Dr. Melissa Johnston is a neurobiologist from New Zealand who investigates complex cognition in birds. Following the completion of her PhD at the University of Otago, Dr. Johnston was awarded a Humboldt Research Fellowship to continue her research in Germany at the Eberhard Karl University of Tübingen. Her research focuses on the interplay between the brain and behaviours such as working memory, timing, and probabilistic reasoning in a range of avian species, including pigeons, jackdaws, and carrion crows.
Episode Description
While Edgar Alan Poe reported that the raven quoth, “nevermore,” crows may respond with, “more likely” when forced to choose between two options. That’s according to our guest on this episode of Stats+Stories Dr. Melissa Johnston
+Full Transcript
John Bailer
Well, Edgar Allan Poe reported that the Raven quoth, “Nevermore”. Crows may respond with “more likely” when forced to choose between two options, at least according to our guest on today's episode of Stats and Stories, where we focus on recent work suggesting the crows demonstrate statistical inference skills. I'm John Bailer. Stats and Stories is a production of Miami University's departments of statistics and media, journalism and film, as well as the American Statistical Association. Rosemary Pennington is away. Our guest today is Dr. Melissa Johnston, a neurobiologist from New Zealand who investigates complex cognition in birds. Following the completion of her PhD at the University of Otago. Dr. Johnston was awarded a humble research fellowship to continue research in Germany at the Everhard Carl University of Tuebingen. And now she's a postdoctoral researcher at the animal minds lab. Her research focuses on the interplay between the brain and behavior such as working memory, timing and probabilistic reasoning and a range of avian species, including pigeons, jackdaws, and carrion crows. She is the lead author on a Current Biology paper. Crows flexibly applies statistical inferences based on previous experience. Melissa, thank you so much for being here today.
Melissa Johnston
Thank you so much for having me, and giving me the opportunity to talk about my work.
John Bailer
Oh, it's a pleasure. What don’t you know now, Millie? Well, I'm tempted to start our conversation with a question about the names of the crows, but I'm not going to do that. I'd like to get started with a little bit more relevant background. What led you to start exploring statistical inference ideas with crows?
Melissa Johnston
Well, actually, I wasn't the one who started this project. The project was started by our co-author, Dr. Katerina Brecht. So she was the real driving force behind starting this up and training the birds. I was just lucky enough to take over from her and get to do a deep dive into the data and present it to the world, I guess. She's also worked with a lot of bird species and their various cognitive skills. We've been on a couple of papers together now, another one from the same lab, where we looked at recursive sequence generation and crows as well.
John Bailer
So crows have this reputation for being really clever, and problem solvers. And sort of, you know, and I think I've even read that, that there's some tool that they might even employ. So I wasn't fully surprised to see this was the bird that was being used to demonstrate and dive into these questions. Can you give us just a quick summary of what was the research question that was being explored in this study?
Melissa Johnston
Yep, of course. So for this project, what we really wanted to know is, can these bird species learn to associate, in this case, arbitrary stimuli, such as a square or a triangle, to the probability of getting a reward. And so once we found that they could do this, we wanted to see if they could flexibly apply this information to make judgments about which would be better if they were presented with two options. So you're presented with Option A, where you've got a 20% chance of reward and Option B with a 40% chance of reward. We wanted to see if they could really apply this knowledge they'd learned and pick the more optimal item of where they could get a higher chance of reward.
John Bailer
So you know, this is just, to me, it's amazing to think about the fact that there are nine unique, abstract shapes that these birds are learning. I'm trying to think about if I could get nine shapes down with different probabilities, you know, so you've got this collection for these birds. And, and there's some training process for them to learn that yet, somehow, you know, maybe a filled circle is some, you know, 0%, and I know a triangle is 80%, or a filled triangle is 40%. And it filled squares 100%. How did you do that? I mean, how was this done in this lab?
Melissa Johnston
Yep. So the same way you would sort of do this with any species or even with humans, it's just through experience. And that was one of the key things we were interested in, because many statistical inference paradigms have the information immediately available to the subject making the choice, whereas we wanted to see if they could learn it from experience alone. So essentially, what we did is these birds were already quite experienced and being able to work in operant chambers. So thankfully, that was all done. So all we had to do was repeatedly expose them to the stimuli to let them build up an internal representation of a probability or reward.
