Download madsen-lipo-bentley-saa2008-poster.pdf

and will be published soon along with our entire poster session as an issue of the online Journal of Evolutionary and Historical Sciences.  I’ll talk more about this soon, but this week is incredibly busy and I wanted to get at least one link up for the paper before I forget.

I haven’t written much here since late December, but only because life has reached a fever pitch again, and the brief times I have free away from a full schedule need to be devoted to research and my dissertation, not idle contemplation for my website. But we’re in the thick swamp of an election season, unseasonably early of course, and I haven’t written anything about the candidates, the primaries, the debates, as I did for much of 2004. I can’t promise to get back to regular posting before Super Tuesday, but I hope to soon thereafter. Or as soon as I can get my two projects more firmly underway (one paper, one poster) for the SAA (Society for American Archaeology) meetings in late March in Vancouver. Both are co-authored with Alex Bentley and Carl Lipo, and we’re working on the statistical consequences of expressing formal models of cultural transmission within realistic social networks.

For those unfamiliar with cultural transmission, this is the observation that humans are not born with a hard-coded set of cultural behaviors (in the sense of genetically transmitted) but learn, over the course of child development and throughout life, ways of behaving and believing and thinking through interaction with others in our social groups. In a formal sense, cultural transmission is modeled mathematically through analogues of haploid population genetics models (Wright-Fisher and Moran processes), replicator dynamics and allied models from evolutionary game theory, and the contact and voter models in the study of “interacting particle systems” or spatial stochastic processes by probability theorists and statistical physicists. An open question, whose likely answer is “yes,” is that these methods of modeling cultural learning and transmission are formally equivalent, given appropriate variations of population structure and the focus on deterministic versus stochastic models. But more of that in future posts, hopefully.

Basically, I’m working with some collaborators studying models of social learning and communication, for predictive ensemble or spatial statistical “signatures” in cultural data which are mapped spatially and dated temporally. A “signature” would be a unique pattern of statistical properties which tells us how a given population was structured (in terms of social networks) given the results of how cultural information flowed within the population, and came to be reflected in material objects or artifacts. An example would be a model in which we learn about, and adopt, preferences for songs and music from our social network of friends, but in an unbiased fashion — we occasionally adopt the preferences of a colleague or associate. What statistical properties does this local process of imitation have, when projected into a “global” perspective — statistical patterns within a population, spatial patterns in kinds of data we can map and chart?

Of course, we all know that the model I just described is pretty simplistic. Nobody “just copies” their friends, let alone doing so without any filters, biases, and on a strict “coin flip” or probabilistic basis. But it turns out it sure can look that way when you aggregate the results of many people imitating, choosing, learning, and adopting ideas. So this kind of model is a good “null hypothesis” for a simplistic kind of cultural communication — anything more realistic will have to depart from this simple random model in striking, hopefully unique ways.

Being able to find unique, predictive patterns from more complex models of cultural learning and communication is possible, but not guaranteed — it is easily possible (maybe even likely) that several different kinds of social situations could lead to the same overall patterns at a local, regional, or even global level. We call this problem “equifinality” — the data we have are insufficient to distinguish between several possible processes, so given our models and data, each process is “equally likely” to have caused the observed pattern.

This type of research is what I’ve been engaged in for a long time — at least since 1995, with conference papers, publications, and Carl Lipo’s dissertation research covering some of the results. Now I’m extending our previous work and learning a lot of math, probability, and population genetics in the process. It’s fascinating stuff, but in addition to the job at GridNetworks the work keeps me pretty busy.

This is all by way of explanation for my longish absences from writing something here. I hope to remedy that, as I said, but there’s some serious work between now and then.

Today I got a bit of an inkling at a more psychologically adequate answer.  That’s not to say that it’s the "correct," or complete, or the only answer, but I immediately recognized it as a gut-level truth, at least within the scope of my life.  I attended our department’s reception for graduate students and faculty, held at the beginning of the academic year, and met a number of new and returning students, many of whom (because I’ve been off doing business and other things) I’d never met. 

