Home » agent based modeling
Category Archives: agent based modeling
I’m presenting next week at the Society for American Archaeology Annual Meeting. I’m giving two papers. One argues for parsimonious models when we do agent based modeling. The other reverses the flow of archaeological network analysis and instead of finding nets in the archaeology, I use agent based models to generate networks that help me understand the archaeology. (The session is ‘Connected Past’.) Here is the draft of my talk, with all the usual caveats that that entails. Parts of it have been drawn from an unpublished piece that discusses this methodology and the results in much greater detail. It will appear…. eventually.
Scott Weingart has been an enormous help in all of this. You should follow his work.
My interests lie in the social networks surrounding primary resource extraction in the Roman world. The Roman epigraphy of stamped brick easily lends itself to network analysis. One string together, like pearls, individual landowners, estate names, individual brick makers, signa, brick fabrics, and locations. This leads to very complicated, multi-dimensional networks.
When I first started working with this material, I reduced this complexity by looking only at the humans, whom I tied together based on appearing in the same stamp type together. I called these ‘producer’ networks. I then looked at the ties implied by the shared use of fabrics, or the co-location of brick stamp types at various findspots, and called these ‘manufacturing’ networks.
I then sliced these networks up by reigning dynasty, and developed a story to account for their changing shapes over time.
This was in the late 1990s, and in terms of network theorists I had largely only Granovetter, Hanneman & Riddle, and Strogatz & Watts to go on. The story I told was little more than a just-so story, like how the Camel got its Hump.
I had the shape, I had points where I could hang the story, but I couldn’t account for how I got from the shape of the network in the Julio-Claudian period, to that of the Flavian, to that of the Antonines. I’ve done a lot of work on networks since then; now I want to know what generates these networks that we see archaeologically, in the first place.
In this talk today, I want to reverse the direction of my inquiry. We are all agreed that we can find networks in our archaeological materials. The problem, I think, for us, is to explain the network processes that produce these patterns, and then to use our understanding of those processes to narrow down the possible entangled human & thing interactions that could give rise to these possible processes.
We need to be able to understand the possible behaviour-spaces that could produce the networks we see, to tease out the inevitable from the contingent. We need to be able to rigorously explore the emergent or unintended consequences of the stories we tell. The only way I know how to do that systematically, is to encode those stories as computer code, to turn them from normal, archaeological storytelling rhetoric, to computational procedural rhetoric.
So this is what we did.
One story we tell about the Roman world, that might be useful for understanding things like the exploitation of land for building materials, is that its social economy functioned like a ‘bazaar’.
According to Peter Bang, the Roman economic system is best understood as a complex, agrarian tributary empire, of a kind similar to the Ottoman or Mughal (Bang 2006; 2008). Bang (2006: 72-9) draws attention to the concept of the bazaar. The bazaar was a complete social system that incorporated the small peddler with larger merchants, long distance trade, with a smearing of categories of role and scale. The bazaar emerged from the interplay of instability and fragmentation. The mechanisms developed to cope with these reproduced that same instability and fragmentation. Bang identifies four key mechanisms that did this: small parcels of capital (to combat risk, cf Skydsgaard 1976); little homogenization of products (agricultural output and quality varied year by year, and region by region as Pliny discusses in Naturalis Historia 12 and 18); opportunism; and social networks (80-4). As Bang demonstrates, these characteristics correspond well with the archaeology of the Roman economy and the picture we know from legal and other text.
Bang’s model of the bazaar (2008; 2006), and the role of social networks within that model, can be simulated computationally. What follows is a speculative attempt to do so, and should be couched in all appropriate caveats and warnings. The model simulates the extraction of various natural resources, where social connections may emerge between individuals as a consequence of the interplay of the environment, transaction costs, and the agent’s knowledge of the world. If the networks generated from the computational simulation of our models for the ancient economy correspond to those we see in the ancient evidence , we have a powerful tool for exploring antiquity, for playing with different ideas about how the ancient world worked (cf. Dibble 2006). Computation might be able to bridge our models and our evidence. In particular, I mean, ‘agent based modeling’.
Agent based modelling is an approach to simulation that focuses on the individual. In an agent based model, the agents or individuals are autonomous computing objects. They are their own programmes. They are allowed to interact within an environment (which frequently represents some real-world physical environment). Every agent has the same suite of variables but each agent’s individual combination of variables is unique (if it was a simulation of an ice-hockey game, every agent would have a ‘speed’ variable, and an ‘ability’ variable, and so the nature of every game would be unique). Agents can be aware of each other and the state of the world (or their location within it), depending on the needs of the simulation. It is a tool to simulate how we believe a particular phenomenon worked in the past. When we simulate, we are interrogating our own understandings and beliefs.
The model imagines a ‘world’ (‘gameboard’ would not be an inappropriate term) in which help is necessary to find and consume resources. The agents do not know when or where resources will appear or become exhausted. By accumulating resources, and ‘investing’ in improvements to make extraction easier, agents can accrue prestige. When agents get into ‘trouble’ (they run out of resources) they can examine their local area and become a ‘client’ of someone with more prestige than themselves. It is an exceedingly simple simulation, and a necessary simplification of Bang’s ‘Bazaar’ model, but one that captures the essence and exhibits subtle complexity in its results. The resulting networks can be imported into social network analysis software like Gephi.
It is always better to start with a simple simulation, even at the expense of fidelity to the phenomenon under consideration, on the grounds that it is easier to understand and interpret outputs. A simple model can always be made more complex when we understand what it is doing and why; a complex model is rather the inverse, its outcomes difficult to isolate and understand.
A criticism of computational simulation is that one only gets out of it what one puts in; that its results are tautological. This is to misunderstand what an agent based simulation does. In the model developed here, I put no information into the model about the ‘real world’, the archaeological information against which I measure the results. The model is meant to simulate my understanding of key elements of Bang’s formulation of the ‘Imperial Bazaar’. We measure whether or not this formulation is useful by matching its results against archaeological information which was never incorporated into the agents’ rules, procedures, or starting points. I never pre-specify the shape of the social networks that the agents will employ; rather, I allow them to generate their own social networks which I then measure against those known from archaeology. In this way, I start with the dynamic to produce static snapshots.
We sweep the ‘parameter space’ to understand how the simulation behaves; ie, the simulation is set to run multiple times with different variable settings. In this case, there are only two agent variables that we are interested in (having already pre-set the environment to reflect different kinds of resources), ‘transaction costs’ and ‘knowledge of the world’. Because we are ultimately interested in comparing the social networks produced by the model against a known network, the number of agents is set at 235, a number that reflects the networks known from archaeometric and epigraphic analysis of the South Etruria Collection of stamped Roman bricks (Graham 2006a).
What is particularly exciting about this kind of approach, to my mind, is that if you disagree with it, with my assumptions, with my encoded representation of how we as archaeologists believed the ancient world to have worked, you can simply download the code, make your own changes, and see for yourself. If you are presented with the results of a simulation that you cannot open the hood and examine its inner workings for yourself, you have no reason to believe those findings. Thus agent based modeling plays into open access issues as well.
So let us consider then some of the results of this model, this computational petri dish for generating social networks.For my archaeological networks, I looked at clustering coefficient and average path length as indicator metrics, (key elements of Watts’ small world formulation). We can tentatively identify a small-world then as one with a short average path length and a strong clustering coefficient, compared to a randomly connected network with the same number of actors and connections. Watts suggests that a small-world exists when the path lengths are similar but the clustering coefficient is an order of magnitude greater than in the equivalent random network (Watts 1999: 114).
In Roman economic history, discussions of the degree of market integration within and across the regions of the Empire could usefully be recast as a discussion of small-worlds. If small-worlds could be identified in the archaeology (or emerge as a consequence of a simulation of the economy), then we would have a powerful tool for exploring flows of power, information, and materials. Perhaps Rome’s structural growth – or lack thereof – could be understood in terms of the degree to which the imperial economy resembles a small-world (cf the papers in Manning and Morris 2005)?
The networks generated from the study of brick stamps are of course a proxy indicator at best. Not everyone (presumably) who made brick stamped it. That said, there are some combinations of settings that produce results broadly similar to those observed in stamp networks, in terms of their internal structure and the average path length between any two agents.