John Bailer
So let's just deconstruct that expression operant chamber. So these birds are in a box, and they're picking or pecking one of two choices. Is that kind of the essence of it?
Melissa Johnston
Yeah. So for training, it was just one stimulus. And they're either rewarded or not. And then for the testing, they were presented with the two items, and they had to make a choice, which one they thought was better.
John Bailer
So how many times were they doing it? How long did training take? You know, how many times do you have to see? Yeah, how many times do you have to see a 40% response before you start getting a handle that oh, that you know that that filled triangle is 40% reward?
Melissa Johnston
I mean, I actually think that would be a very good follow up study, because we gave them a lot of trials, but we don't know how long it takes for them to build this representation. You know, if you think like, I always think about what I would do in this situation, and I think 10 trials, maybe I could do that with a couple of different reward outcomes, but for nine of them, like, I would need a lot more than 10 trials of each of those.
John Bailer
Yeah, I was, I was just imagining that it was probably that in the course of doing this experiment, a lot more of the time was spent just getting them to learn the rewards, the pairing of reward percentage with the abstract symbol that they were meeting?
Melissa Johnston
Yep. So they were trained for, I think they experienced a couple of 100 trials of each stimulus, which that's, that's a lot. And the fact is, they had to update their mental representation with each new presentation of the same stimulus. So, you know, the probability of something occurring changes every time that you've experienced it, and it's, you know, roughly something by the end of the representation, forming.
John Bailer
So, so they're doing this, what's the reinforcement? You know, what was the reward for that, that they were getting?
Melissa Johnston
Yep. So they get a mixture of these little tiny pills, which have all the nutritional value they need, and their diet, but also live worms.
John Bailer
I'd work for that. Yeah. Well, I'm also thinking that you're doing hundreds of these trials. I mean, at some point, they're just gonna say, Look, I'm full, I don't care what it is. Yeah.
Melissa Johnston
And they, you know,
John Bailer
They will let you know.
Melissa Johnston
Yeah, they just casually, you know, stop working. They're like, No, thank you, I'm done.
John Bailer
You know, there's some wisdom there, there's some, you know, that's probably a healthy dietary choice for them to stop. Okay, so now, I've got this picture in my mind that we now have these two crows, these are these two All Star crows that are participating in this study. And they now have some representation of some kind of relative frequency by which they're going to expect a reward associated with each of these nine abstract shapes. So these, you know, these shapes are going to be mapped to either 10%, 20% up through 90% reward. Now, here comes this, this is where the study begins, you know, where they have the choice, and they're going to have paired choices. So can you talk a little bit about the kind of the study where you're looking at what's the right choice for the bird to make?
Melissa Johnston
Of course, yeah, this is the really interesting part, as you said, because we presented them with two of the stimuli, and they just simply had to make a choice, there was no right or wrong answer, and that they weren't guaranteed a reward for picking the stimulus with the higher probability associated with it. So they simply had to pick the item, when it was in their best interest to pick the item with the higher reward probability. But that item, be it at the field circle, or the Red Square, or, I don't know, it wasn't always the optimal item. So for example, the 50% reward was the better choice when it was paired with 40, 30, 20, or 10. But it was actually the worst choice when paired with any of the stimuli associated with a higher reward probability. So what's really cool is that we could make all of these different combinations and see if they could make an optimal choice based on the pair that it was currently in.
John Bailer
So when they were encountering these pairs, you know, they're encountering like 40%. And 70%. Pair was sometimes the 40% on the left and other times a 40%. On the right, again paired with that 70%.
Melissa Johnston
Exactly, yes. So there were 36 unique pairings of the probabilities of the stimuli, but then when you CTRL for which side of the screen are presented on? It's actually 72 unique conditions. Okay, so it's very unlikely that we're like, Oh, I've seen this picture before. And it was the right choice when it was on the left, I'm going to do the left again. Because, again, remembering 72 unique combinations is not an easy thing to do.
John Bailer
So I'm going to sort of see if I can remember some lessons that I had learned about memory. So you can help me understand this. Yeah, that is the irony of the question, right. So with short term memory, we talked about maybe four to six chunks of information that can be retained. So really, I don't know what crows have in terms of, what's a crow's short term memory?