One of the students told me of a class taught the previous year where they’d read a paper I’d written with Carl Lipo (and likely others, I didn’t catch the exact citation, but I’m guessing Lipo et al. 1997).  He mentioned it because of the oddity of actually running into and meeting one of the authors, but for me the experience was significant.  Here was somebody who knew something about me before ever having met me — in this case, how I thought and what I thought about a topic.  He’d encountered some aspect of me as assigned reading in a class, and thus was acquainted with something I’d done and thought, years before. 

We hope to be known, ultimately, by our words and thoughts and ideas.  We hope to be assigned reading, or the fortuitous article or book encountered in the library late one night.  We hope to be the idea that causes somebody else’s project or thoughts to finally "gel" and come together.  Just as others served as the building blocks with which we had a tiny nugget of a new idea, we hope to be the seeds of someone else’s new ideas, down the road.  Most of my ideas and published works will not accomplish this, but some might, whether in a small or large way.

And with that, I greet the new academic year, secure in the knowledge that something I wrote is being read.  And that’s why I, and many others, do research. 

Poster sessions were terrific this year, with plenty of space to walk around and see everything; increasingly I find that spoken talks are much less interesting, particularly when people have few slides and read their written-out talks in a flat monotone. How do people expect to convince or interest an audience without a strong presentation style?

Among the interesting papers I saw were two papers on costly signaling theory by Aimee Plourde (of UCL), and Jillian Galle (Monticello). Another paper by Colin Quinn and Ian Kuijt on signaling in the Natufian was also interesting, but I need to see a written copy to follow their argument on how they link Natufian burial behavior to costly signaling.

A paper by Railsbeck et al., in the September 2006 issue of Simulation, is right on the money in saying that the original Objective-C version of the Swarm toolkit is a strong “framework,” as opposed to the “library” style of successor toolkits like RepastJ and MASON. Swarm definitely forced a style of organization onto your simulation models, via the concept of nested “swarms” of agents, observers, etc. I suspect this is much like Ascape, but the latter doesn’t seem to be an active development project any longer (at least given the website – leave a comment if this incorrect). Whereas Repast provides a ton of infrastructure but simulation models themselves seem to be fairly unstructured, as I read various examples and models folks have posted online.

One of the reasons it’s hard to create a generic framework for computational models is that they differ in structure and needs across domains. On the other hand, within a specific research domain, it seems like structure should be relatively standardized, particularly if we want to communicate, share, and replicate results. So the governing philosophy I’ve been using with TransmissionLab is that of a domain-specific standard modeling framework built on a standard generic toolkit, tailored to the research needs of a given domain (in this case heritability/genetics/cultural transmission modeling), but flexible enough to allow the building of many models by different groups of researchers. We’ll see how this works out, but it seems like a good “middle” approach to the problem.

Version 1.4 of TransmissionLab is a reasonable step in that direction. Nearly all of the “boilerplate” involved in building a computational model is abstracted into a series of Java interfaces and abstract classes (which provide templated partial implementations for the interfaces). All of the “machinery” of the simulation deal exclusively with interfaces, such that nearly anything that the “user” writes can be considered a reusable component. For example, in the rules package, TransmissionLab currently includes a rule for “mutation” called RandomAgentInfiniteAllelesMutation. Assuming that “traits” are being represented as integer “tags” (a generalization of this will be in a future version to allow Object traits with internal structure), this rule will operate on any implementing class of the interface IAgentPopulation. This allows a simulation author to reuse this mutation rule, creating populations that are well-mixed, spatially structured, network-structured, and so on.

The central model class (which contains the “main” method, for example) is similarly templated. The “component” classes (rules, data collectors, population factories) all work with classes which implement ISimulationModel. In practice, main model classes extend the abstract class AbstractTLModel, which provides full or partial (template method pattern) implementations of the ISimulationModel interface contract. By design, satisfying the ISimulationModel contract also guarantees that your model class meets the needs of the SimModel contract in Repast, and thus will be runnable by any standard Repast method. But since much of this “infrastructure” is contained in the abstract base class, user-written model classes are much much simpler. In fact, looking at the new implementation of the Random Copying Model this morning (class NeutralCTModel), it strikes me that everything in it could be specified declaratively, in an XML configuration file. So I may get to the point soon where user-written models are just XML files, users write new components as Java classes, and the framework stitches together the rest.

All of this may sound suspiciously like Repast Simphony, so why I am putting in this work?

