Collaboration and the Evolution of Disciplines

Collaboration and the Evolution of Disciplines

Robert Axelrod [7.1.19]

The questions that I’ve been interested in more recently are about collaboration and what can make it succeed, also about the evolution of disciplines themselves. The part of collaboration that is well understood is that if a team has a diversity of tools and backgrounds available to them—they come from different cultures, they come from different knowledge sets—then that allows them to search a space and come up with solutions more effectively. Diversity is very good for teamwork, but the problem is that there are clearly barriers to people from diverse backgrounds working together. That part of it is not well understood. The way people usually talk about it is that they have to learn each other’s language and each other’s terminology. So, if you talk to somebody from a different field, they’re likely to use a different word for the same concept.

ROBERT AXELROD, Walgreen Professor for the Study of Human Understanding at the University of Michigan, is best known for his interdisciplinary work on the evolution of cooperation. He is author of The Evolution of Cooperation. Robert Axelrod's Edge Bio Page


COLLABORATION AND THE EVOLUTION OF DISCIPLINES

ROBERT AXELROD: Let me start with what’s new in the world of cooperation. There's the problem of international relations in which an established power, the United States, is dealing with a rising power, China. The ancient Greek historian Thucydides said that the reason why Athens and Sparta fought was because Athens was a rising power and Sparta was the established power and they couldn’t work it out. More recently, Graham Allison at Harvard looked at the last 500 years for all the cases in which an established power was dealing with a rising power. He found sixteen of them, twelve of which led to war. Those are not good odds.

One of the ways of dealing with this is to try to develop norms and rules of the road for understanding what’s proper behavior. I’m working with Chinese and American delegations who are meeting regularly to discuss things like cyber conflict. For example, if cyber weapons were used on a large scale, it looks unstable in a way that nuclear weapons are not unstable. So, we’re dealing with how to develop norms for understanding cyber tools and cyber weapons. That’s one area where cooperation is important.

Another area, which you’re all very familiar with, is the decline of democratic norms not only in the United States but in many other countries, especially in Europe, where the basis for societal cooperation in a sense of governance are deteriorating. A third area is climate change, where one could certainly look at this as a technical problem. I hope technical progress can be made, but it’s also a collective action problem of getting large numbers of actors to work together.

Interdisciplinarity is another area where cooperation is needed and is not trivial to attain. The research on what makes interdisciplinarity succeed when it does and what its characteristics are has exploded in the last ten years, in part because of the ability to do large-scale analysis of things like citations and see whether people who publish articles together from different disciplines are more successful in, say, achieving citations.

There are a few things that are known. One is that interdisciplinary research has higher variance. It’s not a higher average of success, but it’s higher variance, so sometimes it does very well and many times it does not so well. So, it’s not necessarily that more interdisciplinary work is better. If we could understand better barriers to make it work, we could maybe change that.

Another finding is about preferential attachment. The idea is if you work with somebody, you’re likely to work with them again, and maybe even second-order, where you'll work with people that worked with them again. Another result is that you can map the disciplines in two dimensions such that distance represents the probability of collaboration; for example, you might have a lot more collaboration between, say, physics and chemistry than you would between physics and sociology. That’s not surprising, and the maps look plausible and reasonably stable. But that’s not a lot of knowledge about interdisciplinarity.

The questions that I’ve been interested in more recently are about collaboration and what can make it succeed, also about the evolution of disciplines themselves. The part of collaboration that is well understood is that if a team has a diversity of tools and backgrounds available to them—they come from different cultures, they come from different knowledge sets—then that allows them to search a space and come up with solutions more effectively. Diversity is very good for teamwork, but the problem is that there are clearly barriers to people from diverse backgrounds working together. That part of it is not well understood. The way people usually talk about it is that they have to learn each other’s language and each other’s terminology. So, if you talk to somebody from a different field, they’re likely to use a different word for the same concept. That also comes up with Americans talking to Chinese about military things. That seems to me just part of it.