One such mimics a world where transaction costs are significant (but not prohibitive), and knowledge of the world is limited . The clustering coefficient and average path length observed for stamped bricks during the second century fall within the range of results for multiple runs with these settings. In the simulation, the rate at which individuals linked together into a network suggests that there was a constant demand for help and support. The world described by the model doesn’t sound quite like the world of the second century, the height of Rome’s power, that we think we know, suggesting something isn’t quite right, in either the model or our understandings. But how much of the world did brickmakers actually know, remembering that ‘knowledge of the world’ in the model is here limited to the location of new resources to exploit?
Agent based modeling also allow us to explore the consequences of things that didn’t happen. There were a number of simulated worlds that did not produce any clustering at all (and very little social network growth). Most of those runs occurred when the resource being simulated was coppiced woodland. This would suggest that the nature of the resource is such that social networks do not need to emerge to any great degree (for the most part, they are all dyadic pairs, as small groups of agents exploit the same patch of land over and over again). The implication is that some kinds of resources do not need to be tied into social networks to any great degree in order for them to be exploited successfully (these were also some of the longest model runs, another indicator of stability).
What are some of the implications of computationally searching for the networks characteristic of the Roman economy-as-bazaar? If, despite its flaws, this model correctly encapsulates something of the way the Roman economy worked, we have an idea of, and the ability to explore, some of the circumstances that promoted economic stability. It depends on the nature of the resource and the interplay with the degree of transaction costs and the agents’ knowledge of the world. In some situations, ‘patronage’ (as instantiated in the model) serves as a system for enabling continual extraction; in other situations, patronage does not seem to be a factor.
However, with that said, none of the model runs produced networks that had the classical signals of a small-world. This is rather interesting. If we have correctly modeled the way patronage works in the Roman world, and patronage is the key to understanding Rome (cf Verboven 2002), we should have expected that small-worlds would naturally emerge. This suggests that something is missing from the model – or our thinking about patronage is incorrect. We can begin to explore the conundrum by examining the argument made in the code of the simulation, especially in the way agents search for patrons. In the model, it is a local search. There is no way of creating those occasionally long-distance ties. We had initially imagined that the differences in the individual agents’ ‘vision’ would allow some agents to have a greater ability to know more about the world and thus choose from a wider range. In practice, those with greater ‘vision’ were able to find the best patches of resources, indeed, the variability in the distribution of resources allowed these individuals to squat on what was locally best. My ‘competition’ and prestige mechanisms seem to have promoted a kind of path dependence. Perhaps we should have instead included something like a ‘salutatio’, a way for the agents to assess patrons’ fitness or change patrons (cf Graham 2009; Garnsey and Woolf 1989: 154; Drummond 1989: 101; Wallace-Hadrill 1989b: 72-3). Even when models fail, their failures still throw useful light. This failure of my model suggests that we should focus on markets and fairs as not just economic mechanisms, but as social mechanisms that allow individuals to make the long distance links. A subsequent iteration of the model will include just this.
This model will come into its own once there is more and better network data drawn from archaeological, epigraphic, historical sources. This will allow the refining of both the set-up of the model and comparanda for the results. The model presented here is a very simple model, with obvious faults and limitations. Nevertheless, it does have the virtue of forcing us to think about how patronage, resource extraction, and social networks intersected in the Roman economy. It produces output that can be directly measured against archaeological data, unlike most models of the Roman economy. When one finds fault with the model (since every model is a simplification), and with the assumptions coded therein, he or she is invited to download the model and to modify it to better reflect his or her understandings. In this way, we develop a laboratory, a petri-dish, to test our beliefs about the Roman economy. We offer this model in that spirit.
[edited April 4th to make it less clumsy, and to fit in the 15 minute time frame]
Below is a draft of the first part of my talk for Scholarslab this week, at the University of Virginia. It needs to be whittled down, but I thought that those of you who can’t drop by on Thursday might enjoy this sneak peak.
Thursday, March 21 at 2:00pm
in Scholars’ Lab, 4th floor Alderman Library.
When I go to parties, people will ask me, ‘what do you do?’. I’ll say, I’m in the history department at Carleton. If they don’t walk away, sometimes they’ll follow that up with, ‘I love history! I always wanted to be an archaeologist!’, to which I’ll say, ‘So did I!’
My background is in Roman archaeology. Somewhere along the line, I became a ‘digital humanist’, so I am honoured to be here to speak with you today, here at the epicentre, where the digital humanities movement all began.
If the digital humanities were a zombie flick, somewhere in this room would be patient zero.
Somewhere along the line, I became interested in the fossilized traces of social networks that I could find in the archaeology. I became deeply interested – I’m still interested – in exploring those networks with social network analysis. But I became disenchanted with the whole affair, because all I could develop were static snapshots of the networks at different times. I couldn’t fill in the gaps. Worse, I couldn’t really explore what flowed over those networks, or how those networks intersected with broader social & physical environments.
It was this problem that got me interested in agent based modeling. At the time, I had just won a postdoc in Roman Archaeology at the University of Manitoba with Lea Stirling. When pressed about what I was actually doing, I would glibly respond, ‘Oh, just a bit of practical necromancy, raising the dead, you know how it is’. Lea would just laugh, and once said to me, ‘I have no idea what it is you’re doing, but it seems cool, so let’s see what happens next!’
How amazing to meet someone with the confidence to dance out on a limb like that!
But there was truth in that glib response. It really is a form of practical necromancy, and the connections with actual necromancy and technologies of death is a bit more profound than I first considered.
So today, let me take you through a bit of the deep history of divination, necromancy, and talking with the dead; then we’ll consider modern simulation technologies as a form of divination in the same mold; and then I’ll discuss how we can use this power for good instead of evil, of how it fits into the oft-quote digital humanities ethos of ‘hacking as a way of knowing’ (which is rather like experimental archaeology, when you think about it), and how I’m able to generate a probabilistic historiography through this technique.
And like all good necromancers, it’s important to test things out on unwilling victims, so I would also like to thank the students of HIST3812 who’ve had all of the ideas road-tested on them earlier this term.
Zombies clearly fill a niche in modern western culture. The president of the University of Toronto recently spoke about ‘zombie ideas’ that despite our best efforts, persist, infect administrators, politicians, and students alike, trying to eat the brains of university education.
Zombies emerge in popular culture in times of angst, fear, and uncertainty. If hollywood has taught us anything, it’s that Zombies are bad news. Sometimes the zombies are formerly dead humans; sometimes they are humans who have been transformed. Sometimes we deliberately create a zombie. The zombie can be controlled, and made to do useful work; zombie as a kind of slavery. More often, the zombies break loose, or are the result of interfering with things humanity was wont not too; apocalypse beckons. But sometimes, like ‘Fido’, a zombie can be useful, can be harnessed, and somehow, be more human than the humans. [Fido]
If you’d like to raise the dead yourself, the answer is always just a click away [ehow].
There are other uses for the restless dead. Before our current fixation with apocalypse, the restless dead could be useful for keeping the world from ending.
In video games, we call this ‘the problem space’ – what is it that a particular simulation or interaction is trying to achieve? For humanity, at a cosmological level, the response to that problem is through necromancy and divination.
I’m generalizing horribly, of course, and the anthropologists in the audience are probably gritting their teeth. Nevertheless, when we look at the deep history and archaeology of many peoples, a lot can be tied to this problem of keeping the world from ending. A solution to the problem was to converse with those who had gone before, those who were currently inhabiting another realm. Shamanism was one such response. The agony of shamanism ties well into subsequent elaborations such as the ball games of mesoamerica, or other ‘game’ like experiences. The ritualized agony of the athlete was one portal into recreating the cosmogonies and cosmologies of a people, thus keeping the world going.
The bull-leaping game at Knossos is perhaps one example of this, according to some commentators. Some have seen in the plan of the middle minoan phase of this palace (towards the end of the 2nd millenium BC) a replication in architecture of a broader cosmology, that its very layout reflects the way the Minoans saw the world (this is partly also because this plan seems to replicate in other Minoan centres around the Aegean). Jeffrey Soles, pointing to the architectural play of light and shadow throughout the various levels of Knossos argues that this maze-like structure was all part of the ecstatic journey, and ties shamanism directly to the agonies of sport & game in this location. We don’t have the Minoans’ own stories, of course, but we do have these frescoes of bull-leaping, and other paraphernalia which tie in nicely with the later dark-age myths of Greece
So I’m making a connection here between the way a people see the world working, and their games & rituals. I’m arguing that the deep history of games is a simulation of how the world works.
