Ant Encounters

Key Takeaways

[Music] [Applause] [Music] welcome to another episode of data skeptic animal intelligence today we're going to do a little Bri broadcast this is actually an older episode if you're a longtime listener that we originally aired in our first artificial intelligence season but how fitting to uh speak to the author of the book ant encounters during our more biofocus season Becky were you familiar with this book before listening to the episode I was not but I might have to pick it up now it sounds really interesting it was a great read yeah especially at a time when I was just starting to take an interest in ants would you say this made any specific aspects of ants more interesting to you made The evolutionary reasons that they do what they do more interesting to me because they seem very lopsided so after I listen to the episode I dove into the literature on Ant learning and cognition which is actually pretty scant because one thing you both talked about is they don't necessarily learn really well and what I found are some papers where they learn very specific things really well but they're not broad Learners so I'm really interested in you sociality and evolution and how it produces these automatic sort of algorithmic responses to different stimuli in the in the environment versus when do you need to learn and when do you need to be flexible and I think that's a those are super interesting questions and and y'all talk about that a fair bit in the interview individual ants don't seem to learn except on maybe special occasions how do you feel about this idea that maybe the colony as a whole has its own version of uh first of all learning or maybe cognition or if we really want to uh stretch the definitions maybe it's sentient in its own way yeah we could say that because humans do a lot of Cooperative behaviors too where we have these emergent properties of societies in terms of of things that get done so I think that's an interesting question from an intellectual point of view like what is it and we see this a lot in evolution especially of you social insects where do we measure selection at the level of the colony all the individuals are just the queen and things like that so I think I'd need to read more about that to give you a more definitive opinion sure I think maybe some grants are in order to really answer the question but yeah yeah call up the NSF we're ready but I do think those are really important and interesting questions any other takeaways on your end and you both talk about this just a little bit but I think the point needs to be belabored just a little bit is that ants are a family within the Hop which includes your bees wasps all that stuff but each Gena each genus in that family have very wildly different life histories and evolutionary pressures so you can't necessarily study instinctual Behavior algorithmic type Behavior learning behavior in one ant genus and say this applies to everybody so I think if we keep studying ants we're going to find all kinds of cool little maybe gas isn't the right word but little uh mechanisms for getting things done that don't exist in other species so they're very much worth investigating like I found a study on an ant lassus Niger they punished them for using pheromone trails to find food and both of you talk about pheromone trails and they learn to not use the pheromone trails anymore they were like no I don't want to go get shocked but what they didn't learn is that the good stuff the sugar is in the trail without pheromones like they only got As Good As chance so they did learn to avoid the trail but they never learned that avoiding it is what gets you food so I'm like that's really bizarre I think making sure that knowing that ants are a really big family really diverse life histories and you're going to find all kinds of cool stuff the more you study them they're going to be really different from one another and lots of opportunities to compare and contrast then I think there are some true learnings for researchers to have here in learning theory and agency and all stuff like that yep you made me a Believer we need more funding for the ants absolutely hopefully that comes out of this well let's jump right into the episode Deborah welcome to data skeptic thanks it's great to be here so I completed your book ant encounters earlier this year I want to say something I don't mean this as hyperbole but I really thought it was a page Turner and I couldn't put it down that's great that's not something you often say about non-fiction for those who haven't had the pleasure yet can you describe what the book is all about the book is an introduction to the idea that the way that an ant colony Works without central control is to use patterns of simple encounters so ants respond to the rate at which they meet other ants and in the aggregate that regulates the behavior of the colony yeah it's that frequency that was very interesting to me I had not previously been opposed to that idea could you tell me maybe a practical example of um what ants are doing and how those encounters will alter their behavior one example is the behavior of harvester ants in the desert the foragers go out and search for seeds and then they come back and they're inside the nest for a little while until they go out on their next trip and an outgoing forager uses the rate at which it meets returning foragers to decide whether to leave again since the ANS are searching for seeds the amount of time a forager is outside depends on the availability of food if there are more seeds out there they find them more quickly they come back more quickly and more ants go out so it's a simple kind of feedback that links the amount of food out there to the experience of an ant without an ant having to assess how much food there is does that mean that ants are counting no they don't have to count they only have to use the rate at which they meet so tell me a little bit more than about memory how do they recall th those rates it works like a neuron think about how a neuron decides whether to fire a neuron gets a charge from