“More than the sum of its parts” is practically the slogan of systems thinking. One canonical example is a beehive: individually, a honeybee is not that clever, but together they can function like shapeshifting metamaterials or mesh networks — some of humankind’s most sophisticated innovations. Emergent collective behavior is common in the insect world — and not just among superstar collaborators like bees, ants, and termites. One firefly, alone, blinks randomly; together, fireflies effect an awe-inspiring synchrony in large, coordinated light shows scientists are only starting to explain. It turns out that diversity is key, even in a swarm; variety improves the “computations” that these swarms perform as they adapt to their surroundings. Watch them self-organize for long enough and you might ask, “Is this what people do? What hidden patterns and emergent genius do we all participate in unawares?” If bees and fireflies inspire that kind of question in you, you’ll find yourself at home in this week’s episode…
Welcome to COMPLEXITY, the official podcast of the Santa Fe Institute. I’m your host, Michael Garfield, and every other week we’ll bring you with us for far-ranging conversations with our worldwide network of rigorous researchers developing new frameworks to explain the deepest mysteries of the universe.
In this conversation, we talk to SFI External Professor Orit Peleg (Google Scholar, Twitter) at the University of Colorado Boulder’s BioFrontiers Institute and Computer Science Department about her research into the collective behavior of bees and fireflies. These humble insects can, together, do amazing things — and what science shows about just how they do it points to deeper insights on the nature of noise, creativity, and life in our complex world.
If you value our research and communication efforts, please rate and review us at Apple Podcasts, and/or consider making a donation at santafe.edu/podcastgive. You can find numerous other ways to engage with us at santafe.edu/engage. Thank you for listening!
Join our Facebook discussion group to meet like minds and talk about each episode.
Podcast theme music by Mitch Mignano.
Follow us on social media:
Twitter • YouTube • Facebook • Instagram • LinkedIn
Collective mechanical adaptation in honeybee swarms
Collective ventilation in honeybee nests
Flow-mediated olfactory communication in honey bee swarms
Self-organization in natural swarms of Photinus carolinus synchronous fireflies
Spatiotemporal reconstruction of emergent flash synchronization in firefly swarms via stereoscopic 360-degree cameras
Further Listening & Reading:
Episode 29 — David Krakauer on Coronavirus, Crisis, and Creative Opportunity
Episode 56 — J. Doyne Farmer on The Complexity Economics Revolution
Stefani Crabtree — The archaeological record can teach us much about cultural resilience and how to adapt to exogenous threats
Annalee Newitz — Scatter, Adapt, and Remember
Laurence Gonzales on Behind The Shield Podcast
Michael Mauboussin — The Success Equation
Episode 55 — James Evans on Social Computing and Diversity by Design
@sfiscience on Orit Peleg’s research into honeybee olfactory communication
Orit Peleg (0s): Noise in a natural biological system can actually be beneficial. What we knew from the literature is that bees have a distribution of threshold that trigger the fanning behavior. There's not just like one temperature value distribution becomes wider when there is higher genetic variability. And what we showed was the model is actually having this variability of the threshold point to trigger the behavior can be really beneficial and help them create these aggregations more optimally.
The reason for this is coming from the aggregation process itself. So if you can imagine one bee that is fanning, driving hot air from the hive outside, this bee is standing, The temperature is going to be slightly higher. And then it's going to decrease kind of gradually as you get further away from that bee. If you just had a constant threshold that triggers the fanning, then every bee will just start kind of randomly fanning. But then if there's some bee that has a slightly lower threshold have higher density to start spending closer to the beat that is already funding because of the temperature distribution there.
So having the noise, having the variability in this threshold point is actually really helpful for this process of aggregation.
Michael Garfield (1m 40s): More than the sum of its parts is practically the slogan of systems thinking. One canonical example is a beehive individually. A honeybee is not that clever, but together they can function like shape-shifting metamaterials or mesh networks. Some of humankind's most sophisticated innovations Emergent collective behavior is common in the insect world and not just among superstar collaborators like bees, ants and termites. One firefly alone blinks randomly together.
Fireflies affect an awe-inspiring synchrony in large coordinated light shows. Scientists are only starting to explain. It turns out that diversity is key. Even in a swarm variety improves the computations that these swarms perform as they adapt to their surroundings, watch themselves organized for long enough. And you might ask, is this what people do? What hidden patterns and emergent genius do we all participate in unawares. If bees and fireflies inspire that kind of question in you you'll find yourself at home in this week's episode. Welcome to Complexity, the official podcast of the Santa Fe Institute.
I'm your host, Michael Garfield, and every other week, we'll bring you with us for far ranging conversations with our worldwide network of rigorous researchers, developing new frameworks to explain the deepest mysteries of the universe. In this conversation, we talked to SFI External Professor Orit Peleg at the University of Colorado Boulders Bio Frontiers Institute and Computer Science Department about her research into the collective behavior of bees and fireflies. These humble insects can together do amazing things. And what science shows about just how they do it points to deeper insights on the nature of noise, creativity, and life in our complex world.
If you value our research and communication efforts, please rate and review us @applepodcasts and/or consider making a donation at santafe.edu/podcast. Give you can find numerous other ways to engage with us at santafe.edu/engage Thank you for listening.
Orit, hello. It's a pleasure to have you on complexity podcast.