Another part of it is whether they have common goals. For example, if there are two different disciplines, the researchers might want to publish in journals from their own discipline so that their own peer group will recognize the contribution, and that could be a conflict of interest between them that they need to work out.

The other problem is that what they come up with in a collaborative interdisciplinary activity may not be recognized as a contribution by any field, and this is especially true when there are new fields. One of the things, though, that does make the interdisciplinary activities that I’ve been involved with work is having some tools in common. For example, game theory is understood as valuable and taught in much of the social sciences and the biological sciences, so being able to collaborate with someone who knows game theory gives us a chance to make progress.

Ian mentioned civil war in a body as an interesting aspect. I saw an agent-based simulation of a growing cancer where the agents were the cells. I asked the computer scientist that developed that with a student, "What are the premises that go into that simulation? What are the mechanisms?" They pointed me to an article on the hallmarks of cancer, and it turns out that there are about eight different defenses that the human body has to keep cells in line from becoming selfish and asking for more resources than is good for the host. The common understanding was that a single cell line develops the mutations necessary to overcome each of those defenses, but when I saw the simulation and read about the mechanisms, I realized that that wasn’t necessary.

Let me give you an analogy: If you have two thieves robbing a house and one of them knows how to turn off the alarm and the other one knows how to pick the lock, they don’t both have to know how to overcome the defenses as long as they’re traveling together. In cancer, some of the defenses are overcome by putting out a certain chemical saying, "Build a capillary in my direction," basically asking for more blood and more oxygen, but another cell nearby might be exceeding its normal capacity to do something which would overcome another defense. So, as long as they’re together, you don’t need a single cell line. That would help give you another channel for therapy, which would be to interrupt the cooperation in the cell line.

I went to a geneticist and an oncologist, and we worked out some of the implications of this. First of all, we found that it hadn’t ever been explicitly stated and, secondly, it was biologically plausible, so we wrote this up. So, here was collaboration between a political scientist, an oncologist, and a geneticist. When we proposed this speculation about how things might be, we got two reviews. One of them said what we were proposing was impossible, and the other one said what we were proposing everybody knows. Anybody had a pair of reviews as challenging as that? Obviously, we didn’t explain ourselves very well, so we picked ourselves up off the floor, rewrote it, and tried to explain why it’s neither impossible nor the same thing that people knew.

What provided the basis for the collaboration is that I was looking at this from a social science perspective of community action and cooperation, and the others had the competence about how cancer works. We were working on a known problem, which is, for example, how does civil war in your body get under control and how does it lose control? One of the opportunities for success is if there’s already a known problem and then you provide another way of attacking it or making progress on it. As long as the problem is accepted in at least one discipline, then it seems to me you could use any tools and any new concepts as long as you could make progress in the terms that the people that care about that problem understand. If it’s a problem that they don’t realize they have, it’s much harder.

Let me talk about the evolution of disciplines, which is one way to think about this. One way to approach disciplines is to see them as an ethnic group or a language group where the people within a discipline are able to talk to each other well. This is because the disciplines have become institutionalized so that anybody that calls themselves an economist or geneticist knows a whole bunch of stuff that almost every other economist and every other geneticist would know. So, they can talk to each other in the areas that the discipline has defined as building on their canon. That’s fine, but it’s like a gravity model in a sense that then the disciplines become more and more coherent over time, and that makes it easier. Then there’s a body of concepts, and terminology, and science, and previous experiments that are shared, and that makes that kind of collaboration easier. But there’s another group over here that has coalesced in a different place in this high dimensional space. As they each coalesce, they become further apart with fewer people in between.

Another analogy might be like Spanish, French, and German. There used to be a whole series of dialects that are more or less continuous across that space. Eventually, those three countries established the canonical way of saying German, French, and Spanish and taught it in the schools, which was very useful in the Industrial Revolution when you wanted people from a distance to be able to deal with each other. Then it wiped out most of the stuff in between, Catalonia being a surviving exception.

