SOLS Podcasts

Transcript - [Printable PDF format]

Peggy Coulombe: Hi, this is Peggy Coulombe with the School of Life Sciences at Arizona State University, and welcome to Science Studio. With us today is Stephen Pratt, a new young assistant professor who studies collective intelligence, specifically how decisions by individuals result in coordinated tasking in large groups. Welcome, Stephen.

Stephen Pratt: Thanks very much, pleasure to be here.

Peggy: Now when people hear the words "collective intelligence" they might think of the Borg, an interplanetary warrior race that was hardwired together and ruled by a queen in the TV show Star Trek: The Next Generation. What types of life on Earth exhibit collective intelligence, and how does fact differ from fiction?

Stephen: Well really there are loads of examples of that kind of collective intelligence. If you think of collective intelligence as being a group of organisms who are tightly integrated in order to carry out complex group behaviors like making decisions or deciding where to live or what to eat, then really there are many examples. My main interest is in the social hymenoptera, better known as the ants, bees and wasps, but there are also many other kinds of social insects. Termites form very complex societies. Actually, there are things you can easily see just walking around: flocks of birds who perform very complex kinds of aerial acrobatics, or fish schools who do similar things in the water. There are sort of more obscure things, collective groups of bacteria, for example, who do remarkably--this is increasingly things that have been found in recent years--carry out remarkably sophisticated group behaviors, where they sort of monitor one another's presence and decide whether or not, for example, to form a complex structure or, to our detriment, to turn on some kind of virulent behavior, and do lots of other things like that as well.

But as I say, my main interest is in the social hymenoptera, mainly ants and bees.

Peggy: So tell me some characteristics that makes an insect a social insect.

Stephen: Well, not to get too technical, but if we think of these ants as, if we use sort of the proper term, and use sociology, really we're talking about a few key features, the most important of which is a reproductive division of labor. So you have a certain minority of the colony members who do all the reproducing. They're the ones who have offspring, who go on to found new colonies. The majority of the individuals within the colony don't reproduce at all, they just help the reproductives--the reproductive queen, typically--produce her own offspring.

But the other thing we have is a sort of overlap of generations. So workers will stick around to help raise their sisters and brothers to become more workers or, for the lucky few reproductives, to found their own colonies. So it is this division of labor, really, that creates the opportunity for the kinds of collective intelligence or complex collective behavior that I'm interested in.

There's much behind the division of labor among all these workers, each of them performing specialized tasks, communicating with each other. Then what you can have is these very elaborate sort of group activities, where for example, hundreds or thousands or even millions of individual workers are exploring an area for food items, finding out which ones are the best ones, coordinating with each other to allocate this foraging force to work to collect only the best of the resources available.

Peggy: OK, I'm going to ask you a little more about that process in a minute, but I'm curious, how did you start working with social insects? Was it when you were an undergraduate at Harvard?

Stephen: In fact it was, yes. I had the good fortune to be there when both E.O. Wilson and Bert Holldobler, who is now at ASU, were on the faculty there, and they had a very, very active, very lively group, and they were open to having undergraduates work there who were sufficiently motivated. So I spent two or three years there, working with a lot of different people, mainly guided by Bert Holldobler and students in his lab.

Peggy: He's kind of like your academic grandfather.

Stephen: Yeah, well in a way he is both my academic grandfather and my academic father, because he was the first mentor I had as an undergrad, but then in graduate school I went on to Cornell, and my Ph.D. supervisor was Tom Seeley, who was himself a Ph.D. student of Bert Holldobler's, so there are a couple of lines of connection there.

Peggy: And was the relationship with Bert Holldobler one of the things that brought you to ASU?

Stephen: Well actually, yes it is. I mean this is the thing about academia, I think. If you're smart you form kind of long-term relationships with the people who help you along and teach you. Bert was kind enough to let me know of this opportunity here at ASU when he and Rob Page and others were forming this exciting new social insect research group here, so I've been fortunate enough to be able to come along and join it.

Peggy: So tell me a little bit about your study animal--and I'm probably going to butcher the name here, but I'm going to give it a shot. It's Temnothorax curvispinoscis?