Orit Peleg (3m 56s): Thank you. I'm really honored to be here. And I'm excited to talk to you today
Michael Garfield (4m 1s): To kick things off in the realm of the human, rather than the realm of the insect, I'd like to invite you to talk a little bit about your path into being a scientist and what got you interested in it, what compelled you to pursue this professionally, why it is that you study these particular topics in your biography and your passion.
Orit Peleg (4m 25s): I grew up in Israel and I actually studied physics and computer science. I was majoring in that in high school and I didn't like it very much then, but then later on, I got really excited about astronomy. Just kept on finding myself, thinking about how I'm just a tiny little creature in a vast, vast universe and how it's so difficult to perceive everything, from my perspective is a small little creature. I think that's kind of what drew me into liking physics and wanting to study physics at the university, which I did.
I didn't end up working on astronomy because I sort of rolled into a summer research program as an undergrad with a professor at the Physics Department. And he was working on biological physics. So things sort of rolled out from there. I still think I kind of try to think about how things look like from the perspective of a small little creature building block in a complex system, only in biological systems, which turn out to be just beautifully and vastly reach.
So that's kind of how I got into it. There's been a lot more hopping from fields and continents throughout the years. And now I'm here at Boulder, also associated with SFI.
Michael Garfield (5m 46s): I think there are so many people who draw a very natural kind of analogy between stars in the sky and fireflies. In some weird sense, it's like nothing has changed of course a lot. It seems to me like actually the right place to start diving into your work is with the honeybee swarm work. And then we can go from there into the firefly work. So the first piece I'd like to discuss with you is this piece for nature physics that you led collective mechanical adaptation of honeybee swarms, and actually seeing you present on this at SFI was one of my first really like exciting vivid memories of sitting in Cowan campus, watching you give this presentation on the swarm behavior of bees and how they respond to someone sitting there shaking a branch off a tree. So this is a rather beautiful and devious experiment that you set up to understand the collective behavior of swarms. Why don't you take us in and give us a little exposition on the thinking that led you into this particular paper and then how you and your coauthors Peters, Saucedo, and Mahadevan went about actually doing this research?
Orit Peleg (7m 6s): So that's definitely one of the most exciting projects I had so far. I think the vision was coming from my postdoc advisor Mahadevan who is really good in seeing people who work on different systems and what kind of skills they have and how this can be projected to new and exciting problems. So I was working on microscopic biological active matter at the time, and he sort of saw that connection to a larger biological active matter. You were kind of illustrating it to use your hands, but I'm guessing our listeners don't get to see that.
So the structure that the bees create is material. So the bees really connect. They're linking their bodies. They're not flying. They're actually holding each other or holding a tree branch. They do this in very large numbers. There's tens of subtleties of individuals who are creating the structure and they have to deal with a lot of environmental perturbations coming from just static forces coming from gravity, but then also more dynamic ones coming from wings. And it's really just from a mechanical perspective, so amazing that they're able to maintain clinical stability and not break.
So how do they know how to do this when the swarm is so large and an individual is so small that it can only directly coordinate with nearest neighbors? And sometimes these mechanical stability actually demands the simultaneous collective coordination throughout the material. So that was the questioning in mind. The area, as you already pointed out, was a little bit crazy I would say is maybe one way to describe it. One approach to study these kinds of systems is to just observe them in their natural habitat, unperturbed.
Another way tends to be really helpful is to do something through the system to change something in the environment, and then from seeing how the system responds to these probations. Maybe we can learn something interesting about its underlying principles. So that was the logical idea behind shaking a swarm of 10,000 stingy honeybees. Turns out the main result was that when we continue checking the bees right and left, then these start moving around in the swarm and they collectively change the swarming to a more flat structure and more flat comb.
And then when we stop, they start slowly go back to their steady state configuration. We combine this also with the computational model that helped us isolate and point to possible rules of local behavior with the bees are following. And in that case, it was local mechanical forces that they're sensing between them and the bonds that they create to their neighbors. From that point on, we just kind of extrapolated on that and came up with a set of local rules where they sense these forces and go up the strain gradient towards the attachment surface of the swarm to get everything to spread out.
And this is all based on local rules of behavior. So only they're sensing local information about the mechanical forces responding to it, but that in turn leads to a global change in the system,
Michael Garfield (10m 19s): Just to help people get a picture of this, what we're talking about is actually like in a laboratory setting There's a plate hanging from the ceiling of the bees. They're hanging from the plate. The plate is on a motor and it's just vibrating back and forth until you get the bees to basically hunker down except it's hunkering up. So as the structure changes as bees at the tip of this column are responding to the basically big the end of a swaying rope, having to climb up to a place where they're not jiggling around quite so much.
I'm curious if you think that the analogy I'm about to offer is a valid analogy. One of the thoughts that I have listening to you talk about conversations I've had, and I've heard with numerous other people at SFI. One was captured in episode 29 when David Krakauer and I were talking about mass extinctions and how the depth of an ecosystem changes due to perturbations of the entire ecosystem and commit hits the earth at the end of the Cretaceous period. There's like a limbo bar that anything larger than I think it was like five kilos or something was almost guaranteed to go extinct.
I wonder if you think that these same kind of mechanical properties that you're observing in swarming behavior have a fundamental relationship to the distribution of niches and metabolic functions and body size and all of that when you're perturbing not just a swarm of bees, but you're perturbing entire ecosystems.
Orit Peleg (11m 54s):So I actually think a lot about what you just described. My sense is that these systems of large swarms of insects for example, are kind of working as distributed agents. They're making decisions based in a distributed way based on perhaps local information. So when something happens to the environment, it can be something catastrophic, like shaking them or hitting them out, things that makes them really uncomfortable and can potentially lead to devastating consequences. It seems like these systems of insects are really good in adapting to these unexpected and very harsh changes to the environment.