Several reasons. Simphony is still alpha code, and although it’s coming along very nicely, I’m not yet sure what I think of it as a modeling framework. Clearly many of the goals are the same as I’m discussing here, and I may port TransmissionLab to Simphony down the road.

But Simphony is a bit of a “black box” at the moment: I’m not quite sure how it all works under the hood, and that bothers me. Exploratory simulations are fine if I don’t understand the fine detail of the runtime, but if we’re going to be sure that we’re getting scientifically usable results out of a computational model, we’d better darned well know exactly how the model works, period. This is the problem I have with NetLogo (other than proprietariness and the fact that it’s well suited to specific types of models, not generic simulation).

So I’m still working with RepastJ, and plan to continue doing so until there’s a compelling case to switch. And working with RepastJ is a lot simpler if you have just a bit more structure. So TransmissionLab will continue to evolve in the direction of a domain-specific modeling framework using RepastJ. If you’re doing simulations of cultural communication and transmission phenomena, check it out.

Carl is digitizing all of Archy 497, the first of two quarters of archaeological theory. In 497, Dunnell focused on “formal theory” — concepts, key conceptual relationships, and the classification tools necessary for all explanation in archaeology. In 498, which likely will be the next digitizing project, Dunnell focuses on “explanatory” theory and the history of archaeological theory.

For those readers unfamiliar with R.C. Dunnell, he was my former academic advisor, longtime Chairman of the Department of Anthropology at the University of Washington, key initiator and driver of Darwinian approaches to explanation in archaeology, and scourge of generations of first-year graduate students. Dr. Dunnell retired in the mid-1990’s and now resides in the Southeastern United States, surrounded by Mississippian mounds, archaeological sites, and decent BBQ joints.

Both the batch-mode parameter file and library dependencies have slight differences from Version 1.2, so be sure to grab both otherwise you’ll encounter errors starting up a simulation. In particular, this release adds a dependency upon the Jakarta Commons CLI library for command-line parsing, since this isn’t a strong suit of the Repast libraries.

This version also adds one statistic to the OverallStatisticsRecorder data collection module. For each simulation run, we calculate the average number of agents who have traits (measured at each model tick) which are listed in the “top N” list of traits. In other words, if you’re working with a “top 40″ list of song-analogues, this statistic measures the number of agents whose chosen trait is a song in the top 40, as opposed to a trait that wasn’t frequent enough to make the top 40 list. This statistic is thus paired analytically with the parameter for the size of the “top N” lists, and the combination of the two should be interesting to examine across a range of mutation rate and population size parameters.

On a related note, LiveScience has an article on the upcoming article by Alex Bentley, Carl Lipo, Harold Herzog, and Matthew Hahn. I recommend it for a somewhat popularized account of the main conclusions of their 2007 paper. Since much of what we’re doing with TransmissionLab at the moment is going further along the lines suggested by Bentley et al., and earlier Fraser Neiman, Carl Lipo, and myself, it’s a good clue to the kinds of phenomena we can explore purely assuming that choice among alternatives is statistically random or neutral.

I’ve been busy on the simulation modeling front this week, coming out with a “final” version of the RandomCopyModel from an upcoming paper by Alex Bentley, Carl Lipo, Harold Herzog, and Matthew Hahn. I didn’t work on the original model or paper, but Alex graciously allowed me to use the original code as the basis for some future experimental models we’re working on, as well as simulation setups I’ll need for my dissertation research. In return I refactored and made the original simulation a bit clearer to follow in terms of the code, so we’re providing version 1.3 of the RandomCopyModel under the terms of a Creative Commons-GPL license for non-commercial use.

In the meantime, I’ve also been planning the next “version” of the model, which is now a separate codebase managed by Google Code under the name TransmissionLab.

The goal of TransmissionLab is to accurately represent theoretical models of CT (e.g., random copying, prestige-biased transmission, frequency-biased transmission) within a variety of population structures (e.g., complete graphs/well-mixed, sparse random graphs and social networks of varying topologies, spatial lattices), and using a variety of update algorithms (e.g., Moran processes, Wright-Fisher processes, various other birth-death processes). TransmissionLab seeks to also make data collection and “observation” of simulated populations simple, with modules which are completely separate from the simulated population itself thus preventing observational “side-effects” on the model. Analysis flows from data collection, and can be done in a variety of ways.