In disciplines we’ve converged, the convergence is not just on subject matter; it’s incredibly well institutionalized so that departments not only represent disciplines like physics and economics, they also control careers. They decide whom to hire and, therefore, professionals have a strong need to be attractive to at least one of those disciplines. Not only that, but they control the entry. They control the training process to determine what it takes to get a PhD in X or Y. They also control, to some extent, the journals and the major professional conferences. They don’t control the smaller journals or smaller conferences. So, when a group like us gets together with different backgrounds and tries to communicate, there are several questions about whether there’s an emerging discipline of brain intelligence and neuro and cognitive psychology, because all those disciplines are so well established and institutionalized. It’s not easy.

It’s easier to get a center going perhaps, but it’s very hard to then get the established groups to give tenure lines, faculty, resources, course credits, and the ability to grant the PhD under their label. The way these things have coalesced to some extent is accidental, though not completely because there is a difference between what chemists study and what physicists study—in the matter of scale, for example. In other fields it’s not as obvious where the boundaries would be if you started over again or, more to the point, where they should be now.

You can’t erase the boundaries and just redraw them. Several places have tried that. Carnegie Mellon and Irvine are famous for having redrawn boundaries, and you can see that the problems they have include the fact that they can’t develop a cascade of reorganization across the academic community. So, they're at a disadvantage, say, on whether their PhDs are hirable and whether the cluster of things that they teach in one of their structures doesn’t correspond to what anybody else does.

The evolution of disciplines seems to take several forms. One is the splitting off of a single discipline into several disciplines or usually one. Maybe astronomy is on the edge of being separated from physics in some places and not others. Clinical psychology is quite different from developmental or cognitive psychology, but they’re still holding together.

Sometimes a new discipline can arise from the territory between. Biochemistry in some places is a new discipline. One of the constraints that helped define this is how much can a PhD candidate learn in five years? They can learn a set of tools, concepts, and experiments. When a single discipline is in the situation where some of it takes five years to learn and other parts take a different set of five years, then it’s pretty ripe for separating those things out and giving them different names and then having fission, and that’s certainly one way that it happens.

Another way, though, is more typical of this room, which is where people from many different disciplines are working on some problem area, like questioning what intelligence is and how the mind works, and how can we accomplish more effective AI and what would it mean to do that. So, we can gather together in this room and try to understand each other, which is certainly a significant task that can be promoted by having repeated meetings of largely overlapping people, but it’s hard to buck the established institutional frameworks.

Caroline mentioned just before lunch the topic of immunology and how the immune system has a kind of intelligence. Let me give you another collaborative example dealing with the immune system. I had worked with evolutionary biologist Bill Hamilton on evolution of cooperation in biological systems. A couple years later, he came to me and he had a theory of the origin of sex, and the theory was that it’s an adaptation to resist parasites. That seems very strange, and it goes like this: Parasites have an incentive, a biological selection pressure, to look as much like you as they can. If they look like you, your immune system will not identify them as non-self. You can imagine a high-dimensional space, basically the antigens in which you’re located here and the parasites can evolve to become more and more similar to that and eventually get to the point where you don’t recognize them as foreign. They have an advantage because they can reproduce perhaps 100 or more times faster than you can. So, they can outrace you as you run away from them. When I mean you, I mean your progeny over generations.

Bill Hamilton’s idea of what sex does for you says there’s one adult here and one adult here, and they’re quite different in how they present themselves to their immune systems and to parasites. If you could take some of the genes from this one and some of the genes from that one, you’ve made a huge jump in this high-dimensional space. You haven’t just moved incrementally. If you had asexual reproduction, your children would be very much like you, but if it’s sexual, then from the point of view of the antigen you’re very different. Therefore, sex is an adaptation to resist parasites. The problem it has to account for is that only half of the adults have offspring. This is a tremendous biological disadvantage. It could be up to two for one. That’s a lot to overcome, so there’s got to be some powerful things on the other side to show that at least it’s plausible. He said he tried to model this as being explicit about the three characteristics of your adult, and then if you have three others from the other adult and you mix those and then model that, but he couldn’t do the math after about three. It doesn’t work for three.

