Revisiting Couzin et al. 2005

In a paper published in Nature in 2005, Iain Couzin, Jens Krause, Nigel Franks and Simon Levin modelled how information can pass between members of groups without signalling, and when members don’t know which, if any, of the other members have information. Couzin and colleagues showed that a very small proportion of informed individuals can lead to great accuracy in movement decisions, and this proportion is inversely related to the size of the group. Their model provided novel understanding of the processes underlying leadership and decision making in groups. Thirteen years after the paper was published, I spoke to Iain Couzin about motivation to do this work, how the group of authors came together, and what what we have learnt since about leadership and decision making in animal groups.

Citation: Couzin, I. D., Krause, J., Franks, N. R., & Levin, S. A. (2005). Effective leadership and decision-making in animal groups on the move. Nature, 433(7025), 513-516.

Date of interview: 31 December 2018 (via Skype)

Hari Sridhar: What was the motivation to do the work presented in this paper?

Iain Couzin: Ever since I was a kid – ever since I can remember – I’ve been interested in animal behavior, specifically in social behavior of animals. I was very fortunate, as an undergraduate, student, to work with Mike Ritchie at University of St. Andrews, not just for my thesis, but also over the summers. I worked in the Pyrenees and in a group of islands in Scotland, studying sexual selection, looking at how crickets choose mates. Long-standing theory, such as the Hamilton’s hypothesis, would suggest that individuals should go for older males because they’ve got proven genetic quality. In fact, we found the reverse, that females were strongly going for songs of very young individuals. That was, kind of, my first foray into some scientific research. But for my PhD, I really wanted to study ants. I had this great fascination for social insects, in how they can coordinate in such a sophisticated way and solve problems collectively. Looking back to when I was growing up, my dad was very influential in that interest. He was very interested in self organization and chaos theory and fractals and so forth. We had all these books at home about these topics and that really got me excited and interested in the process of self organization. There’s no one orchestrating the other individuals and telling them what to do. A first approximation would be that each individual is following relatively local rules and these scale up to the emergent properties of the colony. This, at the time, really fascinated me. One of the greatest challenges was quantifying these emergent collective behaviors. I was very fortunate to do my PhD with Nigel Franks, one of the co-authors of the paper that we’re going to discuss. One time, I was watching the colony and realized that even if I videoed this and I watched the video 300 times, for each of the 300 ants, I would still only have a very qualitative understanding of what’s going on. This was really a parallel processing system, and so I needed to be able to see what’s going on among all of the components simultaneously to even hope to understand what’s going on; and even then it’s going to be very challenging. That’s when I started getting interested in computer vision, using a computer program to see the world for us. The computer isn’t going to be limited by having to focus on one individual at a time. It can observe all of the individuals. Humans are very good at qualitatively classifying patterns; we’re very bad at quantitatively doing so. Using a computer gives us the possibility of quantifying how the individuals interacting with each other. That’s what I spent a good part of my start of my PhD doing, trying to develop software that would allow me to track the motion of hundreds of thousands of individuals simultaneously. This was in the mid 90s, so this was this was not a straightforward proposition, especially since I had absolutely no background in computer programming. We were all self-taught in the lab. And then once you have these trajectories, these movements of the animal, of course, we need to understand them. That’s what I realized and started learning about simulating behavior. I learned to create virtual organisms in the computer and simulate the types of algorithms we suspect they’re using. There are, sort of, two district directions, but it’s actually the same problem – simulating the movements and interactions and understanding what is going on. That’s really how this interest in doing these experimental tests and modeling these collective dynamics came about.

HS: Just to make sure I understood you correctly: did you say that you taught yourself the methods to extract data from the videos and analyze them during your PhD?

IC: Yeah, exactly. Within our lab, there were three or four of us PhD students. And we were all interested in self organization and complex systems, broadly speaking, and so we helped each other out. We all came from different backgrounds. My background is in biology, another student’s background was in mathematics, a guy who was in the lab was more familiar with complexity science; so, we actually taught each other. We even went to my parents’ place in Scotland for Christmas one year, where we taught ourselves the core elements of programming and the skills that we needed. You can’t learn programming in a week, but it was really where we broke the back of the problem, and then continued teaching ourselves the skills that we needed.

