in spite of my best efforts. - Is there something about those that make them ideal species? I know fruit flies have a lot of reasons why they get studied. Could it be along similar lines? - I think so. They're just really nice model organisms. And in the United States, you don't have to go through all the welfare approvals if you're going through, if you're using invertebrates. I don't think that's the reason why all our guests were studying invertebrates. If you listen to the episodes, you can tell that's not why. But they do end up being really good models. So I think you're right about that. - Do you think that'll change? I heard a recent, I think it was on Qantas. They said some, I don't know if it was like an open letter or some publication, but a group of scientists, well-known people are now formulated a new definition of consciousness that is, I wish I could tell you the finer points of it off the top of my head. - So I think what you're talking about might be the New York declaration on consciousness. - I believe you're right. - Yeah, so recently I went to a conference at New York and it was all about wild animal welfare, which is like a brand new science. So there are philosopher scientists, things like that. But I believe that group at NYU had something to do with like kickstarting the New York declaration. And yeah, I think there is a lot more interest in sentience, especially in animals that look really, really different, like invertebrates. There's some really interesting philosophers talking about that kind of thing. And for me, I'm really interested in the nitty gritty. So I study pollinators, plants make all kinds of secondary metabolites that are psychoactive chemicals. And I'm really interested, like, does the bee know when it's got caffeine? 'Cause to me, that has a lot to do with sentience and are you monitoring your internal state, right? So I'm really interested in those scientific nitty gritty questions as opposed to like, should we have policies on invertebrate welfare? But I think it's getting a lot more attention which animals are sentient, which are not. What does sentience entail? I think that's really catching a lot of steam. - Yeah, the philosophers have, and for good reason, struggled to give us firm measurable definitions of these things. That's probably intrinsic in what they are. But I'd love to see more about this gradient. I consider myself more conscious than the bee, but I too would love to know if they're aware they're hopped up on caffeine or not. - Yeah, and I think insects are funny, and this is just conjecture on my part because their nervous systems are so small. So I think if they're gonna be flexible about things, it's gonna be very specific things. And then other behaviors, you really need to save space in that nervous system and hardwire it and not maintain all that plasticity. So like, my bees are great at learning to solve little puzzles because flowers all have different ways that you have to access them. So they're really good at pulling things, manipulating things, watching other bees. But when they get out of the box in my lab and suddenly all the lighting is really different, they get lost and they get stuck up in the fluorescent lights and I have to catch them. So you see suddenly very unintelligent behavior that's based on their experience being raised in a lab, but I think there's just some things they're super good at and other things that they're not. And with people we're much better at a lot of different things. So yeah, I think that gradient is there. - I bring an AI bias to the table that if I can think of it as an agent where I can see it source code, I kind of regard it as non-conscious and like getting stuck in a loop or getting stuck in a light has that feel about it. Yet I've learned this season that they're surprising learners and a wide variety of contexts. Unlike maybe answer, not quite as good at learning might be more mechanical, but it doesn't take much, I guess, in the brain, maybe more than the sea elegance we learned about early on, which can now be modeled completely in a machine, but it's not too far off on whatever our scale is before you start to see some really emergent, surprising behaviors. - Yeah, I think a lot of people get surprised by that once they start digging into the invertebrate, just literature and science. So I see it as if you have sentience or what I say sentience, I'm usually meaning some sort of subjective affect. So like in computer terms, if you see all the code, does it have subjective affect? Does it feel anything? And that's actually not a scientific question, right? That's not falsifiable. - In a way, yeah, right. - So to me, it's a really fascinating question 'cause I think it's highly likely that you have an internal affect. - I hope so. - And I have some indirect ways to measure that, but it's not truly a falsifiable question. So this is where I just kind of get nerd out and I think this is super fascinating 'cause we can't really answer that question very well. Is that be an agent with a nervous system generating a computer program or does it have some feeling internal affect or not? Those are just super hard questions. And I really like thinking about that sort of thing. - And in terms of your own work and your career, what do you see yourself for? Maybe remind listeners what you're doing now and where that work is going? - Sure, I'm a PhD candidate at the University of Missouri, St. Louis, and I study tarantulas and bumblebees. With the bumblebees, I'm looking at just what I told you. Like, do they know when they're on drugs? How does that affect their foraging behavior? With tarantulas, I do welfare and conservation. So we're looking at heart rate and tarantulas as a measure of excitement, whether that's stress or happy they got food or anything like that. And that has got a surprising amount of pushback when I talk to other arachnologists or even veterinarians that work with invertebrates because their nervous systems are so different from ours. Like, there's not a sympathetic parasympathetic division. So like, you can't assume they have fight or flight. But for me, regardless of the nervous system architecture, if you are an animal that needs to move quickly, that you might be preyed upon, one of the best adaptations you can have is getting more oxygen to your muscles. If you have any kind of heart, even though their circulatory system is kind of open, if there's a threat, you're gonna need more O2. And we just have really preliminary data on that 'cause it takes a long time to train a spider and habituate them to taking their heart rate. But we do have some habituation curves where their heart rates start out really high and now are consistently really low. And that's what you would expect if an animal was habituating to what you were doing and heart was a good measure of stress. And we have lots of other analyses with that that we're doing. So that's a big thing that I'm working on. And then also tarantula conservation here in Missouri. The couple of populations that we study seem to have really low population densities. So we're curious why? 