I learned about the genetic algorithm from John Holland where you can have long strings of chromosome simulation. "Seventy is no sweat," he said, "that’s just what I need." So we did some simulations, which were enough to make the search problem hard or to make sex valuable.

The problem of why we have sex is a well understood problem. It’s well understood that that’s a serious problem in Darwinian theory because the two for one disadvantage is so great. There's another explanation for why sex is the answer, but this one looks pretty cool. This allowed me to take something from computer science search techniques and adapt it for a simulation of an evolutionary biology technique.

We are now faced with the question of intelligent AI systems, and that is a lot like disciplinarity. The humans have some set of concepts, and the artificial intelligence system will have another set of tools, concepts, ways of organizing the world, and thinking. How can we promote the effective collaboration of humans and intelligent systems? Then the other question is, how do you guys do it, and what is your experience with effective collaboration across disciplines?

* * * *

STEPHEN WOLFRAM: Has anybody made a giant wall chart of the evolution of disciplines over the last few hundred years?

AXELROD: I’ve looked for that, too. No. There are some histories of universities that have been around a long time, like the University of Padua, which separated philosophy from law. But I haven’t seen it more generic. That would be interesting to do.

NEIL GERSHENFELD: Stephen, could you derive it from all the data you’ve ingested?

WOLFRAM: I was wondering. The Web of Science, through which you get the Science Citation Index is hard to get access to. There’s now this open citation project where journals are contributing their citation metadata, which started maybe two years ago or something and it’s gathering steam, but that data is mostly fairly recent data.

GERSHENFELD: Erez Lieberman Google Book data, and he did a surprisingly good job of deriving history.

WOLFRAM: In academia, there isn’t management of research. In a company, there is management of research. It’s interesting what the tradeoffs are between managed research and unmanaged research.

CAROLINE JONES: Historians like to see the pattern of disciplines as infinite proliferation. So, the phrase "renaissance man" was invented in the 17th and 18th century to describe a lost nostalgic moment of wholeness, when you possessed all disciplines in one person. It was already acknowledging that there were divisions happening.

PETER GALISON: What do you see as the main difference?

WOLFRAM: For my company, and I can’t say this very scientifically, but over thirty-something years we’ve developed the concept that we’re going to put together these teams of people with different expertise and they will work together. In earlier times, that was hard to achieve, but we finally got to the point where, culturally, we expected that people from different backgrounds would work together in teams. I don’t know whether that happens in universities as effectively. It’s something that took a long time for us to achieve.

DAVID CHALMERS: I don’t remember who it was who said the best interdisciplinary conversations take place inside one person’s head.

IAN MCEWAN: I attended a university in which the vice chancellor’s project in the early ‘60s days of great optimism in Great Britain was to redraw the map of learning. Interestingly enough, every student of the humanities was required to read three books. One was Turner’s Thesis on the expansion of the American West, one is Jacob Burckhardt’s Civilization of the Renaissance, and the other was Tawney’s Religion and the Rise of Capitalism. This was based on the understanding that you could not approach the humanities without a background in historiography, the nature of history, and the changing ways in which history is studied. That whole project lasted about fifteen years and then was swept away. Now, when I go back to my old university, there’s the History Department and the English Department—a kind of inertia dragged it back.

GEORGE DYSON: You already gave the answer to your question, or Hamilton did in his beautiful explanation for the origin of sex, which was the same reason you should have these interdisciplinary things because it allows you to outrun the parasites who build up in the History Department.

SETH LLOYD: Sex is the ultimate interdisciplinary act.