HS: Were you always interested in mathematics and computer programming, or did it develop because of this need during the PhD?

IC: I liked mathematics early on at school. But then, I dropped it pretty quick, because I got involved in arts and creative writing. And so, I never really had a training in mathematics, and I still don’t have any training in mathematics. Programming is a very different type of experience, and that was all self-taught during my PhD.

HS: Was your father also a scientist?

IC: My dad was a geneticist, but he also has a great interest in astronomy and physics. In fact, since he retired, he’s done a degree in physics. He was always playing around with programming on our home computer. We had a home computer with hardly any memory, but he was always tinkering with programming and playing around with this, and I was always really impressed by the sort of patterns that he was producing with very simple algorithms. I think it was a very exciting time. There was a lot of influence from Santa Fe and other places, on complexity and fractals and chaos theory, and all of these types of ideas were bubbling to the surface, even to the general public. I was fascinated.

HS: Immediately after your PhD you went on to do a postdoc in Leeds. Was that on fish behavior?

IC: Yeah, I had the opportunity to work with Jens Krause, who is an expert on fish behavior. I really wanted to be able to devote more time to do experimental work. I realized during my PhD that despite a long standing interest in phenomena such as bird flocking and fish schooling, there really wasn’t any good data to test these types of ideas and to develop further theory. So, I wanted to get some experience in the lab working with these animals – not just in the lab; we also worked in Canada in the field – to have those skills as well and integrate both into an experimental and theoretical research program.

HS: How did this paper come about?

IC: Well, during my PhD, I’d already been developing some models of collective movement, very general models that transcended the specific systems: birds or bats or fish or even herding ungulates. I’d also worked on pedestrian crowds and ant swarms. There are commonalities to the programming fundamentals underlying all these simulations: motion is very important and local interactions are very important. The computer can really be used as an extension of the brain to think through these problems. With just a verbal argument, we can’t think through how these local interactions scale up to collective properties. That really is the power of using a computer, to, sort of, expand one’s thought processes when thinking about collective dynamics. It’s a long time ago, so I can’t remember the exact specifics, but I remember being interested in how information transfers through these systems, because we really didn’t find any evidence that individuals were signaling to each other. We also found very limited evidence that individuals could recognize who is who, yet it appeared that information was circulating through these systems. I wanted to use the computer, to formalize the thought processes, to understand if and how information can be processed within the collective.  That’s what led to the development of this concept.

HS: How did this group of authors come together and what did each of bring to this piece of work?

IC: Nigel Franks, as I mentioned before, was my PhD advisor. He gave me the creative freedom to explore anything I wanted to explore, whether I it was ants or humans or fish. So really, I developed my initial interest in all these topics in his lab, and we kept in touch after I left his group. As I mentioned before, I worked with Jens Krause to expand my empirical understanding. Jens and I developed all sorts of ideas together – I’ll come back to that in a moment. Simon Levin was my postdoc advisor at Princeton University; after Leeds, I went to Princeton. Simon is mostly known as a spatial ecologist, but he has also developed some of the very early models on collective animal behavior, everything from, how groups join and split and how that results in group sizes that we see in populations, to how these migration of wildebeest herds might be the result of simple self-organized properties. So, it was a really great combination of people coming from different angles but who all find this topic very interesting. And then, Jens Krause and I, largely driven by Jens, got a fellowship from this Institute of Advanced Study at University of Bielefeld, which is acknowledged in the paper. We had this opportunity to go on, sort of, a mini sabbatical in this enriching environment where we could really isolate ourselves from everything. That’s where I presented many of these ideas to Jens and we discussed them. I was planning to submit this to Animal Behaviour. It was really Jens who saw that this could be interesting to a much broader audience. He suggested that I submit it to Nature. That is advice that has shaped my entire career.