This carries through to more recent periods as well. Herodotus wrote about the coming of the Etruscans to Italy: “In the reign of Atys son of Menes there was a great scarcity of food in all Lydia. For a while the Lydians bore this with patience; but soon, when the famine continued, they looked for remedies, and various plans were suggested. It was then that they invented the games of dice, knucklebones, and ball, and all the other games of pastime, except for checkers, which the Lydians do not claim to have invented. Then, using their discovery to forget all about the famine, they would play every other day, all day, so that they would not have to eat… This was their way of life for eighteen years. Since the famine still did not end, however, but grew worse, the king at last divided the people into two groups and made them draw lots, so that one should stay and the other leave the country’.
Here I think Herodotus misses the import of the games: not as a pasttime, but as a way of trying to control, predict, solve, or otherwise intercede with the divine, to resolve the famine. In later Etruscan and Roman society, gladiatorial games for instance were not about entertainment but rather about cleansing society of disruptive elements, about bringing everything into balance again, hence the elaborate theatre of death that developed.
The specialist never disappears though, the one who has that special connection with the other side and intercedes for broader society as it navigates that original problem space. These were the magicians and priests. But there is an important distinction here. The priest is passive in reading signs, portents, and omens. Religion is revealed, at its proper time and place, through proper observation of the rituals. The magician is active – he (and she) compels the numinous to reveal itself, the spirits are dragged into this realm; it is the magician’s skill and knowledge which causes the future to unfurl before her eye.
The priest was holy, the magician was unholy.
Straddling this divide is the Oracle. The oracle has both elements of revelation and compulsion. Any decent oracle worth its salt would not give a straight-up answer, either, but rather required layers of revelation and interpretation. At Delphi, the God spoke to the Pythia, the priestess, who sat on the stool over the crack in the earth. When the god spoke, the fumes from below would overcome her, causing her to babble and writhe uncontrollably. Priests would then ‘interpret’ the prophecy, in form of a riddle.
Why riddles? Riddles are ancient. They appear on cuneiform texts. Even Gollum knew what a true riddle should look like – a kind of lyric poem asking a question that guards the right answer in hints and wordplay.
‘I tremble at each breath of air/ And yet can heaviest burders bear. [implicit question being asked is who am I? – water]
We could not get away from a discussion of riddles in the digital humanities without of course mentioning the I-ching. It’s a collection of texts that, depending on dice throws, get combined and read in particular ways. Because this is essentially a number of yes-or-no answers, the book can be easily coded onto a computer or represented mechanically. In which case, it’s not really a ‘book’ at all, but a machine for producing riddles.
Ruth Wehlau writes, “Riddlers, like poets, imitate God by creating their own cosmos; they recreate through words, making familiar objects into something completely new, rearranging the parts of pieces of things to produce creatures with strange combinations of arms, legs, eyes and mouths. In this transformed world, a distorted mirror of the real world, the riddler is in control, but the reader has the ability to break the code and solve the mystery (wehlau 1997)
Riddles & divination are related, and are dangerous. But they also create a simulation, of how the world can come to be, of how it can be controlled.
One can almost see the impetus for necromancy, when living in a world described by riddles. Saul visits the Witch of Endor; Oddyseus goes straight to the source.
…and Professor Hix prefers the term ‘post mortem communications’. However you spin it, though, the element of compulsion, of speaking with the dead, marks it out as a transgression; necromancers and those who seek their aid never end well.
It remains true today, that those who practice simulation, are similarly held in dubious regard. If that was not the case, tongue in cheek articles titles such as this would not be necessary.
I am making the argument that modern computational simulation, especially in the humanities, is more akin to necromancy than it is to divination, for all of these reasons.
But it’s also the fact that we do our simulation through computation itself that marks this out as a kind of necromancy.
The history of the modern digital computer is tied up with the need to accurately simulate the yields of atomic bombs, of blast zones, and potential fallout, of death and war. Modern technoculture has its roots in the need to accurately model the outcome of nuclear war, an inversion of the age old problem space, ‘how can we keep the world from ending’ through the doctrines of mutually assured destruction.
The playfulness of those scientists, and the acceleration of hardware technology lead to video games, but that’s a talk for another day (and indeed, has been recently well treated by Rob MacDougall of Western University).
‘But wait! Are you implying that you can simulate humans just as you could individual bits of uranium and atoms, and so on, like the nuclear physicists?’ No, I’m not saying that, but it’s not for nothing that Isaac Asimov gave the world Hari Seldon & the idea of ‘psychohistory’ in the 1950s. As Wikipedia so ably puts it, “Psychohistory is a fictional science in Isaac Asimov’s Foundation universe which combines history, sociology, etc., and mathematical statistics to make general predictions about the future behavior of very large groups of people, such as the Galactic Empire.”
Even if you could do Seldon’s psychohistorical approach, it’s predicated on a population of an entire galaxy. One planetfull, or one empire-full, or one region-full, of people just isn’t enough. Remember, this is a talk on ‘practical’ necromancy, not science-fiction.
Well what about so-called ‘cliodynamics’? Cliodynamics looks for recurring patterns in aggregate statistics of human culture. It may well find such patterns, but it doesn’t really have anything to say about ‘why’ such patterns might emerge. Both psycohistory and cliodynamics are concerned with large aggregates of people. As an archaeologist, all I ever find are the traces of individuals, of individual decisions in the past. It always requires some sort of leap to jump from these individual traces to something larger like ‘the group’ or ‘the state’. A Roman aqueduct is, at base, still the result of many individual actions.
A practical necromancy therefore is a simulation of the individual.
There are many objections to simulation of human beings, rather than things like atoms, nuclear bombs, or the weather. Our simulations can only do what we program them to do. So they are only simulations of how we believe the world works (ah! Cosmology!). In some cases, like weather, our beliefs and reality match quite well, at least for a few days, and we know much about how the variables intersect. But, as complexity theory tells us, starting conditions strongly affect how things transpire. Therefore we forecast from multiple runs with slightly different starting conditions. That’s what a 10% chance of rain really means: We ran the simulation 100 times, and in 10 of them, rain emerged.
And humans are a whole lot more complex than the water cycle. In the case of humans, we don’t know all the variables; we don’t know how free will works; we don’t know how a given individual will react; we don’t understand how individuals and society influence each other. We do have theories though.
This isn’t a bug, it’s a feature. The direction of simulation is misplaced. We cannot really simulate the future, except in extremely circumscribed situations, such as pedestrian flow. So let us not simulate the future, as humanists. Let us create some zombies, and see how they interact. Let our zombies represent individuals in the past. Give these zombies rules for interacting that represent our best beliefs, our best stories, of how some aspect of the past worked. Let them interact. The resulting range of possible outcomes becomes a kind of probabilistic historiography. We end up with not just a story about the past, but also about other possible pasts that could have happened if our initial story we are telling about how individuals in the past acted is true, for a given value of true.
We create simulacra, zombies, empty husks representing past actors. We give them rules to be interpreted given local conditions. We set them in motion from various starting positions. We watch what emerges, and thus can sweep the entire behavior space, the entire realm of possible outcomes given this understanding. We map what did occur (as best as we understand it) against the predictions of the model. For the archaeologist, for the historian, the strength of agent based modeling is that it allows us to explore the unintended consequences inherent in the stories we tell about the past. This isn’t easy. But it can be done. And compared to actually raising the dead, it is indeed practical.
[and here begins part II, which runs through some of my published ABMS, what they do, why they do it. All of this has to fit within an hour, so I need to do some trimming.]
[Postscriptum, March 23: the image of the book of random digits came from Mark Sample's 'An Account of Randomness in Literary Computing, & was meant to remind me to talk about some of the things Mark brought up. As it happens, I didn't do that when I presented the other day, but you really should go read his post.]
I’m contributing to a volume on ‘Land and Natural Resources in the Roman World’. Below is my draft, on which I welcome comments and questions.