JONES: It speaks to the internal problem of errors in replication that accumulate without hybridization. Self-replication without hybridization risks a lot of errors and repetition of errors and accumulation of errors.

WOLFRAM: You see a discipline in its first generation, the people who founded the discipline are still around; they still know what the fundamental questions are; they’re still often a bit insecure about the foundational things that the discipline is based on. Then you get to more generations, and by the time you’re at third generation of people, typically, they don’t even discuss the foundations. It is just assumed.

If you look at the period of maximum fertility, maximum lasting effect of a discipline, is it the case that most of it is in the first ten years? Is it in the first twenty-five years? Is it in the first generation? Should disciplines be euthanized after they’ve gone through five generations, for example? They basically won’t produce significant output by the time they’ve gone through five generations of people.

AXELROD: I doubt that. It seems to me that’s like saying, should we get rid of some culture because it’s been around a long time and it’s worked as much out as it can. That seems silly.

WOLFRAM: That’s an extreme version, but my question is, at what point in the curve? It’s true with conferences, for example.

AXELROD: Paradigms can get mature and stale, but disciplines can change paradigms, and that’s a possible form of regrowth. Let me give you an interesting example of biology where the disciplines of biology have been now reorganized to turn 90 degrees. It used to be botany and zoology—plants and animals. Now it’s skin in and skin out. Skin out is ecology and evolution and skin in is microbiology, so they just turned the whole thing sideways. That’s a case where there still is more fluidity in the structure of biological disciplines than there are anywhere else. Maybe that’s because there’s more things being discovered faster.

The disciplines are often tool-based. Microbiology wouldn't be possible without microscopes. As we get tools to deal with artificial intelligence, for example, then it gives us an ability to see psychology in new ways.

TOM GRIFFITHS: An analogous rotation is happening in psychology, where you have neuroscience as one set of methods, which includes social neuroscience and clinical neuroscience, and then behavioral psychology is another set of methods. Those things would have traditionally been clinical psychology, cognitive psychology, social psychology, and so on, but then getting rotated around into behavioral methods versus neuroscientific methods, where the questions that are being asked in those groups are now starting to be quite different.  

AXELROD: One problem with a new discipline or new restructuring is the evaluation of performance. People in the new field will tend to say that anybody in the new field is worth valuing, worth promoting, worth funding. And because it's a new discipline, it doesn't have a well-established hierarchy that you can look to and say, "That journal is the best in the field." Even if you could claim that a journal is the best in this new field, you can’t necessarily say how it would rank among other things in a broader area.

An ability to do high quality and consensual evaluation outside of a new area means that there's a comparative advantage for somebody not very good to go into the new area. That in turn leads to a suspicion of people in the old area that the new area is attracting people that couldn’t successfully do the old thing, not that they weren’t interested in it or that they were pioneers in some sense, but maybe because they see they would thrive in some area where the evaluation is harder.

LLOYD: Sometimes what can happen to an old field, particularly those in which the original founders of the field might not have been the nicest people in the world, is that they can get grumpy and refuse to cite each other. Who was telling us about an NSF panel where there are these three solid-state physicists on this NSF panel, and they recommended none of the proposals be funded.

WOLFRAM: One thing I’ve noticed in the creation of new fields is that you ask who are the people who come into a new field when it’s created? For example, are they young people? Are they old people? What’s the type of person who comes into a new field when it’s created? I’ve been surprised that it’s not just young people who come into a new field. It seems to be the case that there are fields or paradigms that are suitable for particular individuals for one reason or another, and if they’re lucky enough to live in a time in history when their paradigm is one that is being pursued, then they gravitate to that and they start doing it.

My anecdotal observation is that if you look at people who come into a new field when it’s young and then you wait twenty years or something, about half the people who came into the new field when it was young are still in that field, and the other half have gone on to do three new fields or something after that. I’m curious what other people’s experience is.