HS: Was it just you and Jens at the Bielefeld centre, or did the authors also join you there?

IC: The other authors didn’t join us there but I always sent them drafts of the paper and discussed the concepts of the paper with them. Jens and I were the only authors who were at Bielefeld and spending a large amount of time pushing the idea forward and getting it to a publishable form.

HS: Would this have been in early 2005?

IC: No, I think the paper was published in February 2005, so I suspect this would have been in 2004.

HS: A this time, you also had an affiliation with the University of Oxford. Tell me a little bit about that.

IC: Yeah, that’s right. After working with Jens Krause for three years at Leeds, I then went to work with Simon Levin and Martin Wikelski at Princeton for a year. But, I’d also been very interested in working on locusts with Steve Simpson in Oxford. So, I also got a fellowship to work in Oxford. And so, for some years, including the year I worked on this paper, I actually had affiliations both at Princeton and Oxford, and I was splitting my time between two places for a few years.

HS: Could you give me a sense of what this work involved? Give me a rough sense of the timeline from idea to publication.

IC: Unfortunately, I’ve lost all my emails from those days. One thing I do remember, very very clearly, is that the original submission to Nature was only about leadership; it wasn’t about collective decision making at all. I remember that very much. I was working on this model for how information could be transferred without signaling and without individual recognition and finding interesting properties. These individuals are starting at random positions, at random orientations there’s no way they can infer the informational status – they can’t read the minds of other individuals to infer their information status – and yet information is transmitted effectively, and as group size increases, the proportion of individuals that need to be informed becomes smaller.  Those were the main ingredients of the initial submission. And then when he got the reviews back, I think, two of the referees made very valid points saying, yes, this is fine, this is very interesting, but what about if there is conflict? What about if some individuals want to go one way and other individuals want to go in another direction? Can this help us understand, what they considered to be a more interesting problem, that of collective decision making.  How can groups come to a consensus regarding what to do when there’s conflict. So, that part of the published paper came out of a suggestion from the referees. And we managed to incorporate it into the manuscript in relatively little time – I believe we had three weeks from the journal to respond to the comments. I was too inexperienced to realize that I could probably write to the journal and say, look, this is gonna take me a bit more time than I had anticipated. I didn’t do that. I just worked like crazy on this new problem of collective decision making. That was a great suggestion by the referee because it really enriched the paper, and made it much more engaging and  much more broadly applicable.

HS: Did you have to do fresh simulations and analysis?

IC: That’s right. I got the reviewer comments from Nature when I was in Princeton. The evolutionary biology department at Princeton was, and still is, an extremely friendly and supportive environment. I remember emailing the entire department to ask if anyone has any sort of CPU time.  This is before the time of using supercomputers. I was just running this on laptops and desktops, and I was fortunate to get 10 or 12 people who wrote back to me and said, you are welcome to use my computer. I was running around to different buildings, putting my executables on as many machines I could get hold of, and then running around collecting the data as they would come in, to get enough computational power to do the required analysis.

HS: Did the revised manuscript go through review easily?

IC: Yes.

HS: After the manuscript was ready, did you consider other journals too, or was Nature the immediate choice?

IC: Jens really felt that it deserved to be read by a broad audience. Nature was just the first one we thought we would try, and we got lucky. I, previously, had a pretty, harrowing experience with what was really my first first-author paper, which I first tried to publish in Proceedings of the Royal Society, and eventually published in the Journal of Theoretical Biology in 2002. The reviewer at Proceedings was absolutely scathing; absolutely destroyed the paper. His quote was, “I’ve seen more interesting patterns in slime molds”. I thought he really missed the point of the paper. This paper was about the emergence of group dynamics, and properties such as collecting memory and spatial sorting within groups. And because that was my first paper that I was really writing myself, I lost confidence after reading the review. I wrote to Simon Levin and said, I’ve got this paper that’s just been rejected, and he suggested that I send it to the Journal of Theoretical Biology because it has published some seminal work like Geometry of the Selfish Herd by Hamilton. So, I sent it there and it was accepted within a week. I remember I wrote to the editor – I don’t remember who it was – and said, this was my first paper, and I haven’t received any reviews, and it would be very helpful for me to see the reviews. I didn’t get a reply, and then it was just published. So, despite the initial hammering that has become one of the most cited papers in animal behavior research.