Melissa Johnston
Yeah, that's a good question. And another lab and book them also in Germany, they've had a look at this. And it does appear to be between four and six items very similar to primates.
John Bailer
Okay, so then what you've done in the training is you're trying to move this into longer term memory by just by the repeated exposure to these various stimuli. Yeah, essentially. Hey, I was just curious. I didn't notice when I looked at your paper, did you ever pair the same stimuli just to see if they if they picked left the same frequency as they picked? Right?
Melissa Johnston
Oh, that would have been a great probe trial? Yeah, no, we didn't, that would have been a really good condition to have.
John Bailer
Okay, I just was curious, because I was thinking, Well, you know, you answered part of my question about kind of whether there was any side preference, but you balance that out as part of this experimental probe that you're doing. But I just thought, Wow, I wonder if they did that. And if they did, do kind of half a chance of picking left or, you know, sort of that would sort of, in my mind, kind of be this nice reinforcement piece?
Melissa Johnston
So, yeah, definitely, definitely something to keep in mind for the future.
John Bailer
So there's my, that might be my future here. You know, thinking about these types of studies?
Melissa Johnston
Have you considered a career in that?
John Bailer
Oh, my training. I will say that one of my favorite undergraduate courses, I took a wonderful class in animal behavior by someone who was a psychologist studying, actually taste preference, and some of his taste preference were there were some, these were some squirrels that really loved incredibly spicy things. So there's this, there was a really interesting study of food preference and that, so like, I just remember it, he was trying to understand how people develop these types of preferences, too. So just, it was just fun to think about. So yeah, I liked the animal behavior stuff. I think it's really fascinating. You know, as you look at the work that you've done, I guess I need to ask, so what did you see? I mean, I've been burying the lede here. I mean, if Rosemary was here, she'd really be mad at me, because I have this great story. And I didn't get to the key part until halfway through the episode. So what was the big takeaway that you saw from this work? So the overall results? Yeah, particularly, you know, so they're, they're given a choice where you've got a higher probability, abstract thing, and a lower probability abstract thing paired together as a choice. So you would know, if they didn't use any information, you'd expect it's a coin flip where they might pick. So what did you see though?
Melissa Johnston
Yep, so what we found was that they were really good at this task, they were really good at picking the optimal choice, or the one with the higher likelihood of reward. And this, they were even better when the distance between the two items was larger. So when 90 was paired against 20, it was much easier for them compared to when 90 was paired against 80. So the bigger the distance, the you know, the easier it was for them, or the more they picked, the higher likelihood.
John Bailer
You're listening to Stats and Stories. Our guest today is Melissa Johnston, postdoctoral researcher in the animal minds lab at the University, Todd tahmina. de Barcelona. Well, Melissa, what was the hardest part of working with crows?
Melissa Johnston
Sometimes, just, oh, how do I phrase this? I don't know if there is anything difficult. These crows are really professional, you know, you train them to fly to your arm, they fly to you. And then you put them in the box and they work away and then they stop when they want to and then you take them home. I guess the hardest part is making sure that we are testing what we think we're testing. Is that that training to get to the point where we're like, Okay, this is what we really want to look at. This is what we're testing.
John Bailer
So how old are these crows when they get they start working with the lab?
Melissa Johnston
Probably about a year or so. Okay, yep. And they can live for you know, quite a long time, especially in captivity because they're very well looked after.
John Bailer
So I'm going to come back to one of the things that you were mentioning just before the break, and that was you were talking about kind of that that one of the things that was reinforcing this, this string result that you saw is that as you saw greater separation of the stimuli, so, you know, a 60/40 selection, there is not quite the same strength of frequency of picking 60, as you would see from a 90/10 pairing of stimuli. So that was what you call the distance between them, right. Okay. What else did you kind of learn in terms of the comparison of the stimuli and how it might have driven the responses that you observed?
Melissa Johnston
Yep. So the other big thing that we saw was, as you mentioned that that's what we call the distance effect, where the bigger the distance, the easier discriminations are. Basically, there's also the magnitude effect. So what that is, is where the distance, the difference between 20 and 40, while it has a difference of 20, is not treated the same as a difference between 60 and 80. And that's simply because 20, the difference between 20 and 40 is double, basically, but the difference between 60 and 80 is not double. So this is what we call the magnitude effect where it or the ratio, the ratio difference effect, where it's not just about the absolute number of differences between these numbers, or this reinforcement schedule, but also the relative difference as well. So
John Bailer
So when you're thinking about these ratios, all of a sudden, I'm picturing some kind of point two to point four comparisons, and then point three to point six comparisons and then point four to point eight comparisons. Do you see that you'd have the same relative selection of the optimal choice and each of those pairings?