ALISON GOPNIK: There’s an interesting tension that comes up with what Tom was saying about when fields bifurcate in terms of methodology or when they bifurcate in terms of content. In some ways, the methodological differences make it harder to communicate because the tools are different.

On the other hand, my experience has been that successful interdisciplinary cases are those that occur when you get people using different methods who were trying to answer the same question. Sometimes you get questions that are even narrower than the question you’d typically think of as being a domain question. For example, rather than asking how we're going to solve the problem of the mind, we ask how we're going to figure out how causal inference works. That’s a nice example of where we succeeded in getting real interdisciplinary work.

Another example is the question of how are we going to figure out how people understand what’s going on in other people’s minds, which came to be called theory of mind. That promoted genuine interdisciplinary work, and that was because you had people with different methods who were trying to solve the same problem. I can’t even think of examples where what would happen is that you get a bunch of people together because they said, "We’re all using the same methods and we want to find out more about the common methods even though we’re solving different problems."

AXELROD: Does anybody have examples of collaboration across disciplines that didn't work?

LLOYD: Like you, I’ve done a lot of interdisciplinary work and effectively invading a number of fields. Some fields very much don’t like being invaded, but also at the same time they’re flattered that somebody is paying attention to them. Starting about ten years ago, I and some friends from the quantum information community realized we could make some reasonable contributions to this field of quantum mechanics and photosynthesis. This is a small field full of grumpy old men who never cite each other’s work. Science proceeds one death at a time. At one of the conferences I said, " I’ve never met a field so closed in. You can only make progress in your own specialty by dying yourself," which graduate students thought was very funny, but the professors didn’t think it was very funny.

AXELROD: Your problem of how does photosynthesis work was obviously accepted as important. Photosynthesis is obviously accepted as an important thing and how it works was understood. It was already understood that they didn’t have a complete satisfactory account and, therefore, if you could provide a better account then they could appreciate that that’s a contribution no matter what tools you used to get there.

JONES: But he’s saying it was cranky and they didn’t accept it.

LLOYD: It ended up being accepted. They did need to learn about it because they didn’t understand what was going on with quantum coherence, and the methods that we supplied did allow them to figure that out, but boy they were dragged kicking and screaming to this and still don’t want to cite the work.

JOHN BROCKMAN: You mentioned that you’ve gone back to science rather than writing this book you were talking about. So, what is your science? How would you define advances in the science? I don’t quite understand what the field is.

AXELROD: I guess I’d put it two ways. The core of my interest is in international relations, especially great power relations and issues in times of war and peace. I'm also a math modeler. I’m looking for opportunities to do math modeling, in particular, agent-based modeling and Santa Fe Institute kind of complexity work. I’m an opportunist and a curious person, so if I see a model of a simulation of cancer, I try to figure out how it works. I make a real effort to meet people and talk to them often over lunch and sometimes at meetings like this one.

So, my field could be something different. I haven’t done anything specialized in artificial intelligence, but I’m fascinated by it. As a social scientist, I see that one approach to autonomous vehicles—which are our best example currently of sophisticated artificial machines—is to make them more sophisticated and better able to understand the environment and avoid mistakes. A whole other approach involves legal liability questions. The problem is, who’s going to take responsibility for the accidents, and how do we institutionalize that judgment between the insurance companies, between the manufacturers, between the owner of the car, between the person that sets the parameters?

LLOYD: Would you recommend to a junior faculty member to pursue interdisciplinary work?

JONES: There are fields that privilege single author and fields that privilege multiple authors, and I think the answer would be different based on these two models.

AXELROD: In order to get tenure, the tenure committee wants to know how good a bet you are in the long run. Let’s say all of your work is done with the same senior person. Well, that’s not a way to build up a record that then could be evaluated. If you’re going to do collaborative work, you should collaborate with different people to so that your work is distinctive. The other is to make sure that you’re single-authored work is among your best work. Work with different people and make sure that your single-authored stuff is among your best.