Stephen: That's almost right, curvispinoscus--but yeah, it's a mouthful. Yeah, these are the ants I've been working on for oh, the past I guess six or seven years now, at least.

Peggy: And what makes this particular study animal special for the work that you do?

Stephen: Well they're great animals, partly for reasons that probably mean that most people have never seen them, and that is that they have very small colonies--there are only maybe 100 or 200 ants in a typical colony--and so they can be quite obscure out in the wild, where they live. But it means when you bring them into the lab, where I do most of my work, then you have kind of a complex society in miniature, and there aren't so many parts in there, so many individuals, that it becomes hard to track them all as they carry out some complex task. So it makes it a lot easier for me to describe in detail what they do when they're making a decision or doing something else interesting.

Peggy: You mentioned tracking. How is it that you could track the actions of individual ants, say, even 300 of them. It's almost impossible to watch even one of them.

Stephen: It would be, yeah, exactly. And the first time, as an undergrad, the first time I did experiments I did where I was watching ants, I was just looking through a microscope at a bunch of ants in an artificial nest. And it can be very trying, very difficult to follow the behavior of any individual, even just if you keep your eyes fixed on them. If you want to describe, as I do, what all of them are doing simultaneously, then the only way that works is if you mark them individually, and this is something that is appreciate by a lot of people that have studied social insects. So there are a lot of techniques out there--everybody's got their favorite one--for how to individually identify an ant or bee or wasp in a colony, so you can, over time, track everything she does.

Peggy: My understanding is that you've developed an ant harness?

Stephen: Yeah, that sounds a little strange, but these ants, Temnothorax, I learned how to mark them when I did a post-doc in the laboratory of Nigel Franks, then at the University of Bath. The basic trick there is you've got to knock them out, got to immobilize them, and you can do that with CO₂, that will put them under for a few minutes. But then you still have to fix them down, and the way I do this now is, I take the ant and I put it on top of a small cosmetic sponge with just a single hair taped across the top. Then you can just kind of slip the ant underneath this little harness, and that will hold it still enough for the minute or so you need to mark it. You can then bring a single-bristle of a paint brush up and put, say, three or four drops of distinctively-colored paint on the ant, and then you're good to go.

Peggy: And you let the ant loose, and you give it a challenge, and you videotape?

Stephen: Well exactly. So let's say you've spent a few hours taking 100 ants or 150 ants, marking them all. It's a bit of a trauma for them, you give them a few days to survive, and then yes, you have this colony of marked ants, it's a very useful resource. And then, exactly, the thing then is to give them some kind of interesting problem to solve. What I work on, principally, is how does a colony of these ants choose a new home?

Peggy: So this is, for example, out in the wild. The colony is living in an acorn, for example, and somebody comes along and steps on it.

Stephen: Exactly, yeah. So these ants, Temnothorax curvispinoscus, that's an ant that lives in Eastern North America, in the forests, and so if you want to collect them you'd normally go out to the woods and look for old acorns, hollow acorns. That's a favorite nest. Sometimes it might be a hickory nut or a hollow twig. So that's where they happily reside, but you can imagine, if you live in something like that, as you say, lots of things can go wrong. A deer can step on you; the thing can just rot; you might get pushed out by other ants--it can be a very, sort of, competitive world there, the world of ants. So at that point a colony would need to explore, say, an area of several square meters, and find a new home.

Not only that, it is likely to find a lot of places there that might be suitable homes, like other acorns and hollow twigs, etc., and they've got, ideally, to figure out which one of those is the best and move into that best one. And also, ideally, prevent the colony from splitting and, say, some of them move to this acorn and some of them move to that acorn. They really want everyone to end up at home, and that can be a really big challenge for little ants.

Peggy: How is it that a small ant can contribute to a group decision?

Stephen: Well this is the real, the sort of crux of the kinds of work I try to do, the questions I try to answer. And I think it's important to stress something here--actually, it goes back to your first question about the Borg. One thing about--I mean, I like the Borg, I think it's a nice example; they really do a nice job in many ways, with that, of giving some of the feel of what it must be like to be part of these kinds of highly decentralized collectives. But even there, and partly I think it's for the sake of drama, but partly just because we have such trouble getting our heads around this idea, they have to kind of cheat. So they have a kind of queen who functions as almost like the chief--what that word "queen" would mean to us under human circumstances, someone who is actually in charge, someone who knows what's going on, someone who makes the decisions.