another neuron and when enough stimulation accumulates the neuron fires but that stimulus leaks because electrical charge leaks out the body of the axon each stimulus has an impact which then decays if it gets enough stimuli quickly enough before the rest of them have decayed then it'll fire and that's how the ants assess interaction rate also so we used in fact a model based on this idea of a leaky integrator which is what neurons are to measure the behavior of the ants and it fit very well so the idea is that each time an ant experiences an interaction with another ant it stimulates a neurophysiological response which decays and if the ants get enough interactions before the last one has decayed then it pushes the over a threshold where it's likely to leave the nest and forage but if it doesn't get another interaction for a long time then it kind of forgets anything ever happened and the process has to start again so one of the Impressions I took away from the book was and and obviously you will see my bias here as a student of computer science but I felt like oh these ants they're very simple programs they're executing very basic rules is do you find that that's a fair assessment yes so then where does the intelligence of an ant line why well amps are really not very intelligent uh you could maybe say that colonies are intelligent in that they can adjust to changing conditions and do a lot of complicated things but really it's not a matter of intelligence it's a matter of how all of those very simple interactions um in the aggregate produce an outcome and that's very analogous to a computer yeah there's certainly an emergent property to it when these simple procedures kind of fit together very well I guess yes would it be fair to say then that ant behavior is really more the product of evolution than of learning well absolutely yes so these algorithms these simple rules that the ants are using are the product of evolution so what's evolving is the way that the uh rules that put together simple interactions produce some outcome in the behavior of the colony and that's what natural selection acts on so do ants get you know I know they specialize to some degree in tasks are they able to improve and learn something about their task over their lifetime well first of all they don't specialize as much as people may think MH I know it's part of the popular version of ants that each ant has its job you know in the movie ants each larva is assigned its task at Birth by a bureaucrat with a clipboard but in fact ants do move from one task to another early on in many species and of course we don't know what goes on in most species but the ones we do know about the ants work inside maybe taking care of the other younger ants the juvenile forms the larv and the pupy when the ant is Young and later in its life it moves to work outside the nest and to forage an ant moves through a series of tasks but to get back to your question we don't really have any evidence that an ant gets much better at a task by doing it with ants then not being Specialists the way popular opinion seems to have formalized how does an ant determine if and when they should switch their task again they seem to use the rate of counters we can do experiments in which we change the need for ants performing some task and ants will switch tasks in response it seems as though when there's some condition that creates a need for more ants maybe the ants meet for example um more ants doing cleanup work and there's some positive feedback on that so ants are more likely to switch when they meet other ants doing a certain task probably the rules for switching task depend on encounter rate and we have found that for some TKS in some species uh we don't know yet if that's generally true but it looks like it may be yeah that's another great point I mean ants seem to be a pretty diverse uh label can you talk a little bit taxonomically about the amount of species there are and the diversity and evolutionary path they've taken I want to give a personal thank you to our latest sponsor delete me they have helped me out text Data dat to 64,000 they'll text you back so you can start exploring if this is the right solution for you delete me finds and removes any personal information you don't want online I was very surprised how much they found on me they can get things removed and they can make sure it stays off delete me is a subscription service that removes your personal information from the largest people search databases on the web and in the process it helps you prevent possible ID theft doxing fishing scams and a dozen other situations where you got to waste hours of your time cleaning up a mess someone else caused I'm glad to have delete me set up I'm done with rooc calls I'm almost done with calls in general as a result of all the rooc calls sign up and provide delete me with exactly what information you want deleted and their experts take it from there delete me will send you regular personalized privacy reports showing what they found cuz it's not just a onetime service they're constantly monitoring delete me has your back take control of your data and keep your private life private by signing up for delete me now at a special discount for our listeners today we get 20 % off delete me Plans by texting data data to 64,000 that's Text data to 64,000 the only way to get your 20% off is to Text data to 64,000 that's data to 64,000 message and data rates may apply see terms for details can you talk a little bit taxonomically about the amount of species there are and the diversity and evolutionary path they've taken there are 14,000 species of ants that have been named so far there are probably many going extinct in the tropical forest as the tropical forests are being cut down while we're having this interview so we don't know how many there are there are probably many more species than that they've been around for um about 130 million years and they're very very diverse so there are ants that live in every conceivable habitat on Earth they're everywhere and they make a living in all kinds of amazing ways they