Towards the computational study of the Roman economy
Shawn Graham, Carleton University, Ottawa Canada
“Economies are complicated systems encompassing micro behaviours, interaction patterns, and global regularities. Whether partial or general in scope, studies of economic systems must consider how to handle difficult real-world aspects such as asymmetric information, imperfect competition, strategic interaction, collective learning, and the possibility of multiple equilibria. Recent advances in analytical and computational tools are permitting new approaches to the quantitative study of these aspects. One such approach is Agent-based Computational Economics (ACE), the computational study of economic processes modelled as dynamic systems of interacting agents.”
Most models of the Roman economy do not take into account this idea that micro-behaviours, feedback, and local interaction provide the circumstances out of which emerge those larger issues about which we are typically concerned. Before we can ask questions about growth, or market integration, or the degree to which Rome was ‘primitive’ versus ‘modern’, we have to focus on individual decision making. In this paper, I outline an agenda for how we might be able to do this. It comes down to this: we have to draw out and understand networks of individuals at all geographical scales and then use those networks as the substrate for computationally simulating individuals’ economic activities. Thus, archaeology and ancient literature become united through computation.
It is not necessary, I think, to rehash the historiography of studies of the Roman economy; the broad outlines of the debate are well known. What I find exciting though is the emergence of the New Institutional Economics of Douglass North. These works all draw attention to ideas around the consequences of individual decision making from incomplete knowledge, of how ‘good enough’ or satisficing decisions push actors and economics towards path-dependence (where because of past decisions one is locked into a particular mode). The idea that network relationships (and the institutions that emerge to promote these) are the mechanism through which ancient economies deal with incomplete knowledge is a powerful one because we can find and outline the traces of these networks through archaeology.
Bang takes this idea further and through cross-cultural comparison with Mughal India pushes our attention to the bazaar: “a stable and complex business environment characterised by uncertainty, unpredictability and local segmentation of markets’. Bang shows how the bazaar helped shield the individual from (while at the same time encouraging) fragmented information and instability. For Bang, one of the key mechanisms for exploring and understanding a bazaar-like economy is social networks, the way they form, and how information flows through them. We need to focus on the social differences between actors in a market situation.
Even the Emperor can be incorporated into such a perspective. In considering the Emperor as both the embodiment of the state and a private person intervening in its economy as one more player amongst other private citizens, Lo Cascio draws attention to the interplay between the Emperor’s euergetism and private markets, that he solved the problem of feeding Rome “by leaving unchanged the market mechanisms at work” and by “establishing and enforcing the rules of the game”. He does so to ensure that individuals neither engage in rampant speculation to force up prices, but also to ensure that prices do not become set to low, that is, to ensure an adequate profit. The success or failure of the Emperor to do this should depend on his position in the network.
This is a very complex view, too complex for any one individual to hold in one’s head and to be able to understand the non-linear outcomes of so many interacting parts. We need the computer. To understand the Roman economy, we need to simulate it. The best way to do this is to ground our simulation in what our evidence actually gives us: the actions of individuals in the past, whether we find that evidence in the archaeology or the ancient literature. We have to build up from individuals before we can begin to understand what ‘the Roman economy’ might actually mean.
Roman Economics Needs to be Networked
Networks and network analysis are currently in vogue in many areas of research. Tom Brughmans has neatly summarised the historiography and main issues surrounding networks and network analyses as applied to archaeological research. In essence, he argues that all of the varied approaches to network analyses are united by the idea that networks are ubiquitous and influence decision making; through networks both tangible objects and intangible influences spread (and in turn, promote or hinder network growth). Through studying networks, we are able to bridge the study of parts through reductionism to the emergent whole. The methodological advantages enumerated by Carl Knappett can be tied explicitly to the advantages that the New Institutional Economics bring. A network perspective forces one to consider relations between entities; they are spatial whether considered in social or physical terms; they work across scales; they can incorporate people, objects, and time.
Schortman and Ashmore make the argument that a by-product of social networks is the emergence of power, through collaborative or cooperative action. Debates around structure and agency in the social sciences revolve around questions related to the concentration of power and the generation of hierarchies across multiple spatial scales. It is from this contending for assets that politics becomes necessary and structures for the same emerge. 
Achieving power over others involves monopolizing some aspect of the production, distribution or use of materials. On the other hand, others can contest this by using their own social networks to gather resources (whether material or social). In this way, action through social networks results in political structures which are themselves the aggregate of flows of materials and ideas through social networks.
Perhaps the most well-known term in network studies is the idea of the ‘small-world’, first coined by Stanley Milgram. A small-world is not just a metaphor, but rather a precise concept in network terms, where a randomly connected network with mostly local connections has a few long distance connections which allow the entire network to be spanned in only a few steps. This is a crucial concept and one we should look for archaeologically or as a by-product of our models. Brughmans writes
“ … Such a specific topology has direct implications for the processes underlying it, like the transportation of materials, the spread of religious ideas or the enforcement of political power. These processes would largely take place between the highly connected nodes [actors of whatever kind] and they would only reach the larger number of less connected nodes through these [linkages]. In a small-world network, on the other hand, nodes within the same small-world are more often directly connected to each other, while only processes involving other small-worlds (e.g. long-distance trade) would go over the bridging nodes.”
How can we draw out networks from archaeological materials? Objects carry the resonances of who make them. They mark out membership and social identity. “Exchanging these items thereby manifests and extends crucial social linkages… Networks, therefore, come alive, in part, through the transfer of items that partake of their members’ social essences”. Objects therefore could be taken as proxies for social actors. Artefacts are the result of human, individual, decisions. They influenced other individuals at the time through their complex resonances of thing and place and object life-history. By considering artefacts and their relationships explicitly in terms of social network analysis, we reconnect with the individual in the past, and we obtain a perspective that allows us to see what kinds of actions were possible in the past, patterns of agency and structure, that those actors themselves could not see.
Anytime one can discern a relationship, a network is possible. Fiona Coward reminds us that the archaeological record is not a passive by-product but rather is in fact social relationships: “The patterning of material culture is a direct result of the social relationships between individuals and groups in which these objects were caught up. A network perspective provides a much more realistic picture, not only of objective sociality, but also potentially of individuals’ subjective experience of their worlds”.
But as Scott Weingart warns us, we also have to take into account the dangers of methodology appropriation. Network analysis comes to us from graph theory, from statistics, from computer science. The methods, philosophies and concerns of those disciplines are not necessarily congruent with archaeology. Brughman’s recent work will go a long way to addressing the potentials and perils of drawing out networks from archaeological materials.
If we can draw out networks from the ancient literature or the archaeological record, what then? We re-animate these networks using agent based (or ‘individual based’) modelling. This is a technique where a population of autonomous, heterogeneous ‘agents’ are created within a computerized environment. They are given rules of behaviour which they implement given a particular situation (whether when interacting with other agents, or with the environment). They are goal-oriented and to a degree, self-aware. That is, if we are interested in something called ‘the Roman economy’, we simulate the agents that compose that economy and their behaviours: not the economy itself (in contrast to traditional economic models with their equations and ‘rationalizing’ assumptions; we simulate at one level of complexity below our ‘target’.
Lea Tesfatsion discusses a number of ways in which ‘agent-based computational economics’ can be used, and classifies different studies according to their objectives. The one which should concern us here is what she calls, ‘qualitative insight and theory generation’: “how can economic systems be more fully understood through a systematic examination of their potential dynamical behaviours under alternatively specified initial conditions?” A key feature of agent modelling that differentiates it from other approaches is that once the starting conditions are specified, all subsequent events are driven by agent interactions. The researcher then is not so concerned with the final results of the simulation as its evolution over time, its history. Thus, the focus is on the process. That is to say, we study the model’s history to generate insight into real history.
Research into agent based models of economies has found that the topology of interactions matters. It is not just the pattern of social ties that matters, but also the environment in which these actions take place. We can explore the environmental aspect by setting the simulation on top of a cellular-automaton, a chess-board like arrangement of squares, where each square represents a unit of land and its holdings, and which responds to rules about growth, climate, geology, and so on. In many ways, a cellular automaton represents a dynamic geographic information system and which could obviously be drawn from archaeological GIS. Combined with an agent-based model representing the decision making agents, we then have a powerful tool.