Ant colonies have queens, but they don't play that role at all. There is nobody, there are no individual ants in the colony, not the queen, not any of the workers, nobody who acts as a sort of boss and tells the other ants what to do. So instead, individual ants have to contribute to a decision, not by, say, reporting to a boss and taking orders. They do it in a highly decentralized way. They gather a little bit of information on their own, they share that information with some of their nest mates by specialized communication signals, and they use sort of specialized decision rules of how to act on all this information, locally, in such a way that lots and lots of ants all following similar rules will sort of generate a good decision as an emergent property.

And that is actually a very hard sort of thing, I think, to understand in an intuitive way. I think what it really needs to do is get into the details, and so then a lot of what I do involves watching what these ants do as they, say, pick a nest; describing what each of these individually marked ants did in the course of this decision, and then trying to reconstruct what happens.

So to give an example of what I see when we look at that kind of thing, we see that an ant whose nest has been damaged will just go off and explore her surrounding forest until she sees a likely-looking nest. She'll assess it, she'll come up to a judgment of how good she thinks it is, and then at some point she'll begin to recruit other nest mates to visit it. But it turns out that she takes that assessment, that initial assessment of how good she thinks it is, and she uses that to tell her just how rapidly she should start to bring nest mates over there. So if she doesn't think it's very good, she'll take her time.

Peggy: And what kind of assessments are we talking about?

Stephen: You mean, what makes a good home?

Peggy: What makes a good home?

Stephen: Yeah, from an ant's point of view, it's all sort of, in a way, common sensical. I mean, they want a spacious home--most of us do--but they want it to have a small entrance, and not too many entrances. Ideally just one entrance, because again, they partly live in these nests to defend themselves against predators. They want it to be dark in there. They want it to be humid, but not too humid. They want, if it's a crevice, they want to be not too tight--you know, enough room to move around in. And probably a lot of other things that we haven't really come to grips with yet. But all these features they seem to be quite picky about.

Peggy: So now your individual ant has assessed this site, and let's say they think it's terrific. What happens then?

Stephen: Well that's when this recruitment begins; and if she thinks it's really good, then she may spend just a few minutes assessing the nest, visiting it a few times, and then go back and immediately start to try to get her other scout ants from her colony to come and visit the nest, too, and generate their own opinion of it.

So to do that, they use this very odd-looking behavior called "tandem running." What they'll do, in essence, is that first scout who found the nest and liked it will lead a single other ant in a sort of, as a little pair, from the old nest all the way to the new nest. If you watch them it's a very distinctive behavior, where there's lots of stopping and starting, because it's hard for them to remain in contact as they do this.

But over time, very slowly, that initial discoverer will be able to get one of her nest mates to come and see the nest, the potential new home, make her own assessment of it, and then if she likes it, she too may begin to lead these tandem runs.

Peggy: One of the things that I find confusing is, you have an ant that finds a great place, that runs back and finds another ant, they form a tandem. Isn't this happening all over the nest? How does a group decision evolve out of all these individual bits of excitement, and how are those things ranked?

Stephen: Yeah, that's a very good question. I mean, essentially, you can imagine what's happening is, there are two or three or more of these potential new homes, each one of them is being visited by a different scout ant, and maybe there's one that's particularly good, and so the ants who find that are really rapidly going and leading these tandem runs and bringing new ants to visit it. So let's say that's Nest A. There's Nest B over here that maybe isn't so good, but still, somebody found it, and after a while, unless it's really bad, she'll begin also to lead tandem runs. So you'll end up with these ants who really don't know anything about what one another is doing. You've got ants visiting A, you've got ants visiting B.

So what the colony needs is some way of increasing the likelihood that ultimately the colony is going to move to A, the better one, and not to B. So what we find is, one way they have of doing this is they actually pay attention to how many other ants are at the site that they are thinking about. So over there at Nest A, where there are lots of ants being relatively rapidly brought over, the population is going to build pretty fast, lots of scouts will end up there. At B it will be building slowly because everyone is a little hesitant about starting to recruit and doing these tandem runs.