HS: Was this during your PhD?

IC: Yeah, that was mostly part of my PhD. I was very slow at getting stuff out back then. Most of the core ideas of that paper can be found in my thesis. The Nature paper came afterwards.

HS: Can we go over the names of the people you acknowledge, to get a sense of who they were and how they helped?

IC: Sure. I will try and recall to the best of my knowledge.

HS: S. Pratt

IC: Stephen Pratt is now a professor at Arizona State. I’d known Stephen for a long time; we were both together in Nigel Franks’s lab. He was a postdoc when I was a graduate student. And we both happened to be sharing an office in Princeton when I got the reviews for this paper. Steve provided lots of support and lots of ideas. Steve, I believe, suggested the graphical depiction of these bifurcation plots. I had the data and I wasn’t quite sure how best to display the data, and Steve suggested that this would be a nice way to spread to display it. He was just generally supportive of my work for many many years prior to this.

HS: D. Rubenstein

IC: Dan Rubenstein was the chair of the Department of Ecology and Evolutionary Biology. He was also extremely supportive and influential. I met with them very regularly to discuss these ideas. We still remain very good friends. He comes to visit me in Germany and we’re still working together now on zebra. He, again,  was never ending in his support and encouragement to the whole process.

HS: D. James

IC: That’s Dick James from the University of Bath. Dick is a physicist. I mentioned to you before that when I was doing my PhD I had no background in mathematics or computer science. I was very fortunate to find a physicist, a proper bona fide physicist, who was fascinated by these problems too. We would meet regularly for coffee and discuss these ideas. He would give, sort of, a physics perspective on these problems, which were extremely influential to me at that time.

HS: A. Ward

IC: That’s Ashley Ward. He is a professor at the University of Sydney now. He was also a PhD student with Jens Krause. Ashley is a fish genius.  He is the person in the world who most intuitively understands fish collective behavior. So, he is always someone I like to discuss ideas with. I’m sure we had plenty of discussions about this work back then.

HS: Did you do most of the writing for the paper?

IC: Yes, I did.

HS: Was this in Princeton?

IC: Probably, in Oxford and Princeton, maybe even a little bit in Bielefeld, although the time in Bielefeld was spent mostly running the simulations and getting the results. 

HS: At the time when the paper was published, do you remember if it attracted a lot of attention, both within academia and in the popular press?

IC: Well, not the right type of attention, I would say! We were very fortunate to also have the front cover of Nature, a beautiful cover by a photographer whose name I can’t remember, but I’m sure is acknowledged in the journal. I was in Oxford when the paper was published, and, as a postdoc, to have the cover of Nature  was a real thrill. But it turned into a bit of a nightmare, actually. The next day, I woke up and when I checked my email – I didn’t get many emails back then – but I had tons of emails, including from my co-author Simon Levin, from people all around the world, many of them contacted Simon directly, saying that there have been allegations that I plagiarized this work. This was staggering to me, and a real shock. What should have been a really exciting time for a young scientist, turned into a nightmare very quickly. These allegations were made by a scientist on a blog about swarm behavior. Now, this is a relatively small community, but, of course, if someone’s suggesting that someone plagiarized a paper, the news goes around the world very quickly. I was extremely stressed about this and didn’t know what to do; whether I should get legal help etc. Of course, I hadn’t plagiarized anything. In fact, the paper that I’d been accused of plagiarizing was in a journal sufficiently obscure that neither Oxford nor Princeton had a subscription. I couldn’t actually access this paper. And so, I wrote to the person that made the allegations, and I sent them the information from the Nature website about how you can formally make these allegations, and I said I felt it was very inappropriate to make it in the public domain.  I also said that I couldn’t comment on the similarities because I couldn’t access it, and so, if you could send me a copy of the paper, then I could comment upon this. I remember also, in the allegations, he said Couzin and co-authors cite several sources of funds, but we wrote our paper on a flight from across the Atlantic. I remember when I sent the PDF of that paper to my co-authors, one of them said that it did look like a paper written on a flight! There really was very little similarity whatsoever. But, of course this was incredibly stressful. I would also add that at no point did any of my co-authors ask me, “did you plagiarize?”, or, “have you seen this paper before?”. I always thought that that was really wonderful implicit trust. They knew that I would never have done something like that. We wrote back to the author saying we had never seen this paper previously and we felt that there was very little similarity between the two, and that the allegations were quite far-fetched. He then wrote a public apology and acknowledged that we’d never seen his paper, and so forth. But it was a really difficult time, and took time and effort to resolve.  I wonder whether, nowadays, with social media posts, whether such an issue will be resolved quickly, or perhaps will escalate even further.