So one example is what we just talked about is the mechanical forces. We've also shown more recently that the bees are able to harness and sense locally fields off the airflow to handle temperature and ventilation, and even more recently fields of chemical concentrations. And they somehow mastered the ability to sense these complex physical fields just locally and know how to manipulate them. I don't know if that completely answered your question, but they thought this is where you're going with this.
Michael Garfield (13m 8s): I definitely want to get to these other papers that you just mentioned, but just to clarify that question, another example might be obviously looking at bees in the aggregate in this way has a lot in common with looking at human beings in aggregate behavior, in socio physics models, where you're looking at the entire riot, the entire marathon or you're doing economics modeling where the individual decisions are just sort of below the level of resolution of the way that you're actually watching changes in the market.
We just had J. Doyne Farmer on, and he was talking about people who just follow the behavior of other people. And he was describing a model in which you see the entire economic system deform due to indigenous strains in the same way that it seems that you're talking about these swarms, just one of these questions of to what extent can we derive insights from this kind of research into understanding collective human behavior and collective behavior in systems that are not composed just of one species, but of many species.
Orit Peleg (14m 23s): All of these are fascinating problems. I always get my brain gets a little fuzzy when I try to think about all the complexity of all the ecosystems and how sensitive it is to all these small interactions between individuals. I wouldn't go ahead and say that these are more resilient. It's hard to compare the resiliency of these insect forms to human behavior, collective behavior. But that's showing is that also with the bees was probably a point where things can catastrophically go wrong and the system is not prepared to handle these situations.
Michael Garfield (14m 59s): That seems to strike at the heart of that question, which is if you shake the thing hard enough, it does not cohere? The swarm actually breaks apart or the bees leave. I'm just thinking about Stephanie Crabtree wrote a piece for the transmission essay series that we did last year, where she was talking about the end of the Chaco Canyon culture, as, as we understand. It was not like these people all died. It was that their agricultural system was so perturbed by climate change that they basically scattered. IP Fest, panelist Annalee Newitz wrote a book about this exact thing, scatter adapt and remember about human society is basically getting to the point where the social graph fragments, because things are just too crazy.
But rather than linger on this, I would like to talk about this piece that you mentioned a moment ago about collective ventilation and honeybee nests and how bees are working together to regulate the temperature inside their hive. And I think it's worth just because this is such a juicy vocabulary word it's worth bringing up as you do in this paper and defining for people stigmergy, such a core concept here.
Orit Peleg (16m 13s): So in the more classical sense, stigmergy is a process where two or more individuals are communicating through cues in the environment. So one very famous example of that is ants when they lay pheromones when they're foraging for food. So each ant is the positing a little drop of pheromones on the way to and on the way back from a food source. So the nest and then other ants who forage kind of randomly have a higher probability to go into directions where there's a higher concentration of pheromones, which means it's ants were already there, which creates this positive feedback loop that leads the ants in the shortest path, from the food to the back home.
So that's just one example in more classical sense. These cues that are laid in the environment that were described so far were tend to be cues that diffused passively. So if it's pheromones, they kind of diffuse passively by the laws of physics and then just follow that kind of spreading behavior. So that's kind of synergy in the classical sense, but what we showed in the paper that we'll talk about now is the ventilation also with the mechanical shaking and also with the aggregation behavior, is that the bees do something active to change these cues in the environment.
This would then be communicated to other individuals in making things more complex, but also much more interesting, at least I think so.
Michael Garfield (17m 43s): Again, to lay out the experimental design a little bit, I think one of the things that I found interesting, the way that you're kind of required to constrain some of the variables of this experiment was that you and your coauthors chose to use a particular kind of beekeeping hive, the Langstroth beehive, which has this long narrow entrance. And so you've controlled for rather than looking at wild beehives with like all of these sort of different organic configurations that emerge again out of this kind of collaboration with environmental factors, you've set up a situation that makes it much, much easier to actually observe what's going on and compare it from hive to hive.
So that's just a point of to celebrate, like thinking about how to design experiments like this.
Orit Peleg (18m 36s): First of all, I just want to give a shout out to the rest of the coursers. My contribution here was mostly analytical modeling. My peers did all of these beautiful experiments. It ended up being really convenient that most of the commercial hives were actually helping us reduce the dimensionality of the problem from complex three-dimensional entrance sizes and shapes to something that is almost one dimensional,
Michael Garfield (19m 0s): A clear sense again, of how does simplify this in a real world setting and not just in a mathematical model, but now let's get into your model. This is cool. The way that you unpack, it's a different kind of strain then a swinging swarm. You're still seeing a redistribution of labor based on the bees making these local assessments and then moving to sort of balance out that distribution of labor. So you wanna talk about some of the details of this model and then what emerged from it, and then how that compared to what you were actually seeing in the observation of flesh and blood beehives?
Orit Peleg (19m 40s): The main thing we were trying to explain is how do bees decide where to ventilate their hive. A little bit of background there is that you already mentioned this really, really narrow one dimensional entrance. And then usually for these commercial hives, there's also a little wooden piece in the beekeeping. Jargon is called the porch, and then the bees can just sit on their porch and what they do to ventilate the hive is fan their wings. So they're really just like your fans. And then the question is where should they stand?