So that means that A will reach a threshold, what we refer to as a quorum, and once these ants realize that that threshold has been met, then they really go into high gear, and they stop doing these tandem runs. They just go back to the old nest and they just start picking up nest mates and carrying them over to the site that they've been visiting. And that goes much faster. And so what you can have happening in a case like this is that that superior nest will experience, reach the quorum, and then everyone will rapidly carry all of the colony's other members to that site before the not-so-good nest, Nest B, has managed to reach the quorum. Then you just have the relatively number of scouts who were thinking about that site, who then have to find their way over to Nest A.

Peggy: Just mop 'em up.

Stephen: Yeah, exactly, there's a little bit of a mopping-up operation to this.

Peggy: I understand that you create algorithms, sort of an ant map, to explain these choices and these behaviors.

Stephen: Well that's, yeah, that's right. Because the thing is, again, it all comes down to the difficulty in intuitively understanding this kind of process. It's very different from the way we organize ourselves most of the time. So we come up with these descriptions, as detailed descriptions as we can, of what the ants are doing; but if we really wanted to be able to say, "Well, are these rules I've been describing? Are they really enough to explain why the colony is able to pick the best nest and move into it unanimously?"

The only way to really know that is to essentially make a simulation, make a virtual ant colony in the computer. So what we try to do is come to as detailed a description as possible of the algorithm that each individual ant is following. I've described a couple of parts of it, in terms of these quorum rules or the tandem running initiation rites, and there are other ingredients too, large and small. We put as many of these as we can into a detailed computational description. Then we simulate a colony of 100 or 150 or 200 ants, following these rules, and we see whether the simulated colony behaves the same way as the real colony does, and that gives us an idea of, have we really nailed down how it is that the ants are able to carry out the collective intelligence, if you will.

Peggy: How can finding out about things like colony intelligence be applied to solve practical problems?

Stephen: Well it's interesting. Obviously, the ants finding a home is a practical problem from the ants' point of view, but obviously it's not much applied value in an obvious sense to humans. But it turns out that a lot can be learned from these kinds of algorithms that the ants are following.

I can particularly point, not to the problem I've been describing now, but to how ants use chemical trails to select food sources. That is probably something everyone has seen, ants walking in trails, and those trails actually consist of chemical signals laid down by other ants who've found, say, a good food source, and are trying to get their nest mates to visit it. It turns out that in kind of a similar way to what I've been describing, a lot of ant species can use these trails to select the best of, say, multiple or many different food sources, and get all the workers to go and eat there.

That, the details of that kind of recruitment communication by trail pheromones, has been adopted as a kind of model by a school of computer programmers. They've generated this sort of approach to computer optimization called "ant algorithms, " and essentially what they do is create programs that mimic the behavior of ants. The programs essentially have sub-units that are like individual ants, who explore a problem space to the solution for some kind of practical problem. Like for example, scheduling information, or rather, scheduling the transmission of information packets on a mobile phone network, or scheduling the delivery of, or a delivery route for a trucking company.

These are really rather difficult optimization problems, and what they do is have these kinds of ant algorithms simultaneously explore lots of possible solutions, and rate each one into how well it seems to be performing, and then if it's performing well, they'll recruit other ones of these sort of software ants to explore that solution further and maybe try to improve upon it. And this is proving to be a quite efficient way to solve problems where it is totally impossible to consider every possible solution; these are problems where there are just too many possible solutions, but this gives a practical way to efficiently explore part of that, or the most promising part of that solution space, and come up with an optimal and near-optimal answer.

Peggy: So it could be something like the scheduling of a Fedex truck in New York.

Stephen: Exactly, yes. Or garbage pickup in any city.

Peggy: Hmm. What other projects are you working on?

Stephen: Well lately I have gotten interested in applying some notions about decision making in general, from human psychology, to trying to understand better how the ants are making decisions. So, I mean in general what I have been talking about is complex collective behavior, but more precisely in terms of, for example, the house hunting case, I am interested in how these colonies make decisions. Of course, we make decisions all the time, and there is a very rich tradition in literature of studying how humans make decisions.