A number of researchers have applied ABM and cellular automata to studying problems of structural change in agricultural societies, exploring everything from government intervention, to the diffusion of innovation in agriculture, to the beginning of European-style agriculture in Indiana in the early 1800s. These studies point to ways in which economic path-dependence emerges, under what circumstances innovation may diffuse, and decision making processes at different spatial scales.
Archaeology and Simulation
Simulation has a long history in archaeology. Agent based modelling is an outgrowth from several different fields, primarily game theory and complex systems studies. John Barret, in considering the relationship between agents and society, drew attention to how agents both form and are constrained by, the social structures that emerge from their interactions. He argued therefore that the level of ‘society’ should not form the basic unit of archaeological analysis, but rather the individual. It is social learning that creates a society (or an economy, for that matter).
There are now many agent-based models of past societies. Many of these are quite complicated, with many moving parts, which can make it difficult to understand what the model results may actually be telling us. I advocate instead for extremely simple models, exploring only a limited aspect of the phenomena in which we are interested in. After building a series of these, we can consider their results in aggregate.
Building a Simulation
How do we translate our arguments over the Roman economy into an agent framework? Tesfatsion suggests a four-step method for recasting that understanding in a way that can be modelled and explored:
“• Select as your benchmark case an equilibrium modelling of an economy from the economic literature that is clearly and completely presented and that addresses some issue you care about.
• Remove from this economic model every assumption that entails the external imposition of an equilibrium condition (e.g., market clearing assumptions, correct expectations assumptions, and so forth).
• Dynamically complete the economic model by the introduction of production, pricing, and trade processes driven solely by interactions among the agents actually residing within the model. These procurement processes should be both feasible for the agents to carry out under realistic information limitations and appropriate for the types of goods, services, and financial assets that the agents produce and exchange.
• Define an “equilibrium” for the resulting dynamically complete economic model.”
Tesfatsion remarks that when she tries this exercise with economics students, they find it difficult to understand the economy at this level working with individual agents. The challenge of this method is that it foregrounds survival: that the needs of subsistence, of surviving over time (death is always a possibility in these models) are the bedrock on which everything else is based.
Thus, we build our rule-sets that our computational agents will follow from our understanding of how an individual [Roman; collegium; military unit; family; city] acts in particular situations. While every agent might have the same suite of variables, each agent is heterogeneous. Its particular combination of variables is unique. We might all be playing basketball by the same rules, but my abilities are different than yours. We model the appropriate underlying environment. We specify the initial starting conditions, and then set the simulation in motion. We run the simulation over and over again to explore the complete behaviour space for all of the particular starting conditions, to see what emerges when and how.
To know that we have found something new about the Roman economy from such a process, I would suggest that we ground one of the model’s behaviours in social networks. There are various ways this might be accomplished. We might draw from an understanding of how patronage worked in the Roman world. Or, we could draw from Bang’s arguments about Rome-as-bazaar. Then, when the simulation has run its course, we can measure the emergent network and compare its features to real networks known from the archaeology. When the two correspond, then the model settings for that particular run have something important to tell us about ancient society. Alternatively, we could reverse that and specify networks found in the archaeology as our starting point. Do the model outcomes, when set from a starting point known archaeologically, make any sense according to what we believe to have been true about the Roman economy? If not, then perhaps our rule sets are flawed.
Doran underlines some of the main impediments to simulation building amongst archaeologists, limitations in computing power, and the disciplinary boundaries that create barriers to knowledge building. As Doran points out, the computing power issue is largely no longer an issue. Quite complicated simulations may be built on a desktop computer without too much trouble in terms of hardware. The second problem is more difficult. One way we can break those boundaries down may lie in the choice of simulation environment: the simpler and more intuitive the framework, the easier to use, develop, and communicate results. In my own work, I use the open-source Netlogo modelling environment, which I recommend to anyone interested in exploring the possibilities of this approach. While Netlogo has its genesis in efforts to educated school children in complex systems thinking, it is now in its fifth major release and is quite powerful. The learning curve is not overly steep, and much can be accomplished through tweaking the many models that come pre-packaged with the software (more on this below).
A simulation is an argument in computer code about the way the world works, and so represents a kind of ‘procedural rhetoric’. It is object-oriented, meaning that each individual behaviour exists as its own object. One then arranges the objects (which can be conditional) in the order the agents should carry them out, given a particular position in the environment or social position vis-à-vis other agents. One can have quite sophisticated simulations running after a day of working through the included tutorials. This should not be taken as a sign that Netlogo is simplistic. Quite powerful models have been built in it, including modelling the emergence of cities in the third millennium BC.
An excellent place to begin is with the included model, ‘Wealth Distribution’. In this model, a ‘world’ is simulated where grain is distributed randomly. In some places it grows thick on the ground; in other places it is sparse. A population of agents are introduced to this world. Their goal is to find enough grain to keep living, and to reproduce when conditions are right. The agents have ‘vision’ or knowledge of the world, to differing degrees. They have ‘metabolism’, or a preset amount of food they must consume with each time-step or they shall die (which again varies by individual). The amount of food collected above this metabolic rate becomes ‘wealth’. When this simulation is run, it becomes apparent that the differential distribution of resources in this world is sufficient in itself to create a partition of the world into classes where there are a few extremely wealthy individuals and a vast mass of others who are in constant danger of ‘dying’ from not having enough food.
We could then extend this model to represent something of the Roman world. We could give the agents a way of looking for help, of becoming a client of someone a bit wealthier than themselves. In return we could imagine that these ‘clients’ could offer support to their patron in turn. Perhaps the number of followers, and their relative ‘wealth’, could be translated into a score for ‘prestige’ which in turn affects the ‘patron’s’ ability to extract wealth from the world. What kind of artificial society results? In my ‘Patronworld’ model, which had its inspiration in the Wealth Distribution model, chains of connected individuals (that is, networks) do emerge from this dynamic, but they are fragile. One result seems to be that extremely high levels of gift-giving seem to go hand in hand with network collapse. It seems to do this by destabilizing the networks: too many people outside particular chains of patrons-clients are shut out of the system. Given the competitive building that characterizes the late Republic, this result is intriguing. It is a very simple model, to be sure, but one that foregrounds an important element of new models of the Roman economy: networks and social life. This kind of modelling also has the virtue that if one disagrees with the assumptions of the model, the code can be easily modified and adapted. In this way, model building is not an end point of research but rather a first step of a larger conversation (my own models may be found at the digital data repository Figshare.com and I welcome their use, adaptation, and improvement).
We need more and better networks drawn from historical and archaeological data. It is not enough, in contrast to Malkin, to use ‘networks’ as simple metaphor. Particular network topologies have different implications for the actors which make them up. If we are to make progress on the Roman economy, we need to explore the multiple networks of individuals and objects at multiple social and spatial scales. We need to turn our archaeological geographic information systems into computing environments in which agents can interact: at a stroke, we will have unified landscape archaeology, ancient history, and the study of ancient economics. Once we have this data, we can take our current understandings of the ancient economy, whether ’consumer city’, ‘primitive’; ‘modernising’, ‘bazaar’, NIE, or something else and translate them into an agent based simulation. If we can generate analogous networks to the ones we know archaeologically, then we might just have the wherewithal to argue that we have a model that tells us something useful, something new, about the past.
Agar, M. (2003). ‘My kingdom for a function: modelling misadventures of the innumerate’, Journal of Artificial Societies and Social Simulation 6.3. http://jasss.soc.surrey.ac.uk/6/3/8.html (accessed 28/05/2012).
Bang, P. (2006). ‘Imperial Bazaar: towards a comparative understanding of markets in the Roman Empire’, in P. Bang, M. Ikeguchi, H. Ziche (eds.), Ancient Economies, Modern Methodologies: Archaeology, Comparative History, Models and Institutions. Bari, 51-88.
Bang, P. (2009). ‘The ancient economy and New Institutional Economics’, Journal of Roman Studies 99: 194-206.
Barabási, A.-L. and R. Albert. (1999). ‘Emergence of scaling in random networks’, Science 268: 509-12.
Barrett, J. (2000). ‘A thesis on agency’, in M. Dobres and J. Robb (eds.), Agency in Archaeology. New York, 61-8.
Barrett, J. (2001). ‘Agency, the duality of structure, and the problem of the archaeological record’, in I. Hodder (ed.), Archaeological Theory Today. Cambridge, 141-64.