HS: Did the journal get involved?

IC: No, not at all. I mean, I sent the author all the information to make a formal complaint, but, of course, that didn’t happen.

HS: In your email to me, you mentioned another response to the paper, a phone call from a major scientist warning you to stay away from this field. Would you like to say a little more about that?

IC: I don’t really want to talk about that, but I think this is important to sort of reveal how young scientists can get treated sometimes. On this occasion, a very senior scientist, whom I respected very much, called me up very early in the morning the next day after it was published, before I’d even read the flood of emails, to basically say that he considered himself to be the person who should be publishing in these types of journals,  on this topic be published in journals and I should stay away from these journals, by which I assumed he meant Nature and Science and PNAS. This was very disappointing to me because this was someone I respected very much.

HS: How much do you think this paper has influenced what you have done subsequently in your research?

IC: I think this paper being published in Nature made a huge difference to my career. I guess it got read much more broadly than it otherwise would have been. It’s connected to lots of different topics that I subsequently worked on. In fact, it is only very recently that we have been able to test these ideas. For example, we have looked at consensus decision making in wild baboons, using GPS technology, which we published a couple of years ago. So, it has had this strong influence, and I think it is because it has relevance in so many contexts, in cells, in many different animals, in decision-making in the brain. It was a topic that I didn’t directly revisit until 2011, when I published a paper looking at the role of uninformed individuals in these groups. In the Nature paper we had almost written, isn’t it amazing that they can do this despite all the uninformed individuals. It was only later that I realized that, in fact, many of these properties are observed because of the uninformed individuals. So, this paper has had a long-lasting influence.  And, it’s a topic we’re still working on, for example, in the context of how the brain makes decisions. We find similar dynamics within the brain.

HS: In addition to from the empirical extensions of the work, have you also improved upon the model itself in any way?

IC: Yes. In the 2011 paper in Science on the role of uninformed individuals, we have expanded the theory, showing that the principles extend to social networks. And we’ve also expanded this in a paper in Physical Review Letters on physical spin systems. As I mentioned before, a lot of my interest in self organization and collective dynamics came from complex systems and statistical physics. And so it is really nice and rewarding that our work can also be applied to physical systems, and there are physical principles, inspired by this work.

HS: According to Google Scholar, the paper has been cited over 2000 times. Are you surprised by the impact it’s had on the field?

IC: Yeah, extremely surprised, especially since it’s a theory paper. I was told by many people back that such a paper wouldn’t be accepted in a journal like Nature. They said you need to have some empirical work to support it. Now we have had the opportunity to find its relevance to a wide range of experimental systems from fish schools to flocking birds and even human crowds.

HS: And do you have a sense of what it mostly gets cited for?

IC: I’ve never really examined that carefully, but very broadly, everything from psychology to robotics to cell behavior to animal behavior to engineering. I think it touches upon a broad range of areas as topics, and various themes within this broad range of topics.

