Transcription
Rebekah Warren: This is 89.1 WEMU. For more than three decades, Issues of the Environment has brought you conversations about the environmental challenges and opportunities shaping our communities. I'm your host, Rebekah Warren. When you watch birds in your backyard or on a walk through a local park, you may notice far more than which species are present. You might see a blue jay chase another bird away from a feeder or notice several species feeding together. These small details, recorded by people who take the time to observe, can help scientists understand how wildlife interacts and how ecosystems change. A new study published this month in the Proceedings of the National Academy of Sciences grew out of research right here at the University of Michigan. It explores how artificial intelligence can help researchers work with millions of bird-watching comments that might otherwise be too numerous to analyze by hand. The important starting point, though, is human observation. People in the field see what happens, describe it in their own words and create the records that scientists can study. My guest today is Dr. Hengxing Zou, a quantitative ecologist and former postdoctoral researcher at the University of Michigan's Institute for Global Change Biology in the School of Environment and Sustainability. He led this research into how large language models can help unlock information about species interactions. Dr. Zou, welcome to Issues of the Environment!
Dr. Hengxing Zou: Thank you for having me. It's been a pleasure listening to this wonderful introduction that you gave me.
Rebekah Warren: All right, Dr. Zou. Let's start with the person, with the individual, just watching birds in their backyard or at a neighborhood park. They see something interesting and add a short comment to a platform like eBird. What can that observation tell a scientist? And how does it become part of a much larger picture of wildlife interactions?
Dr. Hengxing Zou: Yeah. So, basically, like you just covered, people notice all the things when they're watching birds. They might see things like, "Oh, this bird is having some kind of trouble flying," or, "This bird is being chased by a hawk in my backyard," and this kind of thing. And they would just notice it in the notes. They put these things in these places called comments, where a lot of these platforms will permit the use of these natural language comments to be attached onto those observations. And you would be amazed by how many information people actually put in these comments. One of my favorite comments would be, like, there are two birds that engaged in a mid-air collision, which sounds really strange and ridiculous, but it just happened. And you also get things that are more quote-unquote serious, like a hawk actually feeding on a bird and successfully capturing a bird. So, these things, they are all kind of natural observations that people make, and it kind of goes back to the origin of ecology as a field of study is that people look into nature, and they notice things. They notice where things are, they notice when flowers are showing up, when birds are showing up, they notice the kind of interactions between species of birds and other kind of environmental factors. So this is just basically how people are doing natural history notes in the current, visual age, if you will. And like the old natural history notes that were, from the beginning, the roots of ecology as a study, these natural history notes nowadays still carry a lot of very important information. And it is for us ecologists to do, to actually distill this information into something we can actually use.
Rebekah Warren: What kinds of details are most useful for observers to include? Is there a simple way that people can make notes clearer without needing to be a scientific expert?
Dr. Hengxing Zou: Yeah, absolutely. So, in terms of this study, where we looked at mostly species interactions, it would be pretty important for people to look into, like, the exact two species that are interacting, right? So, if you're observing birds, I would assume that you would have some kind of basic knowledge of identification. And those IDs can go into the comments, and that will make our lives so much easier, rather than saying something a bit more fuzzy, like, "Oh, there's a bird rather than a, for example, black-capped chickadee." So, the latter, if you can observe this species and make sure that it is actually the species that you're going to be able to identify and that's a species that was engaged in this kind of interaction, that would be awesome. And the other thing is, like, there are some species interactions that are a bit more obvious than the others. For example, if you see a hawk grabbing, like, a small bird, that's definitely like a predation, right? Versus if you see two birds fighting, they may be fighting for food. They may be fighting because one bird is trying to eat each other or what would the other. They might be just there and suddenly ran into each other. So those are more ambiguous. Like, we don't really know what exactly is happening behind the aggression of these species. In that case, if you could just say, "We observed aggression between these two birds," that would be great because it has been left for us to decide what exactly is the root cause of this aggression. If you are a birder and you are interested in contributing to this kind of research, then I think user language a bit more precisely would greatly help our process of identifying these species interactions and the species that are involved.