Naturally we’re not the first people by any means to do agent-based simulation models of cultural transmission, imitation behavior within populations, or the diffusion of innovations. Heck, this isn’t even the first simulation of these sorts of phenomena I’ve been involved with. Where I hope that TransmissionLab differs is that nearly all of my previous simulation models have proven fairly special-purpose, expedient models for working on one particular question or problem, and I’m trying to make TransmissionLab a common platform that can span projects both for myself and my research group.

This is important for a number of reasons:

1. Stable, well-used models tend to be well-structured, well-tested models. The issues of whether the simulation is showing us artifacts of writing software or the theoretical behavior we’re trying to describe can only be solved by deep investment in design, coding, and testing.
2. Relating the results of one study to another when each uses different simulation code is a difficult one, for obvious reasons. To the extent that we use the same code framework and models to perform multiple studies, we can make arguments (and even measurements) which relate the results of several research studies to each other.
3. The relationship issue mentioned in the previous item could span multiple research teams if the model is well-structured and tested enough that others adopt it for research.

Thus, I’m putting some reasonable effort into developing TransmissionLab, and if you have an interest in agent-based modeling and cultural phenomena, I hope you give it a look in a version or two. Right now I’m moving from the older RandomCopyModel to a new set of development tasks, which will be outlined shortly at the googlecode wiki for the project. Once these are checked in, there will be some interesting elements beyond the earlier model to explore. I’ll post when that happens.

Ernst Mayr, the great biologist and architect of the Modern Synthesis, wrote in a 1959 essay that Darwin’s great contribution to biology was anti-essentialism, or what Mayr called “population thinking.” By this, Mayr meant that Darwin was one of the first biological thinkers to offer a theory of the evolution of species which did not rely upon changes to, differences in, or transformation of the “essence” of a species. In the older view, which goes back to Aristotle (at in terms of codifying this view; the origins of this view are much more ancient and are likely bound up in the cognitive science of “natural kinds”), each species is characterized by an “essence” or definition, which tells us what characteristics an animal, plant, and so on must have in order to belong to that species. The fact that each individual in a species is unique, and that often many individuals lack one or more essential characteristics, but are still considered part of the species, is explained away in Aristotelian essentialism as simply noise or reproductive error that causes the real world to be an imperfect reflection of the species’s underlying “reality.” Variation is thus explanatorily unimportant when species are viewed as characterized by pre-Darwinian biology. And the evolution of one species into another, over time, seems to run up against a massive gulf between two “essences.” One sees echoes of this “problem,” for example, in the objections of many contemporary anti-evolutionists when presented with what biologists believe is the abundant empirical evidence of evolution: the key words “intermediate forms” pop up as tell-tale signs, even though such a notion really arises only if one thinks a species has an “essence” between which a form could be “intermediate.”

Darwin’s great contribution, at least according to Mayr and others, was in founding modern evolutionary biology on a firm basis of anti-essentialism. In this case, what Mayr originally meant by “population thinking” (philosophers will recognize it as a variety of anti-essentialism) is that variation among individuals is not “noise,” nor is it meaningless error — variation among individuals is both the cause of evolution, and the great engine that powers the development of successive adaptations through the continuous (if occasionally roundabout) process of selection. Variation isn’t just important for selection, variation is critical for selection. No variation, no selection, as pointed out by the great mathematical biologist and statistician Ronald Fisher. In fact, selection might simply be the statistical consequence of having variation, some of which makes a difference of our life chances, in an environment where there aren’t enough resources, or enough room, or enough time, for every individual to succeed equally. Selection almost comes naturally when you think about the world through the lens of Darwin’s population thinking.

What has this got to do with social science and Darwin’s contribution to the human sciences? Potentially everything.

The social sciences have long been characterized by modes of explanation which draw upon Aristotelian essentialism for their conceptual structure, just as pre-Darwinian biology did. A simple example from my own field of anthropology illustrates the problem nicely: for many years, anthropologists classified whole societies as belonging to one of several “levels” of organization and thus social evolution. Societies could be “bands,” “tribes,” “chiefdoms,” or “states.” Furthermore, societies were often seen as moving through this progression, from simpler to more complex forms of organization. While this scheme is not often used in its raw form today, and the notion of a straight-line, unilinear “progression” from one form to other in rigid fashion has been thoroughly debunked within the social sciences, it’s not completely gone. One still reads professionals characterizing a particular social group being at the “chiefdom level of organization,” as if there were a natural “reality” to these categories that all societies had to fit. Even if the people in question aren’t really meaning to allude to unilinear forms of social theory, the association is there and the concepts hard to get away from.