HS: In the last sentence of the paper you say, “The mechanism of coordination we propose here requires only limited cognitive ability, and demonstrates that individuals can respond spontaneously to those that have information. This is important to our understanding of group foraging, social learning, migration and navigation, and may provide new design protocols for information transfer among grouping robots.” Do you know whether such protocols have been developed?

IC: Yeah, absolutely. For some years, I’ve been working with control theorists developing interactions between cameras, vehicles. Several of our studies, including this, have had direct applications in robotics to allow information transfer among the robots. If you can achieve information transfer without signaling and without individual recognition, it is both more energy efficient and easier to implement in robots, than otherwise.

HS: In the methods, you say that, due to computational limitations, you restricted the group sizes to sizes of schools, flocks or herds, but not larger aggregations such as honeybee colonies. In subsequent work have you been able to overcome these limitations and work with larger group sizes?

IC: Yes. Actually, relatively shortly afterwards, in late 2007, graphics processing units – GPUs – became programmable. Working with Vishu Guttal, whom you know, we realized that implementing these models on GPUs just massively increases processing power available to us. So, instead of being able to simulate hundreds of individuals, we could simulate, literally, hundreds of thousands of them. We could also use this extra processing power, not just to look at the mechanism of interest, but also the evolution of these types of individual strategies. That was a very, very important development – the massive increase in computational power afforded by massive cheap parallel programming units. That really allowed up to explore and develop these ideas in powerful ways. 

HS: How important was this paper in your career, e.g. in getting your first job?

IC: I don’t know. I was subsequently a faculty member of Princeton, so I now have some idea about the selection process. The faculty at Princeton always pushes strongly to not care about the journal in which something is published, and focus more on the science. Even when we have a long list of faculty applicants, we read all their papers. So we actually read the science. So, I don’t think that it would have mattered so much that this was published in Nature. I don’t even think it was particularly highly cited by the time I got there. But you can never know. I’d like to think that it didn’t actually matter too much, at least for that opportunity. But I’m also realistic and realize that my CV was quite lean, so this may have had an influence. But I think it was the novelty of the work that was more important than where it was published.

HS: Have you ever read the paper after it was published?

IC: No. Not from front to back. I’ve gone back to double check a couple of technical points – what standard deviation of noise did I use and stuff like that. But no, I’ve never gone back to reread it thoroughly. I’ve perhaps read the Introduction when thinking about related things. Not wanting to read one’s own papers is, probably, fairly common.

HS: Would you count this as one of your favorite pieces of work?

IC: Yes.

HS: What do you like about it?

IC: I think I like it because, on one side, it’s so simple, and on the other side, the deeper you get into it the more rich this topic is. And, it’s a topic that is directly relevant to such a wide range of collective decisions. I think, for something so relatively simple to have such wide relevance is really appealing. It was very rewarding when we got the data from the wild baboons, tens of thousands of events of just numbers of individuals moving in different directions and the changes in angles and so forth, and we could then replicate these bifurcation diagrams with a real primate system. That was extremely rewarding, to say that this model, despite its simplicity, despite the fact that it was a deliberately toy model, deliberate simplification of reality, was able to explain the data.  This result is so strong, it’s so robust, that it does help explain a very wide range of systems in nature.

HS: What would you say to a student who is about to read this paper today? Would you guide their reading in any way? Would you point them to other papers they should read along with this? Would you add any sort of caveats to that reading?

IC: Well, I think it’ll be quite useful to get an introduction to self organization and collective dynamics.  Reading a very short letter to Nature is probably not the best introduction to the field. There are better papers in other journals for that, papers by myself and others. In all honesty, there’s nothing really I would change paper even now.  The caveat would be that uninformed individuals are playing a very interesting role that was unexpected at the time. But that’s exactly the process of doing science. Papers are always an updates on the current thinking, the way things are at that moment in time, which is then expanded by future work. The really exciting thing about science is that it’s always moving; it’s always changing.