Rebekah Warren: You're listening to 89.1 WEMU. I'm Rebekah Warren, and this is Issues of the Environment. My guest today is quantitative ecologist, Dr. Hengxing Zou. We are discussing how human birdwatching and observations combined with carefully checked data analysis can help scientists study changing relationships among species and ecosystems. Dr. Zou, for this project, you and your fellow researchers describe a goal of building a biogeography of species interactions--a way to understand where and when relationships among species occur. How could that broader view help scientists recognize changes associated with shifting ranges, climate change or development from urban growth?
Dr. Hengxing Zou: Basically, let's go back to where species interactions are really collected in the past, right? So these are usually collected by experts that spend a lot of time in the field, basically observing how the species are associated with each other through either human observation or some kind of camera or some kind of acoustic recording devices, right? So these things are very limited in terms of their capacity because you simply cannot have the person sitting in the field 24/7 for all days in the year. And then a lot of these intensive efforts are also restrained to only a few sites of, for example, natural reserves or field stations. So, in the past, we really don't have a lot of those larger-scale data sets of species interactions that run through, for example, the entire United States or even the entire North and South America. So that would pose some kind of difficulty in us for understanding how the species interactions are changing both through time, over years or within days in a year or across different kinds of latitudes or across the different countries or even in different sites with different kind of degrees of urbanization.
Rebekah Warren: For listeners who might be cautious about the use of artificial intelligence, could you explain its role in this research? What does that process look like, and how do you check whether it's interpreted a note correctly?
Dr. Hengxing Zou: You know, we're using a large language model in its very original kind of design views, which is just to parse information from natural languages. So we're not asking it to generate anything else, like a lot of those more recent kind of generative models are, or like the uses of those models are for. We are just asking it distill information from what we just fed it from the input to us. So that will definitely reduce a lot of the, what we call, hallucinations or, like, the generation of materials that was not there in the kind of the input. And on top of that, the comments are also relatively short. As you would imagine, like, there might not be a lot of people writing, like, hundreds of words just for a single comment for a simple observation. So that will also help with decoding the information and the hallucination problem because it is just based on the text itself, rather than going into all kinds of imaginative space formed by the models. And another important factor is that after getting this information from the model itself, we would rigorously, manually check whether these are correct. And these have gone through manually by trained professionals, like myself and like other people on my team, that will make sure that the information is correct, they make sense and then there's not a lot of like weird or spurious kind of inferences coming from the comments themselves, which, honestly, I would say that it was pretty rare. We only see some kind of misidentification of one interaction into the other or when there should be an interaction and I find no interactions. So, in general, these are well within the margin of errors that we are expecting and the normal downsides of large-frame work model that people usually associated with being, which one of them being hallucinations, we find a less of an issue in our use cases.
Rebekah Warren: Thank you, Dr. Zou, for being with us today. For more information on our conversation, visit our website at WEMU.org. Issues of the Environment is produced in partnership with the Office of the Washtenaw County Water Resources Commissioner, and you hear it here every Wednesday. I'm Rebekah Warren, and this is your community NPR station, 89.1 WEMU, Ypsilanti.
About Dr. Hengxing Zou
"I am a global change ecologist interested in understanding how spatial and temporal changes of species distribution shape their interactions and community structure. I use a combination of ecological theory, data analysis, lab experiment and fieldwork to study species interactions under global change. Currently, my research is heavily focused on North American birds. Read more about my past and current projects here."
"I am currently a postdoctoral fellow working with TJ Clark-Wolf (Department of Wildland Resources, Utah State University), Erica Stuber (United States Geological Survey Utah Unit), and Alex Baecher (Department of Environmental Science, Emory University), funded through the USGS Climate Adaptation Program."
"Previously, I was a postdoctoral fellow at the Institute of Global Change Biology at the University of Michigan, working with Brian Weeks and Kai Zhu."
"I obtained my Ph.D. with Volker Rudolf at Rice University in Houston, Texas. Before moving to Rice, I graduated from the University of Chicago, where I worked with Cathy Pfister and Alison Anastasio on characterizing plant communities on urban contaminated soil."
"Before coming to the United States for college, I spent my first 18 years in Beijing, China."
Resources
University of Michigan School for Environment and Sustainability
University of Michigan Institute for Global Change Biology
New paper: "How nature lovers can aid ecology with an assist from AI"
Non-commercial, fact based reporting is made possible by your financial support. Make your donation to WEMU today to keep your community NPR station thriving.
Like 89.1 WEMU on Facebook and follow us on X (Twitter).
Contact WEMU News at 734.487.3363 or email us at studio@wemu.org.