Melissa Johnston
Yes, that is yes. So your example, where you're comparing point two to point four, and say, point three 2.6 and point four 2.8, that they are treated the same? Or roughly the same? Because the the ratio difference between them,
John Bailer
So that, wow, okay, I was curious, because I was thinking that, you know, maybe I'm trying to channel my inner Crow, and think about these results. And so, as I was thinking about, like, the point two and point four, it's like, I'm not gonna get it either way. You know, I don't care. But you know, and point 6.8. But maybe that's, maybe when I go to point when I'm thinking about point eight 2.4. I'm like, Man, this is almost a sure thing compared to the other, which is that less than a coin toss? Now, maybe my channeling of an inner CRO is not working. But I do, but it's interesting that you're seeing kind of very similar results in that regard.
Melissa Johnston
Yeah, so both of these are quite known sort of effects of the analog magnitude system. So this system or this internal representation that we have of numbers, and yeah, I guess, at the numerical system, basically.
John Bailer
So you also, I think, correct me if I'm wrong, but you also gave them a holiday of about a month, and then you tested them again. Yeah. And they were just as good. They were just, there was no, no loss.
Melissa Johnston
No, not at all. It's incredible.
John Bailer
I find that truly remarkable, because I'm not sure I would expect humans to do that. Well, if they had that kind of a break. Maybe these crows are not cramming for this test. They're actually learning it. And it's really, really has been conveyed to long term memory, and they're using it and they're still able to call upon it. Have you thought about going back? You know, maybe a year later? I mean, is there any sense of revisiting? I love that you went back a month later, but I'm just thinking, Man, I wonder if a year later, they'd still have that?
Melissa Johnston
I would not be surprised if they still did incredibly well on the task. As I said, these birds are very professional, which means they're used on a lot of different experiments. So they've moved on to two other experiments now, but I reckon if we, if we try them back on this one, they'll do just as good.
John Bailer
That's really– okay. So they're there. They're working on other things. Now see, I would wonder if they start to work on other tasks, if that might confound this task.
Melissa Johnston
Of course, of course, you really have to with animals for animal work, you really have to take into account their previous experience. Fortunately, the birds in this lab are very well trained on a lot of different numerical tasks. And so it's all generally in the same wheelhouse, I guess, for lack of a better word.
John Bailer
So what other birds have exhibited this type of reasoning where they can learn a little bit about frequencies associated with stimuli and then apply it?
Melissa Johnston
There's a really great study from a group. The lead author was Roberts, he did it with pigeons and that was one of the main influences for this experiment, but also with Kia as well. They've shown not so much the learning from experience but they still show statistical inference abilities.
John Bailer
We had an episode with probabilistic reasoning with that parrot. So that's uh oh, okay. You mentioned that and I think I saw in part of the review of your paper that this has been observed and explored in primates, as well as, you know, certain avian species. I'm curious about other other animals, one, you know, sort of I think was a, you know, octopi are thought to be very good at kind of problem solving. Has any of this work been done with marine species?
Melissa Johnston
You know, not that I'm aware of, but that's a very good avenue of future researchers checking in with them, we know that fish have good numerical abilities in general. So, you know, we could definitely look at fish to see whether they can do probabilistic inference. I think, generally speaking, most species that need to forage would be able to do this. Because it makes sense to remember Oh, yeah, I had food from there quite often, but not all the time. Or, you know, that place, I got food more often relative to this other spot. So I think it is an ability that animals naturally would sort of have.
John Bailer
I'm really glad you said that. Because I was thinking about why, why was the skill? Why would you expect this to evolve? You know, what kind of selective pressures might lead to this being advantageous for crows? And so if so, you've just sort of generalized it to all foragers? That's so that would be Have there been work with species that are non foragers? And that, you know, they just don't, your description here leads me to think if you studied non foragers, and compared them to the foragers, I'd expect much, much lower performance on such tasks.