Where should they fan along these one-dimensional entrance. We already knew from Jake's experiments that they tend to do this in little groups. So they kind of aggregate at one position along the one dimensional entrance and they all fan their wings together. In terms of explaining how this is achieved based on sensing local information, this is where the mathematical model came into the picture. And the idea was that we know the bees are sensitive to temperature. They can sense temperature. Above a certain threshold of temperature above a certain value of the ambient temperature they have a higher tendency to start standing. So that was one ingredient that we put into our model. The second ingredient was related to a posing airflow. So you mentioned that you have one bee that is standing on the porch, and it's finding its wings driving hot air from inside hive to the colder environment. And then because of conservation of mass, there's going to be some colder air from the environment going back into the hive. But every time we have these two opposing flows, we're going to have some friction that is happening at the interface between these two opposing flows.
So on one hand, we want to have many bees fanning. On the other hand, we want to minimize that friction energy loss that is coming from opposing flows. And we also want to minimize the bees sort of working against each other. If it is a happy group of bees scanning one position along the entrance. Where they're not standing is where the colder air is going to come into the hive. But then if these start sending there and fanning, they're going to oppose that air flow. So there's going to be some energy loss there as well.
So we're kind of posing this as an optimization problem of what would be the most optimal configuration that would allow for global airflow, but still minimize all of these energy loss. And these are the main ingredients that we placed in the mathematical model, I don't know how much you want to go into the details, but we basically it was too complicated to get an analytical solution, but we solved it numerically and received really similar results to what we see in the experiments. We get the aggregation of bees that are fanning the wings together in one position along the entrance.
And that helps them to create that global airflow in and out of the hive.
Michael Garfield (22m 31s): One thing that I found curious about the factors that you're putting together in forming this model was that as it says here in the paper, when you're talking about task threshold model, when the demand for a task is large, more individual bees respond due to broad variation in task thresholds. This variation is higher in colonies with a queen, which has made it multiple times and promote temporal stability with regulation. So what is going on here? The maturity of the queen figures into this, but what is that?
Orit Peleg (23m 3s): Let me try to unpack all of this. So, first of all, more generally, I think it's another really nice example, how noise in a natural biological system can actually be beneficial. Let me explain. What we knew from the literature is that bees have a distribution of threshold that trigger the fanning behavior. There's not just one temperature value distribution becomes wider when there is higher genetic variability, which is comes, you know, to the point with the queen or the mating history basically.
And what's the genetic variability of all the worker bees. And what we showed you is the model, is it actually having this variability of this threshold point to trigger the behavior can be really beneficial and help them create these aggregations more optimally in comparison to a situation where there's really a sharps threshold uniform for all the worker bees. The reason for this is coming from the aggregation process itself. So if you can imagine one bee that is fanning driving hot air from the hive outside.
So where this bee is standing, the temperature is going to be slightly higher. And then it's going to decrease kind of gradually as you get further away from that bee. If you just had a constant threshold that triggers defending, then nothing would really happen here. Everybody will just start kind of randomly no matter where it is along the interest to start fanning. But then if there's some point it'd be that has a slightly lower density to start fending closer to the bee that is already shining because of the temperature distribution there.
So having the noise, having the variability in this threshold point is actually really helpful for this process of aggregation.
Michael Garfield (24m 51s): It's funny how it seems like every single conversation we have on this show gets back to the value of noise and the value of diversity in these complex systems. I'm reminded of a conversation I just heard with SFI Miller scholar, Lawrence Gonzalez. He was on behind the shield podcast recently, which is like a firefighter and emergency response show. And he was talking about the natural distribution in risk tolerance in people. It's good that we have some people that are highly risk averse and some people that are risk seeking.
And of course, thinking about the writing that SFI Chair Economist, Michael Mauboussin has done on risk. So there seems like a loop from that back to the first paper that we discussed, which is about bee. They're all seeking a lower strain. I don't remember this being in that paper, but it seems like there were probably bees that were more likely to try and climb the column and get to an easier place and bees that were down there swinging at the bottom, like, "yes, this is fun."
And that we need this basically. So that swarms and hives, don't kind of like seize up. And when all the neurons start firing at the same time. Maybe this has an interesting connection to the firefly stuff, but I'm getting ahead of us here just to throw one more piece on it for people that are interested in, in what this means in terms of scientific research. It reminds me of the conversation that we had with James Evans was talking about the science of science and how big scientific ideas emerge in these gaps between disciplines.
They tend to be discovered by people that are crossing from one discipline to another, which is inherently inefficient. To cross an inferential distance between one way of understanding modeling the world and another being used by a different domain is energetically costly. And so in a way I feel like your beehive ventilation work is an arrow in the quiver of the argument for interdisciplinary science, because you're saying, "listen, we need people willing to take intellectual risks here, otherwise our knowledge community hive would just overheat and die."
Orit Peleg (27m 13s): I love it. Everything you just said, very more for interdisciplinary science. And if we can actually add another metal level to that philosophical arguments, I think that just to connect it to the swarm shaking experiments, I think that over there, the noise or the flexibility, I would think of more in the transition from phases, from solid to liquid to gas. And let me bind this back to earth. So solid leads and crystals, they have a really pretty fine configuration, so it's very predictable. But it doesn't really allow for exploration.
And that's a really important process for the electrical systems to actually adapt to varying changing environments and unknown conditions. When I think about the swarm structure, it's something in between the solid and the liquid. There's still definitely some constant and strong bonds between individuals, but as it kind of melts and change shapes, it's a bit more behaving, a bit more like liquid.