Bentley, R. and H. Maschner (2003). ‘Preface: considering complexity theory in archaeology’, in R. Bentley and H. Maschner (eds.), Complex Systems and Archaeology. Salt Lake City, 1-8.
Berger, T. (2001). ‘Agent-based spatial models applied to agriculture: a simulation tool for technology diffusion, resource use changes, and policy analysis’, Agricultural Economics 25: 245-260.
Bogost, I. (2007). Persuasive Games. Cambridge, MA.
Bourdieu, P. (1977). Outline of a theory of practice (trans. R. Nice). Cambridge.
Brughmans, T. (2010). ‘Connecting the dots: towards archaeological network analysis’, Oxford Journal of Archaeology 29.3: 277-303.
Brughmans, T. (2012). ‘Thinking through networks: a review of formal network methods in archaeology’, Journal of Archaeological Method and Theory 19.2 Online version: DOI: 10.1007/s10816-012-9133-8 http://www.springerlink.com/index/10.1007/s10816-012-9133-8 (accessed 28/05/12).
Buchanan, M. (2002). Nexus: Small Worlds and the Groundbreaking Science of Networks. New York.
Clarke, D. (1968). Analytical Archaeology. London.
Clarke, D. (1972). Models in Archaeology. London.
Costopoulos, A and M. Lake. (2010). (eds.), Simulating Change: Archaeology Into the Twenty-First Century. Salt Lake City.
Coward, F. (2010). ‘Small worlds, material culture and Near Eastern social networks’, Proceedings of the British Academy 158, 449-479. http://www.fcoward.co.uk/Cowardsmallworlds.pdf (accessed 28/05/12).
Davis, J. (1998). ‘Ancient economies: models and muddles’, in H. Parkins and C. Smith (eds.), Trade, Traders and the Ancient City. London, 225-56.
Davis, J. (2005) ‘Linear and non-linear flow models for ancient economies’, in J. Manning and I. Morris (eds.), The Ancient Economy: Evidence and Models. Stanford, 127-56.
Dean, J. S., Gumerman, G. J., Epstein, J. M., Axtell, R. L., Swedlund, A. C., Parker, M. T., and McCarroll, S. (2006) ‘Understanding Anasazi culture change through agent-based modeling.’ in Epstein, J. (ed.), Generative Social Science: Studies in Agent-Based Computational Modeling Princeton, 90–116.
Deffuant, G., S. Huet, J.P. Bousset, J. Henriot, G. Amon, G. Weisbuch. (2002). ‘Agent based simulation of organic farming conversion in Allier Département’, in M. Janssen (ed.), Complexity and Ecosystem Management: The Theory and Practice of Multi-Agent Systems. Cheltenham, 158–187.
DeLaine, J. (2002). ‘Building Activity in Ostia in the Second Century AD’ in C. Bruun and A. Gallina Zevi (eds.). Ostia e Portus Nelle Loro Relzaioni con Roma (Acta Instituti Romanae Finlandiae 27). Rome, 41-101.
Dibble, C. (2006). ‘Computational laboratories for spatial agent-based models’, in L. Tesfatsion and K. Judd (eds.), Handbook of Computational Economics, Vol. 2: Agent-Based Computational Economics. Amsterdam, 1511-1550.
Dobres, M. and J. Robb (2000). Agency in Archaeology. New York.
Doran, J. (2011). Review of A. Costopoulos and M. Lake, Simulating Change: Archaeology into the Twenty-First Century Salt Lake City. http://jasss.soc.surrey.ac.uk/14/4/reviews/2.html
Dornan, J. (2002). ‘Agency and archaeology: past, present, and future directions’, Journal of Archaeological Method and Theory 9: 303-29.
Evans, T.P., and H. Kelley. (2004). ‘Multiscale analysis of a household level agent-based model of land-cover change’, Journal of Environmental Management 72.1, 57–72.
Frier, B. and D. Kehoe. (2007). ‘Law and economic institutions’, in W. Scheidel, I. Morris, R. Saller (eds.), The Cambridge Economic History of the Greco-Roman World. Cambridge, 113-143.
Gardner, A. (2007). An archaeology of identity: soldiers and society in late Roman Britain. Walnut Creek.
Giddens, A. (1984). The constitution of society: outline of the theory of structuration. Berkeley.
Gilbert, N., and K. Troitzsch. (2005). Simulation for the Social Scientist. Second edition. Maidenhead, Berkshire.
Graham, S. (2006a). ‘Networks, Agent-Based Modeling, and the Antonine Itineraries’, The Journal of Mediterranean Archaeology 19.1: 45-64.
Graham, S. (2006b). Ex Figlinis: The Complex Dynamics of the Roman Brick Industry in the TiberValley during the 1st to 3rd Centuries AD. British Archaeological Reports, International Series 1486: Oxford.
Graham, S. (2009). ‘BehaviourSpace: Simulating Roman Social Life and Civil Violence’, Digital Studies/ Le champ numérique, 1(2). http://www.digitalstudies.org/ojs/index.php/digital_studies/article/view/172/214 (accessed 29/5/12)
Graham, S. and G. Ruffini. (2007). ‘Network Analysis and Greco-Roman Prosopography’, in K.S.B. Keats-Rohan (ed.), Prosopography Approaches and Applications: A Handbook. Oxford: 325-36.
Graham, S. and J. Steiner. (2008). ‘Travellersim: Growing Settlement Structures and Territories with Agent-Based Modelling’, in J. Clark and E. Hagemeister (eds.), Digital Discovery: Exploring New Frontiers in Human Heritage. CAA 2006. Computer Applications and Quantitative Methods in Archaeology. Proceedings of the 34th Conference, Fargo, United States, April 2006. Budapest: 57-67.
Hoffmann, M., H. Kelley, and T. Evans. (2002). ‘Simulating land-cover change in south-central Indiana: an agent-based model of deforestation and afforestation’, in M. Janssen (ed.), Complexity and Ecosystem Management: The Theory and Practice of Multi-Agent Systems. Cheltenham, 218–247.
Jansenn, M. and E. Ostrom (2006). ‘Governing Social-Ecological Systems’ in L. Tesfatsion and K. Judd (eds.), Handbook of Computational Economics, Vol. 2: Agent-Based Computational Economics. Amsterdam, 1466-1510.
Knappett, C. (2011). An archaeology of interaction. Network perspectives on material culture and society. Oxford.
Kohler, T., G. Gumerman, and R. Reynolds. (2005). ‘Simulating Ancient Societies’, Scientific American 293.1: 76-84.
Lehner, M. (2000). ‘Fractal house of pharaoh: ancient Egypt as a complex adaptive system, a trial formulation’ in Kohler, T. A. and Gumerman, G. J. (eds.), Dynamics in Human and Primate Societies: Agent-Based Modeling of Social and Spatial Processes . Oxford, 275–353.
Lo Cascio, E. (2006). ‘The role of the state in the Roman economy: making use of the New Institutional Economics’, in P. Bang, M. Ikeguchi, H. Ziche (eds.), Ancient Economies, Modern Methodologies: Archaeology, Comparative History, Models and Institutions. Bari, 215-236.
Malkin, I. (2011). A Small Greek World: Networks in the Ancient Mediterranean. Oxford.
Manning, J. And I. Morris. (2005). The Ancient Economy: Evidence and Models. Stanford.
Milgram, S. (1967). ‘The small world problem’, Psychology Today. 2:60–67.
North, D. (1990). Institutions, Institutional Change and Economic Performance. Cambridge.
Ourednik, A. and P. Dessemontet. (2007). ‘Interaction maximization and the observed distribution of urban populations: an agent based model of humanity’s metric condition’. http://ourednik.info/urbanization_mc (accessed 28/5/12).
Premo, L. S. (2006) Agent-based models as behavioral laboratories for evolutionary anthropological research. Arizona Anthropologist 17:91-113.
Scheidel, W. (2010). ‘Approaching the Roman economy’ version 1.0. http://www.princeton.edu/~pswpc/pdfs/scheidel/091007.pdf (accessed 28/5/12).
Scheidel, W. and S. von Reden (2002). The Ancient Economy. Edinburgh Readings on the Ancient World. New York.