nest in the ground they nest in trees they nest in the Burrows of other insects they nest all over the place they eat all kinds of things so they're really very diverse and we've studied only a tiny fraction of the ants so there are maybe 50 species that anybody has ever looked at in detail and so that leaves 13,950 that we don't know much about well good opportunity for grad students I suppose yeah um yeah there's a lot to learn about ants well we've had some hot days on an especially hot day I like to get cool one thing that hasn't happened yet is a surprising encounter you mention in the book where might those ants go they might go um anywhere in the uh in the walls of your house um they might come up through the drains um but eventually they'll go back outside because they prefer to be outside oh I was thinking specifically of the calls you sometimes get where people are surprised to find a big big cluster in their home freezers oh yes so they're attracted if you have a a a freezer that has a top compartment um and it has a liner there's something in the liner on the freezer door that they like a thing about Argentina ants and there other species of ants like this is that they lay Trail pheromone everywhere they go so Trail pheromone is a chemical that they put down on the surface where they're walking and uh an ant Will Follow That scent um so it's it's another kind of interaction with a short lag when ant puts out the chemical the other ant finds it and tends to go in that direction but they don't do what you might think is the the ordinary way that ants use Trail pheromone it's not that they go find the freezer and go back and tell everybody they're just putting down pheromone as they go so if a few ants get into the freezer because they're inspecting the liner then they're going to pull other ants in behind them because everybody's laying pheromone as they go and so you end up with a lot of dead ants in the freezer well it's not very adaptive but they haven't really evolved to deal with freezers yeah that was going to be my next question uh that's obvious L kind of I'm thinking of ants as computer programs if you'll humor me with that and this is like a bug in the software uh Evolution will have to fix it but how detrimental is it to the colony in the meantime they tend to be pretty big by the time they're getting into people's houses it's probably a pretty big Colony the part of the colony The Nest that's inside the house is only a small part of a larger Colony that in the summer consists of many nests and then in the winter they contract back into a big single one so losing a bunch of workers probably isn't going to kill the col but they've only had what I don't know 100 years to respond to freezers maybe not that much and in 130 million years that's not very long oh very good observation yeah I think most people have the general idea that ants tend to communicate via pheromones uh can you tell me the extent to which they have like developed would we even would we call it a language of scent how do you think of it PLS do produce many different chemicals so they have a lot of glands Each of which produces a different chemical and they also are covered with a layer of grease that they spread on themselves and on each other by grooming it keeps them from drying out many kinds of insects have this um layer of uh they're called cuticular hydrocarbons and they use those as well but to get to language means that there's a kind of a correspondence or even use of the chemicals as a symbol you know this means this and this means that and seems pretty clear that they're not using the chemicals as symbols instead you could say that they again have these basic rules that do link some kinds of behavior to encounters with some kinds of odors so they use these chemicals spread on their bodies to distinguish Who belongs to the same colony and who doesn't and also at least in some species we've studied the ants performing different tasks come to smell different so they can distinguish a forager from a nest worker listeners I want you to head over to brilliant.org skeptic all one word if you haven't checked it out yet you really need to look into brilliant it's Math and Science Done Right brilliant helps you learn more efficiently than lectures through interactive problem solving based courses these courses are really well paced and subdivided in this excellent way where they go lesson to lesson and you build up Concepts as you go it's the kind of thing you can do in your spare time or really commit a whole evening to either way it'll fit your schedule their courses have memorable examples and they teach in this interactive visual way listen I know probably the majority of you need to brush up on your linear algebra I want you to think about doing that with brilliant they've got a great course in that myself I just started Quantum objects yeah I know a little bit about Quantum circuits but I want to broaden that and understand more about quantum mechanics and brilliant's one of my first choices to help do that see if all the interactive quizzes and guided tutorials are right for you there are tons of courses and more being added all the time in math science and CS check it out for yourself at brilliant.org datas skeptic so if uh individual ants behavior is relatively predictable it seems like we could build good computer simulations uh does that happen to be true in practice yes we do that all the time and how do you measure how closely they describe Behavior cuz it seems to me that the colony emerges into these complex behaviors from Simple Rules and that does seem to be a challenge to sometimes simulate meaning you know it's very stochastic process you know any kind of modeling has the issue that you can construct a model that looks like the behavior you're trying to describe but there could be an infinite number of models that would give the same outcome so when you produce a model and it looks like the behavior that you see in the real world that doesn't prove that that's how the world Works still that's the first step so for example when we were looking