And I think being in that transition where they can switch from these two configuration is really important for their mechanical stability, but then also for their survival.
Michael Garfield (28m 32s): There's just one more point I want to let you unpack here before we get into the next piece, which is something you mentioned in the conclusion of this paper is that if the bees fanned into the nest entrance, rather than out of it, they would have no information about the state of the hive. Interestingly, another cavity nesting, the honeybee species, apis cerana, fans into the nest. So, you know, you just mentioned in passing that this, this other honeybee species is likely using an alternative strategy.
What, like, it seems like everything that you just said about how the, you know, the collective computation of the hive is actually functioning. So what do you think is going on in these contrary situations?
Orit Peleg (29m 20s): The mechanism, the model mathematical model, would actually still work based on only having the information of the temperature at the entrance. You don't really need to know what is the temperature inside and outside. From that perspective, it would work. I just think that for bees who send air outside of the hive, they have this extra piece of information where, because they are driving air, they also might be able to sense what's going on, what's the temperature inside the hive. But I have no idea why this other species of bees do it reversely.
And it just to add to the mystery, there's some really interesting behavior in feral honey bees, where when the entrance size is very small, they don't just aggregate in one point along the entrance because it's so small, it doesn't really make sense. So they start switching from that strategy and intermittent fanning. So they fan for a little bit and then stop then fan for a little bit and then stop, which is really similar to what we do when we breeze, by the way we kind of have we introduce airflow and then stop, introduce effort and stop.
So there are all these wild behaviors in the natural environments that I think would be really cool to kind of explore and understand also in the context of this model
Michael Garfield (30m 40s): Or reckless speculation. I wonder if you pointed to the opening at the entrance of the high, of being of a different size in different species. And wonder if that has any relationship to the economic discussion around scaling trust, like scaling, coupling between people. I've been having a conversation on Facebook about usury and how historically my dad's side of the family's Jewish. Interest bearing debt is something that you don't do within your own community.
It's something that seems to have emerged as a way of accommodating a certain amount of risk for engaging in financial transactions with people outside of a circle of trust. I wonder again, if there's some sort of underlying pattern here in the way that you see bees operating collectively. You basically said as much that it seems like the size and shape of the entrance is a key figure in this. I would be curious to know whether this has something to do with scaling laws. Whether you see the emergence of these new collective properties as a result of the size of the hive, and then the size, the size that the entrance of the hive has to be in order to accommodate a hive of that population.
Orit Peleg (31m 58s): That would be really interesting to see the size of the entrance and see at what point they switched strategies. And I don't know the answer to that, but I think it's a really cool direction.
Michael Garfield (32m 11s): You just published this other study led by Nguyen, Iuzzolino, Mankel, Bozek, Stephens, Peleg.
Large, large Colorado team, mostly Amsterdam.
Orit Peleg (32m 34s): Japan, Germany. So it's kind of international, which has been really, really fun. So all the authors on this paper are just fantastic. And it's been a collaborative effort.
Michael Garfield (32m 47s): This piece flow mediated, olfactory communication, and honey bees swarms helps answer a question that I am sure many people naively had is as a child, which is how on earth are these bees finding their way home? How are they communicating with one another over these great distances over the so-called sight line of the dancing and visual communication? Without getting too far ahead of ourselves and spoiling it there's it seems like there's some really interesting implications here about human telecommunications networks and how we can structure those in a way that draws on the sort of intelligence of evolved living systems rather than designed technological systems.
Can you talk a little bit about this piece please, how you use the modeling and then also the machine vision, which has led for some really interesting videos to share over social that we'll link to in the show notes.
Orit Peleg (33m 45s): This is a piece I'm really proud of. It's one of my first, last authors paper. So I'm very excited about it. The question as you already pointed out is how do bees that are far away from the queen managed to locate her when they're actually relying on pheromones that are very volatile molecules at the cave rapidly in time-space. It was already known that of course, bees use pheromones and they have this behavior called scenting. So what they're doing is basically sticking their abdomen up, which exposes a pheromone gland.
And then they found the wings, which drives airflow on top of that farmland and basically distribute these pheromones in the direction of the airflow. What we found out is that when a worker bee is getting close to the queen, instead of just walking all the way to the queen, it would stop at a certain distance and scent. So amplify these pheromones, and it would do it away from the queen. And then in another, we will do the same thing based on those amplified pheromone. And so it would stop and propagate them backwards and so on and so on.
This relay activation process which basically allows the pheromones to travel distances that are much longer than they would be able to if it was only coming from a single bee. So that's the general process. I should probably also clarify, just related to what you said before. We know that this is helpful on a length scale of, a few meters let's say, but bees actually forage over very large distances of up to five miles, I believe is the number.
They definitely have other strategies to find a way back home passing integration and visual cues and so on. But actually one of the first things was shown with these usage of pheromones is that when there's forging these exiting the hive, there's a bunch of bees setting again on the porch and they're sending these times, so they're creating these pheromones and there's a positive correlation between the number of sending bees and the success rate of foragers to come back to the hive. So it could be that they sort of find a way closer to the hive, but then these pheromones actually on the length scale of a few meters really guides them back to their hive.
So that's kind of the biology. Experiments we basically had a semi two-dimensional arena where the queen is located in one corner. And then we put all the rest of the bees at the other corner, and then we can see how they start exploring their arena and sending inform the rest of the bees. And this is where the new computer vision code came into place. And that was solely the contribution of Greg Stephans and Kesha Bowser who guided to my creating this new machine learning code.