Schortman, E. and W. Ashmore. (2012). ‘History, networks, and the quest for power: ancient political competition in the Lower Motagua Valley, Guatemala’, Journal of the Royal Anthropological Institute 18.1: 1-21.
Symons, S. and D. Raine. (2008) ‘Agent-Based Models Of Ancient Egypt’, in N. Strudwick (ed.), Proceedings of Informatique et Égyptologie. Piscataway, NJ. http://www.physics.le.ac.uk/ComplexSystems/papers/AgentBasedModelsEgypt2008.pdf (accessed 28/5/12).
Tesfatsion, L. (2006). ‘Agent-based computational economics: a constructive approach to economic theory’, in L. Tesfatsion and K. Judd (eds.), Handbook of Computational Economics, Vol. 2: Agent-Based Computational Economics. Amsterdam, 831-880.
Vriend, N. (2006). ‘ACE models of endogenous interactions’, in L. Tesfatsion and K. Judd (eds.), Handbook of Computational Economics, Vol. 2: Agent-Based Computational Economics. Amsterdam, 1047-1080.
Wallace-Hadrill, A. (1989) (ed.) Patronage in Ancient Society. London.
Watts, D. (1999). Small Worlds. The Dynamics of Network Between Order and Randomness Princeton.
Watts , D., and S. Strogatz. (1998). ‘Collective dynamics of ‘small-world’ networks’, Nature 393: 440-42.
Weingart, S. (2012). ‘Demystifying networks’, the scottbot irregular http://www.scottbot.net/HIAL/?p=6279 (accessed 28/5/12).
Wilensky, U., and M. Resnick. (1998). ‘Thinking in Levels: A Dynamic Systems Approach to Making Sense of the World’ Journal of Science Education and Technology 8.1: 3-18.
Wilensky, U. (1999). Netlogo . Evanston, IL. http://ccl.northwestern.edu/netlogo (accessed 28/5/12).
Wilhite, A. (2006). ‘Economic activity on fixed networks’, in L. Tesfatsion and K. Judd (eds.), Handbook of Computational Economics, Vol. 2: Agent-Based Computational Economics. Amsterdam, 1013-1046.
 Tesfatsion 2006: 832.
 Indeed, many economic models fail to recognize that the spatial context of economic action undermines many of the basic assumptions of economic theory, Dibble 2006: 1515.
 The discussion is usefully treated in for example Scheidel 2010, Manning and Morris 2005, and Scheidel and von Reden 2002.
 North 1990, and its application to the ancient world in Frier and Kehoe 2007; Bang 2009; the various papers in Manning and Morris 2005; Scheidel’s forthcoming Cambridge Companion to the Roman Economy
 Frier and Kehoe 2007.
 Frier and Kehoe 2007: 119.
 Bang 2006: 79.
 Bang 2006: 80-4.
 Bang 2008: 197.
 Lo Cascio 2006: 225.
 Lo Cascio 2006: 231.
 Graham 2006b: 111-113.
 cf. Davis’ diagrams in 2005; 1998.
 Brughmans 2012; 2010.
 Brughmans 2012; Bentley and Maschner 2003:1.
 Knappett 2011: 10.
 Schortman and Ashmore 2012: 3.
 Schortman and Ashmore 2012: 2-3 citing Barrett 2000, Bourdieu 1977, Dobres and Robb 2000; Dornan 2002; Garnder 2007; Giddens 1984.
 Schortman and Ashmore 2012: 3-4.
 Milgram 1967.
 Buchanan 2002; Watts and Strogatz 1998; Watts 1999.
 Brughmans 2010.
 Shortman and Ashmore 2012: 4.
 Graham and Ruffini 2007: 325-331.
 Coward 2010.
 Weingart 2012.
 Brughmans 2012; 2010.
 cf. Wilensky and Resnick 1998: 4; Agar 2003; Gilbert and Troitzsch 2005: 199-202. Gilbert and Troitzsch is an excellent resource for learning how to build simulations.
 Tesfatsion 2006: 840.
 Tesfatsion 2006:843.
 Watts and Strogatz 1998; Barabasi and Albert 1999.
 Janssen and Ostrom 2005: 1496 citing Dibble 2006; Wilhite 2006; Vriend 2006.
 Berger 2001.
 Deffuant et al., 2002.
 Hoffmann et al 2002; Evans and Kelley 2004.
 Some of the earliest work being in Clarke 1968, 1972; see Graham 2006a: 53-4.
 The premier journal is the Journal of Artificial Societies and Social Simulation, jasss.soc.surrey.ac.uk.
 Barrett 2001:155.
 for instance Lehner 2000; Kohler, Gumerman and Reynolds 2005; Dean et al 2006; Graham 2006a; Premo 2006; Graham and Steiner 2008; Symons and Raine 2008; Graham 2009; Costopoulos and Lake 2010.
 Tesfatsion 2006: 852-3.
 cf. Wallace-Hadrill 1989.
 Bang 2006; 2008
 Doran 2011.
 Wilensky 1999.
 Bogost, 2007: 28-44.
 Ourednik and Dessemontet 2007.
 Wilensky, 1998.
 Published in Graham 2009.
 Graham 2009: 11.
 cf. DeLaine 2002 on patronage in building projects at Ostia.
 Malkin 2011: 18-9.
These are the slides of the talk I gave at the Land and Natural Resources Conference last week at the Free University in Brussels. That talk was a bit more free form than the slides would suggest, so I’m not quite ready to share the written version (mostly because it’s still in a process of becoming…)
But, I think I’m fairly safe to share at least the opening bit….
Could the extractive economy of Rome (such as mining, logging and forestry) promote structural growth? What would be the archaeological signs of structural growth?
So much in the ancient world appears to rely on connectivity and mobility, either literally in terms of things like road-building (Laurence, 1999) or more abstractly, in terms of social connections, bonds of friendship, amicitia, and patronage. Horden and Purcell (2000) make the argument for multiple connectivities, both physical and social, that bound up the Mediterranean world. In my own work, I have argued for social connections becoming ‘real’ in the physical landscape, as a mechanism for the creation of notions of territory and landscape (Graham and Steiner 2008). Sometimes these social connections can be read from the archaeometry of material culture; other times from epigraphy; sometimes from history. In this paper, I want to explore how patronage intersects with natural resource extraction, and whether or not these intersections could promote structural growth. As Mokyer points out, “There is a qualitative difference between an economy in which GDP per capita grows at 1.5 percent and one in which it grows at 0.2 percent” (2005:286). Understanding which possible conditions could promote growth (and to what degree) therefore is a useful exercise.
I use an agent based modeling framework (ABM) for this exploration because I am interested not in simulating the past, but in understanding how different understandings about the past combine. ABM allows me to systematically test the ways these ideas combine and to generate a landscape of possibilities against which I may then lay archaeological or historical evidence.
Agent Based Modelling
Agent based modelling is an approach to simulation that focuses on the individual (indeed, it is sometimes known as individual based modeling). In an agent based model, the agents or individuals are autonomous computing objects – they are their own programme. They are allowed to interact within an environment (which frequently represents some real-world physical environment). Every agent has the same suite of variables (if it were a model of basketball, ever agent would have a ‘height’ variable, and an ‘ability’ variable), but each agent’s individual combination of variables is unique. Agents can be aware of each other and the state of the world (or their location within it), depending on the needs of the model. What is important to note, especially when we are interested in the past, is that we are not trying to simulate the past; rather, the model is a tool to simulate how we believe a particular phenomenon worked in the past (cf Gilbert and Troitzsch 2005:17 on the logic of simulation). When we simulate, we are interrogating our own understandings and beliefs. What is particularly valuable then is that we can build a model, and when the agents begin to interact along the patterns of behavior that we have specified (drawn from our understanding of how various processes worked), we have a way of exploring the non-linear, non-intuitive, emergent consequences of those beliefs. What’s more, in order to code a particular behavior, we have to be clear about how we think about that behavior. It forces us to make our assumptions explicit. A second investigator then can examine the code, critique these assumptions and biases (or indeed, errors) and modify the model towards a ‘better’ state. In this way, the model is both a laboratory and a crowdsourced argument about the past. In that spirit, I offer the code for this model at http://graeworks.net, and encourage the reader to download, adapt, critique and improve the argument. The model is built using the Netlogo modeling environment and language (Wilensky 1999). Simulations make their argument in computer processes, and like all forms of expression, they carry their own rhetoric which must be analysed (Bogost 2007).