at the reaction of the harvester ants to Encounters in their forging decisions using the Leaky integrator model what we did was to create simulations using the model based on parameters that we got from the data and then look to see if the model behaves in the way that the ants did that is can we use that model to predict the behavior of other ants and when that fits pretty well then we can say okay we have a model that describes the behavior we can't say for sure that the ants might not be doing it some other way but now we have an explanation so we use simulations for that very neat I would imagine that there aren't standard off-the-shelf ant simulation software packages can you tell me a little bit about how you manage the simulations yes uh another one that we just published um is about the trail networks of an arboreal ant species called Turtle ants that I've been studying in the trees in Mexico and uh working with collaborators at UC San Diego Saka naaka and his grad student arjin Shandra secar we just published a simulation of how the ants choose the Junctions that they take in the vegetations these ants stay in the trees they never go to the ground they can't you know fly through air so an ant as it's walking along can only go where there's a little stem or a Vine or something to walk on and at every Junction in this tangle of vegetation the ant has to decide which way to go and they use Trail pheromone like the Argentine ants they lay Trail pheromone as they go so an ant's decision about which Edge to take depends on how much pheromone has been left by the ants that came before we've been able to show in experiments with ants not in simulation that they manage of course to find things they have to find food so they can't always stay on the same Trail or they never find anything so we did some simulations with two parameters one is the rate of decay of the pheromone which sets how long at a given Junction how long an ant is likely to turn the same way as the previous ant before the pheromone which is volatile has decayed so one parameter is the decay of the pheromone and the other par parameter is the probability of exploring you could call it searching or making a mistake or exploring is the same thing the probability of taking an edge off a junction that wasn't the one that had the most pheromone so going the wrong way which is how they find things and with just these two parameters we were able to recreate the behavior of the ants in a simulation which is in a way much simpler than what the ants are doing because it's just a 2d grid a graph and the Ants make choices at the corners and still it was sufficient to show a lot of similarities to what the ants are doing in particular what we did was to look to see what would be the parameter values that would work best in the simulation for example when there's a rupture which frequently happens to the real ants a lizard runs through and it breaks a little Vine and the ants have to find another path and I have been doing experiments in which I cut the branch and see how the path recovers so we look to see what were the parameter values in the simulation that would make the ants most successful in maintaining a trail and in repairing one and those turned out to match quite well the values that the ants are actually using which we could measure so again that doesn't prove that the ants are using an algorithm like the one that we simulated but it shows that the one that we simulated is consistent with the one that the an using and now we're working to incorporate some of the variability that we see in the vegetation into the model so having got this simple model that seems to basically predict how the ants are making these Trail networks now we're going to add another layer of complexity that corresponds to what we see in the real world very neat it seems like the Simplicity of ant Behavior makes this um I guess I want to use the word rigor a little bit more rigorous than some experiments with behavior you know you could do experiments on humans but we're very complicated ants you can have a testable hypothesis and it's as you said you can readily confirm it it makes verifiable predictions and it's also falsifiable if you create a scenario where you predict one and you don't see the behavior it seems like it works like clockwork is is that how you guys explore hypothesis testing when doing ant research yes but I wouldn't say it works like clockwork because ants aren't really fully deterministic you know they don't always do the same thing in the same conditions so you have to be ready for a lot of noise and you have to have a lot of patience because if you expect the ant to do the right thing every time um you'll be disappointed gotcha so those rules are actually probabilistic then is that right yes it's all sastic yes very fascinating I know you've just gotten back from doing some field work can you tell me a little bit about that and why uh you have to go to the field why can't we do everything in a laboratory well in the laboratory we can ask ants to solve problems that we create but we can't learn about the problems that they're actually solving without seeing what they're really doing the main thing that any organism including ants have to do is to respond to the way that the environment changes so in the lab we can't create the Dynamics that ants or any other organism has evolved to deal with so I think it's sometimes we have to do things in the lab because we can't see everything in the field but in the lab ants don't do all the amazing things they do in the field and we can't see the environment that they've evolv to deal with so I love to watch ants in their natural environment because that's where all the cool things that they do are really happening yeah and I was working in in Mexico um with these arboreal ants that I was just talking about the turtle ants and what we were doing this time was to try to understand not just how they make the choices that they do but to try to characterize the topology of the network that they're searching in in order to understand why they don't go some