And the idea was using some training that other to be created, they were able to create a code that would classify and find all of these scenting events. So of course for us, this was a major upgrade to us watching the videos and clicking on all of the scenting events. It allowed us to have a really large datasets relatively quickly. And then the really cool seeing the really like aha moment there was when we played the movie, we stop it at a certain frame and then we can see all the positions that the bees are scenting and in which direction they're scenting.
And then if we take all of these little directions, they create and integrate them into something that looks more like a global map. We saw that it really leads to the queen. So falling all of these local directional cues that are coming from individual bees are trying to find the queen can follow these and it would really lead them to the position of the queen. So that was a really exciting moment. That's kind of broadly and I'm happy to elaborate on any other points.
Michael Garfield (37m 46s): The things that came out of this that I thought was interesting in figure two, you talk about the distance between sending bees is invariant to be density, suggestive that the bees have a computational strategy.
Orit Peleg (38m 2s): Thank you for bringing this up. This wasn't really also another important point. Another important piece of information we got from the experiments. So let's assume that we have a bee that is producing pheromones. For the sake of the argument, assume that it's producing them as a tropically. So it moves, it spreads the same in all directions. If you have another bee that is exploring the space, depending on the distance that be from the sending bee, it's going to sense the particular concentration of pheromones, depending on how far it is from the same thing, the distance is directly to the concentration of pheromones.
That bee would sense. And the fact that they are actually spaced in order to very particular characteristic distance from each other, suggest that they might be using concentration of pheromones to position themselves. So that was one big component that we included in the computational model, which is the bees might be able to send formal concentrations. And once they reach a certain threshold, then they would stop moving and start scenting themselves so amplifying the pheromones in a certain direction and the direction in this case, we chose also to be related to the gradient of concentration of pheromones.
So they would have a tendency to move the pheromones backwards towards the less informed bees.
Michael Garfield (39m 24s): I don't know if this is a valid point, but it strikes me a lot as kind of akin to being on college campus and like trying to find the party. By the time you get to the party, you're done looking for the party. You're not trying to find the busiest room in the party. Instead, you're calling your friends, telling them where the party is.
Orit Peleg (39m 48s): That sounds as long as you're not calling them over cell phones, it would have a long-range connection. Then I think it's a valid analogy.
Michael Garfield (39m 55s): Let's say instead of a party, it's a protest and that we're using a Bluetooth mesh network that's percolating a signal through Bluetooth. It's not just like point to point, but you actually have to, the signal has to diffuse. It seems very, very useful information for people who are organizing rallies and that kind of thing. There's a lot of other interesting stuff about this, but for the sake of time, I want to make sure that we get to your work on fireflies for Photinus Carolinus, such a lovely name and you were last author on a piece in Royal society interface on using 360 degree cameras to actually like start to substantiate or to ground all of the models that people have tried to make about fireflies synchronization in actual data. So there's some really cool coverage of this research that has come out in various press publications we'll link to in the show note. You can see pictures of this extraordinary field work that you got to participate in.
And so this is just, you know, I would love to hear you talk a little bit about this piece and how your team came up with the idea to do this, how that facilitated a more rigorous quantitative approach to studying firefly synchronization.
Orit Peleg (41m 16s): Where it all started is really doing undergrad, where I took some physics classes and we handle the comical systems and mostly focused on the math, but we had a really nice teacher. We have a lot of nice examples, relatable examples. For example, synchronization. And still to this day, remember the example of fireflies and how they synchronize their flashes. So really, I would say kind of grew up on these ideas that fireflies are landmark example of synchronization in nature.
And they correlate that this behavior can be explained by a very wide reach set of mathematical models that are generally hard to as the core multimodal where individual agents are considered to be oscillators that perform some periodic function. And then there is some coupling between pairs of agents that leads them to synchronize their phases in the periodic process. And that's kind of generally. I guess when I was kind of thinking about starting my own lab and what do you want to writing my research statement and all these sort of things.
I also watched a lot of nature documentaries, which is something that I try to do all the time, whenever I have time. And so I came across some beautiful movies of fireflies. And I started thinking about all of these mathematical models and reading a little bit more deeply into the literature. It turns out that there was really very little quantitative data on firefight flashes and there's many good reasons for that. Most of the research was published a couple of 10 or 20 years ago when cameras and technology equipment made it really hard to record large swarms of fireflies, they have to be really close to the cameras in order to be captured.
And also the firefly synchronization behavior is really rare. So you have to be there. They're in the right place at the right time. Fireflies have these interesting life cycle where they are spending they're adults who flash what I think people usually think about when they think about flashing fireflies. So they're adults for about two weeks out of the year, and then the rest of the year, they're kind of underground and waiting, springing down.
Even within these two weeks, they only flash for a few hours every night. So it's really kind of a race against time to try to capture their behavior. So all I'm saying is that it's been kind of difficult to capture this data and somebody who is insistent enough to actually really care about this and go out to the field. So we actually drove to the field site in Tennessee last year and the previous year for a couple of weeks and managed to actually capture some quantitative data about swarms of fireflies.
Michael Garfield (44m 16s): Just to set this up in people's minds, what we're talking about is setting up multiple 360 GoPros so that you get a stereoscopic view two, 360 panoramas, that then allow you to correlate things in 3d.