Melissa Johnston
Yeah, potentially, although they might have a different reasoning for applying this kind of skill. But there is also potentially, you know, the publication bias where no results matter. So we may never know, even if they have done it, or tried to do it.
John Bailer
Yeah. Okay. So that's a really interesting question about what she sees the light of day, although you did mention other avian studies and primate studies where, in fact, this this type of reasoning from learning this pairing of probably, you know, probability of reward with certain stimuli, and then for kind of a fixed choice force choice phenomena where they seem to, to exceed what what chance would predict. So that's kind of nice, it's always nice to see this reproducibility in the same species as well as other species in other contexts. So it seems like there's some affirmation of that, that pattern.
Melissa Johnston
Yeah, for sure.
John Bailer
So when you look at this kind of work, where, you know what, what excites you about doing this kind of work?
Melissa Johnston
I've been fascinated with what birds in particular can do for actually not as long as what I would, what I would care to admit, but they're just so fascinating. Every time we try and look at something new in terms of a cognitive skill, they show that they can do it. So it's so exciting to me to be like, Oh, I wonder if they can do this. And then they can more often than not. And so just being able to be creative and think about, Oh, I wonder if they can do this and then seeing if they can I find that incredibly fun and such a joy to be a part of.
John Bailer
Yeah, I'm always delighted and impressed to think about just the distance that the birds migrate, just the ability to have this type of incredible ultra marathon annually, you know, to see what they can do and how they navigate that, to me is one of those, those great stories of nature. And this is interesting. And I'm curious when you think about bird species, if the longer lived species are more likely to demonstrate this than maybe the shorter lived species. Do you think there's any differences that might be there?
Melissa Johnston
That's an interesting question. I'm not sure if lifespan would necessarily correlate, but I would suspect that there's mounting evidence suggesting that birds or species with more neurons, or more densely packed neurons in their brain show more complex cognitive skills. So perhaps it's perhaps living longer as a consequence of a lot of these other things rather than the other way around.
John Bailer
Yeah, that whole direction of causality is a pretty challenging one in terms of the science game, isn't it? So tell me what, what's next. So what's on deck for investigations for you?
Melissa Johnston
Yep. So at the moment, I've been working on projects, I've almost finished it. So watch the space, but measure the crows ability to keep track of time, which, again is a little bit numerical, but a little bit off center. And yeah, we've been looking at both the behavioral and neural sort of evidence of our birds keeping track of time on a small scale.
John Bailer
Oh, that's Wow. I imagine that one of the great challenges to this type of work is trying to find the right type of experimental investigation that maps to this type of question. So you did this for the stimuli allows you to think about the In this current study allows you to think about by choice, whether or not they could make optimal choices. So now just that seems to be a much easier definition than thinking about some kind of exploration of the sense of time.
Melissa Johnston
Yeah, exactly. I mean, there's so many, like I said, there's so many different things that we can test these birds with. It's, and they keep proving to us, you know, bird brain isn't a bad thing. You know, they're incredibly intelligent. And I think going forward specifically with the, you know, the probabilistic reasoning or the statistical inferences in this style, it would be very cool to see if they can add layers to the, to their understanding or their representation. So, you know, for example, there's a 50% chance of reward. I'll get food from this location when foraging, but only if it's been raining, and there's a 50% chance of rain. So building these layers of representation of, you know, outcome probabilities would be very cool to look at.
John Bailer
That's neat. Yeah, I'm certain now I'm picturing you have this, you're now introducing this wet environment versus dry environment and running all these things so that you you're sort of stratifying this, oh, man, this is well, you know, as you get this working, come back and talk to us again.
Melissa Johnston
Have it a recurring spot on the show?
John Bailer
There you go. And now you know, so now when I tell people to call people a birdbrain No, no, that I mean it and love and affection and with respect. Well, that's all the time we have for this episode of stats and stories. Millie, thank you so much for joining us today.
Melissa Johnston
Thank you so much for having me.
John Bailer
Stats and Stories is a partnership between Miami University's Department of Statistics and media journalism and film and the American Statistical Association. You can follow us on social media, Apple podcasts or other places you can find podcasts. If you would like to share your thoughts on the program, Send your email to statsandstories@miami.oh.edu Or check us out at statsandstories.net and be sure to listen for future editions of Stats and Stories, where we discuss the statistics behind the stories and the stories behind the statistics.