places and do go others so we spent a lot of time mapping the nodes that they didn't take to try to characterize the shape and the connectivity of the vegetation and understand how they make their choices one of the things that was kind of impressive to me in the book was your discussion of an experiment you did to test for diplomacy because I think of that uh with you obviously very human bias here but that's to me an advanced component of intelligence it requires one to have some social instincts and think about other people and other groups of people can you talk about uh that experiment and uh what we know about the degree to which ant intelligence has advanced in the directions human intelligence has ant colonies have neighbors and they have to interact with the other colonies and this is of course an ongoing process that happens through repeated Encounters in the lab we had two different colonies and a plexiglass barrier with a with a cute little door that we could open to let ants go through we had the individuals marked so these are harvester ants they're pretty big and you can mark them by putting a colored paint on them and uh once the paint dries they don't smell it and they're fine and we can tell who's who and we wanted to know whether the same ants would be the ones that would be interacting with their neighbors this is based on some earlier work which showed that colonies respond differently to encounters with their neighbors who they meet over and over than to encounters with ants much further away and so we wanted to see if that was because particular ants were specialized in encountering their neighbors so we look to see whether ant number 72 with a blue dot on its head and a Green Dot on its abdomen would be the one that would go in cross through the the opening that we created and interact with the other colony and we found out that that was not the case and that actually LED um much later on to a model that isn't mentioned in the book because we did it since I wrote the book and this was with Fernando esponda who is a computer scientist and works in computer security we were able to model this Behavior as a distributed algorithm and so the idea is that each ant has a boundary in the space of possible odors that distinguishes the ants that it calls nestmates that it recognizes as belonging to the same colony and the ants that are not nestmates and that over time as it has encounters with other ants that boundary shifts but each ant is different each ant smells slightly different and each ant has a different boundary and so at any time an encounter between neighboring colonies depends on which ants happen to meet and whether they identify the ants that they meet as being a nestmate or not a nestmate maybe the explanation for why colonies respond differently to their neighbors than to strangers is not because certain ants know the smell of the neighbors but because the probability is high that the same ants will happen to meet the same neighbors over and over whereas those ants will never have encountered colonies from far away and so they might not respond to that smell as being a non- nestmate so it is an alternative to the idea that every ant has kind of a passport you know it has a certain smell that it identifies as the colony smell and if it meets an ant with the wrong passport it reacts instead it's more like the immune system in a mammal like in a person our immune system consists of many cells Each of which is tuned to recognize a different pathogen that it's met before that's why we have vaccines so that we will create cells that will recognize a certain pathogen and the immune response is this aggregate distributed response that depends on which cells happen to meet which pathogens it's not that every cell knows everything that's supposed to be us and everything that's supposed to be other but it's an aggregate response I'm certainly impressed by the amount of interesting work that has and advancements and things we've learned about ants and uh you know follow up on some of the bibliographic references from the book but uh as you mentioned there's 14,000 something species there's certainly a limit to what we know what are some of the interesting or questions that most interest you that are open about studying this uh I guess spe uh genus or at what level do we study ants uh they're they're a family with many Genera well I think that ants are a great opportunity to think more generally about the match between the Dynamics of the environment and the Dynamics of their behavior so we know that ants use these distributed algorithms to generate their behavior and that they can adjust to environmental change and it's an opportunity to see what kinds of algorithms work best in which kinds of changing conditions I think it's a great opportunity to make comparisons among species that live in very different kinds of conditions and ask how they operate differently and specifically to the artificial intelligence Community do you have any advice for what they can learn from studying research like yours I think the main lesson of the ANS is that you can get something that looks like intelligence without a lot of intelligence so what's really most intriguing to me about ants is how messy and noisy the system is and it still works well Deborah to wind up uh I definitely want to encourage people to go to Amazon or their book seller of choice and pick up ant encounters I can't recommend it enough where else can people follow you online and learn about your work they can go to my lab website so if you look me up um you'll get to my lab website it's hard to um spell it out but I'm Deborah M Gordon and I'm at Stanford and um it's easy to find and in the on the homepage there's a section which has some of the popular articles I've written and uh links to TED talks and um I think that'd be the best place to start wonderful I'll have some of those things in the show notes for people to follow up on as well Deborah thank you again for taking the time to come on this has been a really interesting conversation thanks very much great to talk to you