Orit Peleg (44m 32s):In these mathematical models, the distance between the flashing firefighter can be really important. So we really wanted to have this 3d representation or fireflies. And the idea for the 360 cameras was actually from a very talented post-doc I have in my lab who came up with this idea, which is really cool because when we position these 360 cameras, we really get a nice view of the swarm from inside the swarm, almost like we're no single firefly inside the swarm. We can see what's going on around in comparison to having external cameras
Michael Garfield (45m 8s): Again, to contrast this to previous work that's been done on this, you mentioned in this paper that previous measurements by Copeland and Moiseff described similar intermittent or discontinuous synchronicity, but your observations were different than theirs. They looked at it as sort of like a square wave where it seemed to them group flashing started and stopped abruptly, but you actually measured a kind of gradation into an out of synchrony, which seems like something you might only notice if you're being able to, as you just said, watch the entire thing at once.
Then also there's a piece in this study on flight and how these are not just static points that are coordinating with each other, but they're all changing their distance from each other in this cloud. So each one is this point that's moving around inside the swarm. And if I'm to understand this correctly, that's like one of the mechanisms behind the sort of slop that you see as the system is moving in and out of synchrony. That it's like jostling inside.
Orit Peleg (46m 11s): So there's definitely some mixing coming from the movement of the fireflies. We don't fully understand it yet. It seems to be like this mixing doesn't have a directionality, so the fireflies are flying. And if you look at the collection of flight trajectories, the reason the particular direction in which they all tend to move to, this is probably coming from a search behavior because what's happening is that the males are the ones that are flying in and flashing in synchrony. And they are the reason for synchronization in the first place is for mating purposes.
So they're looking for females who tend to be closer to the ground and more statuary. But the idea of mixing here is very intriguing in terms of also connecting these two mathematical models,
Michael Garfield (46m 56s): It's also worth mentioning, I think, in this paper for highly analytical geometry nerds that you and your team ran into a bit of a challenge which is that you're filming this in the photospheres that you get from these cameras. It's like the same thing you you're dealing with map projections, you have to take these spheres and then flatten them and then correlate that. So can you talk a little bit about the artifacts of this and how your team addressed this potential source of error in the three-dimensional models?
You were building of these swarms.
Orit Peleg (47m 31s): The idea of 3d reconstruction. So X estimating depths from StereoVision is not new, and it hasn't been solved many times between you had this wheel, basically working with a periodic image on both the rotation angles. So we had to take this into account and adjust all the mathematical formulation to the depths estimation. And there's of course, a little bit of caveats when you are right underneath the camera, but otherwise you can really think about it as a stereo reconstruction, just with a more wider field of view,
Michael Garfield (48m 12s): You took this, and then you compared what you were observing in the wild to, again, this is like one of these, like you get to mess with bugs and mess with their heads for a living, which is just amusing to me for some reason, I guess, it's distant enough. The section where you're isolating male fireflies and you're throwing them into tents and you're measuring what one does in isolation, and then you're, you're controlling it. You're adding them incrementally. And this is another instance where you see a transition as the number of fireflies goes up.
What did you see here? It seems like this is another instance of somewhat similar dynamics to what we're seeing in bee coordination.
Orit Peleg (48m 55s): You're right. It's all faced transitions. You're right. Thank you for finding a connecting theme. It's always hard for me to do actually from my own research, but you're right. There's definitely a phase transition here. The reason we worked with the tents in the first place is that the swarm is really big. So it spends miles over miles. We've never really seen the end of the swarm. So we can really just record a small fraction of it. We really want to have a situation where we know how many firefights we have and what is their field of view and what they are perceiving.
And that was a bit harder to do of course outside in the natural environment. So for that, we just brought in dark tents that were served as a controlled environment where we could isolate a few flies from the swamp and bring them into this controlled environment. I guess the really surprising thing that we found is that as we decrease the number of fireflies that we have in that isolated environment, the synchronization becomes less well-formed to the point where we have individual fireflies.
There's almost nothing periodic about the way that they flush. So if you look at the time between the start of the flash burst, then it spends a really wide distribution. And that's surprising because again, as I mentioned, I kind of grew up on these beautiful ideas and mathematical models that assume that an individual building blocks in these agent-based models also behave like an oscillator. Also performs a periodic function when it's by itself, isolated from the swarm.
And we saw that this is not exactly what the fireflies are doing.
Michael Garfield (50m 39s): This kind of reminds me of the conversation around neurodiversity. And I have a friend from college who used to say, "no man is an Island, but some are very long peninsulas." And just this notion that again, like there's to look at it in a more of a social scale, SFI prides itself on having distance enough from the larger academic environment and staying small enough that it can be weird.
This comports with the experience that I remember as a child of seeing these beautiful swarms of synchronized fireflies catching one in a jar, and then being completely unable to predict when it's going to flash. I'm thinking about this now also in terms of the effects of remote work on professional existence, and just how weird and divergent people get when we're isolated from one another. And we're just sort of allowed to, you know, this is actually, obviously, this is like one of the fundamentals of science as it's practiced in a social context which is that we keep each other in check we're cuing off of each other. We're sinking up in ways that we don't when we go camping. And I mean, arguably that's really good for some reasons. So this is where we get into this thing about, you see this in the experiment you saw in your controls, that there were early flashers, mobile flashers. And again, I'm kind of thinking about that natural variation in the honeybees and the variation in risk tolerance and all of that. So it just interesting how even here in a system where it seems like there's much more regimented coordination between members of this collective, that you still see this sort of emergence of different personality types and like early adopters, the ones that like hang out and stay put.
Orit Peleg (52m 32s): I thought it was a really interesting way to put this behavior of the fireflies in isolation. And I can definitely relate to, you know, becoming weirder in isolation.