'Cause usually tarantula species in the United States are doing pretty good. If you go down to the Southwest or Texas, they're like, they're everywhere. So why do we have these low populations up here in Missouri? So interested in that. So long story, what am I gonna do when I'm done? And this is just sort of realities of the academic job market. Would I love to have my own lab and teach at a university? Yes, but I'm already 37. Can I afford to go post docking all over the world until I can get a job? Maybe not. So I'm looking at nonprofit organizations that do research. So for invertebrates, that might be like the Xerxes society working for a wildlife agency or even the federal government doing freelancing. I'm getting more confident as I get older that maybe I could just do that. So I'm really open. If there was a good teaching position or a postdoc that worked, I suppose I would do that, but right now I'm just super flexible. Maybe this is my bias, but I think we're as a society, as an economy, evolving the shape of work in a lot of cases. It need not necessarily be the traditional nine to five. Perhaps you could have four or five income streams and actually that the diversification could be a good thing in some sense as well. Absolutely. And I'm starting to think like, yeah, I think maybe I could do that. And this experience has helped me with my confidence actually working on data skeptic and doing essentially proper freelance. And I think I can always do that with science communication. That's something I have a good portfolio of. Definitely, yeah, I would love to see more for sure. When it comes to that low population you were mentioning, do you have any working hypotheses? Yeah, so I'll try to make this succinct. So in Missouri, we have the habitats where the tarantulas like to live are called glades. And those are native grasslands usually on hillsides. I'm gonna say southern slopes, but I see a lot of northern slopes and stuff as well that are between forests essentially. And in Missouri, these tend to be really dry. They're really xeric, so they get really warm. They have desert reptiles, they have desert adapted plants. And apparently, sometime I think it was towards the end of the Pleistocene, the Midwest actually got a little bit warmer for a few thousand years. And we think that allowed some of these southwestern or southern species to come up. And then it got cold and wet again. And what do we get? We got little glades that are now little islands. And then as Missouri was colonized, we started changing fire regimes. So these glades were maintained by fires, wildfires over time. And folks just stopped allowing those wildfires to happen. So these glades are getting overgrown by things like juniper trees and they're getting really shady. So if you're an ectothermic animal that's a desert adapted in central northern Missouri, you're going to start getting too cold as the habitat degrades. So I think we're just kind of on the edge of where they can possibly live in central Missouri. Next summer, we're going to start looking in southern Missouri to see if we see the same issues. We also get habitat fragmentation. So these glades are little islands that are not always connected. So you can get inbreeding, lack of diversity. And then the last thing that's been really hard to measure, there's an undergrad in our lab who's kind of working on this, is people collecting them for the pet trade. We don't know how big a factor that is, but these populations, they're so small. In some places, we're talking like 25 animals in a 50 acre area. They're not too hard to find. If someone were collecting just for a weekend, they could wipe out half of those animals or almost all of them. Yeah, I think it's probably mostly an environmental habitat fragmentation thing, but this collection question is kind of tough. - Sure. Or like most things, probably a mix of reasons, but perhaps you can figure out which are the most prominent ones. - Yeah, I suspect we could have found some folks that were doing interesting work modeling species rain shifts in response to climate change. So yeah, I bet you're right. We probably could have found some people doing that kind of work. Yeah, that's an issue. Now I want to go back and find somebody. - Yeah, maybe we'll do some follow ups, always room for that. And Becky, anything else you think we should have covered that we didn't get to? - No, not really. I'd say we could have hit the philosophy people a little bit harder, but there's only so much you can fit into a season. And I think we got some really interesting and cool things in there. So I don't really have any regrets or major things I think we should have got to that we didn't. - Same, we put out a couple of invites. There was one person particular, give their name if I remembered it off top of my head, but they had some interesting perspectives on like animal research and ethics. And their perspective was that we don't get a lot of value out of many of these animal tests and maybe we shouldn't be doing them. And that perhaps there's a more formal statistical way or utility theory based way of doing these assessments. And I think that's novel and interesting. I don't want to say I'm fully on his side, but I'd love to, would have liked to have had the chance to share some of those perspectives. Sort of a Peter Singer variety of things. - Yeah, and I think that would have been interesting. Yeah, just because of the kind of analyses and things that data skeptic focuses on. Yeah, I think that would have been interesting. As a researcher, I'm not quite ready to let go of animal research, especially in the biomedical field. But one of the principles of animal welfare is reducing the number of animals you need to get statistically robust results. So I bet that guy sort of had a lot of really interesting things to say about that. - I love those ideas. Some of it comes down to like, like you say, welfare. If you want to test, can we train dogs and you're going to treat them well and give them a good home? Well, do all the testing you want, I think. If you came to me and you said, we think we can cure a certain particular rare cancer, but we have to kill a thousand chimpanzees. I don't know if I'm on board. If you have to kill a thousand ants, go do it today. For whatever reason, I have that moral trade off in my mind. - Yeah, and I think it's something people also don't know a lot about how it works right now in sciences. That's one critique. When I was at that animal welfare conference, not everyone there was like this, there were a couple of people that basically thought a lot of us animal behavior researchers don't really care about our animals, that they're just a science experiment. And I've never worked in a lab where that was the case. Yeah, even if the animals were euthanized at the end. I have heard stories about bad actors. - As is true in everything. - But if someone, yeah, like animal behavior society, it's one of my favorite conferences. I feel like if people found out you were laughing at the pain of animals or mistreating your animals, you would be shunned at that conference. No one would wanna talk to you. For just your average listener, knowing more about how animal welfare and research is approached in the US and in Europe, they do some interesting things that we don't, would be fascinating. - Well, all stuff we might come back to do some follow-ups on then one day. - Sure. - Becky, thank you so much for not just today's recording, but all of your contributions over the season. It's been a real pleasure working with you. - It's been a pleasure working with you. Thank you so much for giving me the opportunity to contribute, this has been great.