This is nowhere more evident than in popular and political discourse, where one sees echoes of straight-line, unilinear cultural change in assertions that all societies trend towards democracy, or towards capitalism, or towards market economies, or really towards any singular and unitary goal. I don’t need to point any further than Francis Fukuyama or Robert Wright, both of whom defend fairly unlinear, 19th century views on the “necessary” path of cultural and historical change. Their views are far from isolated, far from unpopular, and far from lacking influence.

Darwin’s gift to the social sciences is really the same as it was to biology: the tools to reject such remnants of the “Great Chain of Being” and Aristotelian essentialism by replacing it with accounts of social and cultural change that recognize the importance of variation, of individual action, and of selective and other population-level processes that shape how that variation is lost or retained over time. Darwin’s gift is potentially a richer, more predictive, but also less “artificial” account of where we came from and what our possibilities are going forward. I say this because it treats our diversity of action, thought, and ideas as the critical ingredients for who we are and how we’re changing, rather than simply noise or randomness to explain away once we think we see a clean, neat, global historical pattern into which we believe everyone fits.

Darwin’s gift isn’t a particularly easy one to use, however. Leaving aside the 19th century legacy of progressive, essentialist cultural evolution for a moment, much of the social sciences in the 20th century were under the sway of rational choice theory in various forms. Rational choice theory, and neo-classical economics, are each based on very essentialist ways of looking at behavior. Preferences are assumed to be static, rather than evolving as a consequence of population-level processes as a whole, and given static preferences and decision-theoretic rules which are assumed to be constant within the population (i.e., part of “human nature”), rational choice theory in its various guises predicts the “equilibria” that will arise. These equilibria are the mathematical echoes of the “essences” one sees in pre-Darwinian accounts of species. Departures from equilibria are unimportant in these theories, because our assumption is that rational calculation and exogenous preferences would quickly lead back to equilibrium. Economic and political science “equilibrium theories” thus treat variation as noise — not as something important as part of our explanations.

The assault on such thinking has been slow in coming, but it is happening across the social sciences. It’s occurring in the form of non-equilibrium, individual-based models in economics, in evolutionary game theory which replaces “common knowledge” and rational agents with variation and replicator dynamics, and with strong interest in the dynamics of change rather than simply the endpoints or “equilibria.” I take these trends as good signs that we’re learning to use Darwin’s gift. Development of new models and theories, not to mention tests and experiments, to replace rational choice, essentialist social classifications and Homo economicus don’t happen overnight. But the effort has been going on sporadically for 30 years or so, and it’s gaining steam.

Interestingly enough, the strong interest in critical, Continental, or “post-modern” philosophy within the social sciences is also a reflection of Darwin’s gift. Despite disagreements about the nature, place, and methods of science and inquiry (and in particular, the useless fights over “scientism” and “relativism), both evolutionary social scientists and post-modernists are simply trying in different ways to figure out the consequences of anti-essentialism for describing and understanding the human experience. We have much to talk about, but we currently still speak very different languages and there are few, if any, good translators (the Richard Rorty of Consequences of Pragmatism being a notable exception, in my view). Bridging the gulf here, in pragmatic rather than contentious or zero-sum ways, is important to a unified and diverse social sciences, but it is an effort that’s hardly begun, let alone one that is gaining any ground.

Clearly there’s a lot one can say and argue about this topic. Before this day ends (and it’s getting quite late), I wanted to simply recognize the core contribution that Darwin’s “population thinking” and anti-essentialism has begun to make in the social sciences, and look forward to a day when we have more results from its deployment to show, particularly to those who remain skeptical of its “difference” from traditional modes of inquiry in the human sciences. And in recognizing this, to celebrate in a small way the life and anniversary of one of our greatest scientists and thinkers. He changed us and our ways of thinking more than he knew, and even more than we probably still recognize.