Michael Garfield (52m 42s): I mean, maybe this seems like the right place to talk about the last piece that you sent me, which again, this is, you know, you're the last author on a piece with Raphael Sarfati and Julie Hayes. So this is like a little subset of that research team, continuing investigation of this phenomenon, the relationship between the density and the correlation of fireflies. One of the things that comes out of this is just the importance of the environment in shaping self-organization and collective behavior, which I think has been a strain through the entire conversation.
But I would love to just invite some remarks on this piece and how this piece sort of deepens and expands on the insights of the prior one.
Orit Peleg (53m 26s): This piece is just focusing on the fireflies outside in the forest, in the natural environment. And there is something that we were able to capture a little bit with our cameras it's of course, much more. It's just amazing to be there and see live. So there's this way of the onset of a burst that propagates throughout the entire forest. And you can really see it coming towards you in the forest. And it's really magnificent. We're able to direct our cameras in a direction that actually captured some of these propagating waves and also get some statistics on other directions.
This is again where quantitative data becomes really useful. We are not the first one to discover this. This has been discovered before, but the tool that we have today, the researchers did before is that one researcher sued a hundred meters away from the other researcher. And then they just shouted whenever the wave got to their point and they kind of measured it this way. But now we're all doing the best we can do in the given technology that we have. And now we can actually capture and do the 3d reconstruction of these propagating waves.
Propagating waves, of course, tells us something about the mechanisms of synchronization and how information is transferred between individuals. So it's a very interesting and important quality to have. What we found out in this paper is that the speed of the wave that is propagating is order of magnitude faster than the speed of which individuals are flying. So this has to come from some interactions between individuals and triggering of the flashing behavior based on the flashings of other individuals.
And another important thing that we found out is that these, all of these synchronization and wave propagation occurs inside very dense vegetation. So those trees and those bushes and the fireflies just kind of move around them. And we were able to capture with the 360 cameras, the wrecking of symmetry in terms of the directionality. So if you put them in different points, there's going to be certain directions in which we will see more flashes in comparison to other directions.
And that's really just coming from the visual occlusion that is there in the environment. This paper is really just highlighting the importance of considering the environment and the conductivity of the network of interactions, how it affects that network when thinking about the synchronization of the entire swarm.
Michael Garfield (55m 59s): And the conclusion to this piece, talking about the result mixture of short range and long range interactions, you make an interesting point here, which I think we just spoke to really, which was that this self-organization allows for the possibility of an individual to position itself, to be more or less connected by flying higher from the ground, lower to the ground. They have some ability to modify the line of sight or like the wavelength. If I'm getting this right, this kind of reminds me of, again, diversity within various societies and various animal species of spy hopping behavior in whales may be, might be an example, or the loaner weirdos that wander off. You would ask yourself, if being a social organism, like a chimpanzee is so valuable, why do you have these individuals that wander off? Again, it speaks to the way that variation within a given population allows for cohesion.
The fact that this is all occurring within a mating context, it's funny because you think if people are competing for mates, why would they copy each other?
Orit Peleg (57m 13s): Let me just throw out like a teaser that's possible in the first paper that came out in the fireflies. We did find out that there is, even though the firefighters are synchronizing, their flashes there's actually importance to the movement. So they're kind of drawing a flash as they move. And that kind of shape actually has some influence on the synchronization of the entire swarm could also be, this is just speculation, but could also have some influence also on what, not just the males are perceiving, but maybe what also the females are proceeding.
It's a really interesting question of how do you the male fireflies distinguish yourself from the rest of the swarm while still getting the attention of the females that we already know from the literature that they are responding to more punctual swarms. So it's a very hard problem to solve. And this is kind of a future direction that we're taking this in my life.
Michael Garfield (58m 12s): It stands to reason I think that answering this question will help us or could help us understand how to allocate science funding a little bit more effectively? How to distribute money to these sort of large projects that are occurring within a given paradigm. And then like how much money we as a society should consider giving to the crazy ones that wander off.
Orit Peleg (58m 44s): It was up to me. It would be just the crazy ones, but that's the whole different story.
Michael Garfield (58m 49s): You then don't get any synchrony. In closing, what would you, as someone who has a clear and infectious enthusiasm for this world and these ideas and research methods, what would you suggest to people that are looking to get more involved in the study of collective behavior or the properties of complex systems? I mean, obviously like join your lab, right?
Orit Peleg (59m 18s): Joining the lab or taking some of the nice courses SFI is offering. That would be my first guest besides this. I would recommend being in nature, observing these animals in real life or watching nature documentaries. If you can't do that reading popular science books has been a really major inspiration for me. It's like discovering whole new worlds, you know, all these weird animals and weird behaviors.
And kind of that always brings a question of why aren't you doing that? And what happens if you tweak the behavior slightly that way? So that's kind of what I recommend people, nature movies that seems like a fabulous place to ground this a little concrete. Get outside.
Michael Garfield: Orit, this has been an absolute pleasure.
Orit Peleg: Thank you so much. Again, it's such an honor to be on the show and I really enjoy it.
Michael Garfield (1h 0m 33s): Thank you for listening. Complexities produced by the Santa Fe Institute, a nonprofit hub for complex systems science located in the high desert of New Mexico. For more information, including transcripts research links and educational resources, or to support our science and communication efforts. Visit Santafe.edu/podcast.
Transcribed by machine at podscribe.ai and edited with help from Aaron Leventman.