One particularly interesting aspect of this concerns whether or not humans make rational decisions--rational in the sense that they consistently rank things according to a consistent scale--and this would apply, among other things, that decision makers should show transitive preferences. So if somebody says they like Option A over Option B--you know, any particular decision--and they like Option B over Option C, if they are only given those two to choose from, then if you give this person just Option A and Option C, then logically, rationally, they should prefer A over C. It turns out, though, that there are lots of cases where humans don't do that, and they have to be rather carefully set up by psychological experimenters, because if people realize, if I put the problem as blankly as I just did and people realize that they really ought to prefer A over C in that case.

But very often in real-life decision-making situations, in particular where it is very, very hard to say which option is better, maybe because they vary in multiple attributes--you know, maybe you're choosing among jobs, and one job is in a city where you want to live but doesn't pay very well, and the other job is in a not-so-nice city but pays really well, how do you make the decision, how do you balance these different attributes? When people face tough decisions like that, they very often make irrational choices, intransitive choices.

Peggy: Like flipping a coin.

Stephen: Well, you know, they may come up with lots of rules like this. Sometimes what people will do is, they'll come up with some kind of rule of thumb that is, say, relatively easy to apply, but doesn't necessarily lead you to consistently make the best choice. So I've gotten interested lately in trying to see if ants are similar. There is some work that says that individual animals, including bees and hummingbirds and jays and other individuals, will show the same kind of irrational decision-making humans do if you give them the right kind of difficult circumstances like I've been describing.

What I'm interested now is, will ant colonies do the same thing? If you ask a colony to make a decision and, say I have my ants and they are choosing between different kinds of nest sites, and maybe one nest site has a nice spacious interior but a kind of overlarge nest entrance, and another one has a really nice small entrance, but it's a little bit cramped in there, how do they balance those things? And if they are forced to balance them, do we see colonies also showing this sort of intransitive behavior or other forms of irrationality? Hopefully, what I'm hoping this will be is a way to probe more deeply, what these kinds of decision rules and algorithms are that colonies use in order to solve difficult decision-making problems.

Peggy: I find it kind of comforting to think of ants as being somewhat irrational.

Stephen: Indeed. Well, because it's always possible that they'll prove to be more irrational than we are, and that could be a little bit of an unpleasant surprise.

Peggy: I wanted to bring up one other thing. We talked about the painting on the ants, and you in fact got a photography prize in the Art of Science competition held at Princeton University, for one of your photographs.

Stephen: Yes, that was an interesting kind of event. I mean, as I was saying earlier, anybody who studies social insects--or as many people as study social insects--will have some way of marking them individually, and painting them is really just a practical tool in order to make it possible to do these experiments. But when you paint, you take a whole nest of these ants and you paint them all these bright colors, and then look at them all sitting there in, say, a glass-walled nest, it really is quite a beautiful image.

So I took pictures of them, mainly just to illustrate to people at scientific talks how this method works and what the value of it is, but it really makes quite an attractive image, and so I saw a poster on the campus at Princeton University that said, it had a catchy title. It said, "Art is Stupid, Science is Boring. Prove Us Wrong." They get quite a good turnout, people sending images in like the one I did, and all kinds of other things from people's scientific work. It was a really interesting show, a lot of great things came out of it, and I was lucky enough to have gotten one of the prizes for it.

Peggy: And I understand they also published it in Wired magazine.

Stephen: Yeah, it got a little bit of attention, indeed, since I guess it's sort of at the interface between a striking image and what I like to think of as an interesting topic, and that is how these collectives manage to be smart.

Peggy: Well Stephen, I want to thank you so much for joining us today.

Stephen: Oh, my pleasure.

Peggy: For those of you who'd like to see some of the colorful photography by Stephen, and would like to read more about his research, he is featured in the Spring 2007 School of Life Sciences newsletter, and his photograph is on the cover.

This is Peggy Coulombe and you've been listening to School of Life Sciences podcast Science Studio. School of Life Sciences is in the College of Liberal Arts and Sciences on the Tempe campus of Arizona State University.

Collective intelligence, science fiction or science fact?

Transcript - [Printable PDF format]

SOLS Podcasts 2007