Did Rome experience growth? If so, in what ways did that growth occur? What was ‘modern’ about the Roman economy? Why did not Rome make the leap that Europe did? If it is any consolation, historians of the Industrial Revolution are as puzzled by how it happened, as we are by why Rome’s didn’t. An important consideration though comes from Mokyr’s analysis of the intellectual foundations of the Industrial Revolution (2005). For Mokyer, it comes down to the idea not just that ‘useful’ knowledge available had increased, but that the social setting for this knowledge had expanded (2005: 287). For Mokyer, useful knowledge relates directly to the physical world, and how it works.
This is not a view that would’ve been foreign to the bien-pensants of antiquity. Columella, Varo, Pliny all set about to catalogue and categorize the world around them. The difference though is one of quality; an Enlightment description of a phenomenon attempted a degree of accuracy and thoroughness that was alien to the Roman mind.
For Mokyer, the easier it becomes for individuals to access that knowledge, the more likely technological change was to happen, thus resulting in sustained economic growth (Mokyer 2005: 295-6). Though Mokyer doesn’t say it explicitly, he is talking about how information passes through social networks. As Mokyer points out, this useful knowledge did not have to percolate down to the many (301). It simply had to reach those in a position to act on it (a figure he reckons to be at most a few tens of thousands in all of Europe, 301). In Roman terms, to talk about social networks of influence is to talk about patronage.
Once we have created a model that encodes our understanding of the phenomena under question, it remains to interpret the results. A framework for understanding our model of resource extraction in the Roman world is provided by the Canadian economic historian and media theorist Harold Innis (Innis was the mentor of Marshall McLuhan).
My contribution bears the provisional title – ‘Simulating Patronage & Resource Extraction: an agent-based Roman economic model’
“Starting with the idea that the Roman economy was socially and politically embedded in networks of patronage, this paper explores the ramifications of that understanding for natural resource extraction, using an agent-based model. Agent models employ hundreds of autonomous, individual software agents, interacting in a digital environment, according to the rules we specify. In this case, the rules are drawn from our understanding of how patronage worked in Roman society. The initial pattern of interactions is based on resource-extraction networks visible in the archaeological record. The environment is one in which the agents extract a scarce, yet renewable, resource (coppiced woodland). Under what circumstances is such a system sustainable? When - and how – can it break down? The patterning of results suggests a framework for understanding archaeological patterns of resource exploitation in the Roman world.”
Let me tell you today what happened when I introduced the idea of simulating the past to my first year students. The phrase ‘digital history’ does indeed appear in the title of the course, yet a significant proportion of the class told me during the first week that they were technophobes, that they could post an update to Facebook, but don’t ask them to do anything more complex! You can see how this is a wee bit of a challenge.
The solution though presents itself when we remember the benefits of simulation:
Benefit #1: Crystallization of one’s own thoughts on history. If I can’t express it, I can’t code it.
Benefit #2: To critique the simulation, one has to perform a close-reading of the code, a prime historical skill.
Benefit #3: Seeing where what did happen fits into the realm of then-possible outcomes; a way of assessing the impact of Caesar’s decision not to cross the Rubicon as it were.
I firmly believe that one of the most important benefits of learning to program is that it provides the discipline necessary to engage closely with texts, to take them apart to understand what the author, the programmer, intended to happen. I built my lesson around Benefit #2, but to ease them into the idea, I introduced them to Chapter 1 of Wheelock’s Latin Grammar.
I’m no Latinist. But even a glancing familiarity with the language (or with any language’s grammar, for that matter) helps to put the student into the correct mindset for dealing with code. Grammar forces us to look at syntax and construction, and how elements combine to create meaning.
The meaning of a word in Latin depends on its function in the sentence. That meaning is signaled by the combination of a root word and a stem. The verb ‘to praise’ has as its stem laud ; the first person in the present tense is indicated by the addition of -o : laudo I praise; I am praising; I do praise.
We spent a considerable portion of that first day working through Chapter 1 of the Latin grammar. The average Canadian student these days has no knowledge of how grammar actually works, so we had a bit of remedial work to do. The idea that word forms change with meaning, that the structure of a sentence – which words modify which others – can be deduced from the forms of the words even when you don’t know what the word actually means was a bit of an eyeopener.
And programming is rather like composing a sentence in Latin. In Netlogo-speak, you have ‘primitives’ that fit together in particular ways (like when in Latin you put a feminine noun with an adjective, the adjective takes a feminine ending to show agreement with the noun); you put these together into a single unit of meaning, called a procedure. In Latin and English, we’d call that a sentence.
We then looked at the ‘Termites model that comes bundled with Netlogo. I hid all of the code related to setting up the world and its displays, and had them concentrate on the main run-time procedures. I showed them how to recognize an ‘if…then’ type construction, and how to know where a procedure started and stopped. Then, I ask them to scan the code, and sketch out which procedures came first, any loops they encountered, to follow that through to the end of the model. They were to highlight any bits of code they didn’t understand.
Of course, at the end of the exercise, lots of the individual bits of code were highlighted, but as a class, they were able to develop the flow chart of how this simple model works. They’d begun the process of learning how to close-read the code of a simulation. And if you don’t know how the pieces fit together, how can you trust the results of the model? If you don’t know how the pieces fit, how can you know what it means?
Let me conclude with some remarks by Alun Salt on my last post:
[...]One of the benefits of computer models is that reading code requires close-reading which is a useful historical skill. Yes it is, but my gut reaction is why learn close-reading for history by examining code, when you could close-reading for history by examining historical texts? The gut is not noted for its large number of brain cells, and this example demonstrates why mine is no genius. Close-reading for code should be simpler. It should be unambiguous and lacking the complexities of meaning that words in historical text have. It’s an easier way of learning the skill that you can then take into more complex situations effectively making a shallower learning curve.
Having examined the way the code is written, my students are now looking at what the code produces, where it breaks down, and what emerges when you let thousands of individuals, interpreting the same instructions under different circumstances, interact. As one student said, ‘It’s kinda creepy’.
[Originally posted at Play the Past; reposted here for an archaeological readership]
My students – especially my first year students – sometimes wish for direct, first person testimony. Wouldn’t it make life easier if we could just interrogate them, read what they thought, directly? Seeing as how most of the people in question (in my classes) are Romans, this would require a wee bit of necromancy, which is illegal in many states. There are other ways to achieve a similar effect however. We’ll do it in silico.
Jeremiah McCall makes an argument in ‘The Unexamined Game‘ that,
Simulation games are models, and representations, of no particular value for deep learning unless they are reflected upon, dissected and analyzed [...] If analysis and critique, problem solving and skill building, are the most important goals, it matters less whether teachers use simulation games that are more or less accurate, so long as they provide the necessary support, structure, access to evidence, and guidance to allow students to critique simulations.
So to my mind, why don’t we just skip the gaming part altogether, and just build the simulation? Let’s toy with history, and see what we come up with. To do that, we’ll have to resurrect the dead. This presents certain logistical problems, but if we use an agent-based modeling perspective (and related ideas from complex systems theories) we can simulate what we believed to be true about processes in the past.
“Ah. So you’re not simulating the past, but rather how you think x worked in the past.”
That’s right. My preferred simulation toolkit is Netlogo, from the Center for Connected Learning at Northwestern University. Happily, it was created as a tool to help high school students explore and understand complex systems. This means that the learning curve – while steep in parts – is gentle enough that us poor digitally inclined humanities folks can begin to create powerful simulations and models.
Moreover, unlike most games, you can interrogate my code, my simulation. You can see my biases, assumptions, and mistakes. And you can add, delete, or modify the simulation to suit your own. Because the computer does not tolerate woolly programming, creating a model forces us to clarify our thinking in order to translate our understanding of the past into code.
What does the code of a simulation look like? My simulation exploring the way information diffuses through the Antonine Itineraries, the third-century lists of places one must travel through to get from town A to town Z, (1) is simplicity itself:
if anybody-here? with heard-it = false
“Anybody-here?” is a procedure for determining who else is in a particular location, built from smaller building blocks called ‘primitives’. ‘Heard-it’ is a variable of each agent, a yes or no. ‘Spread-the-news’ is a procedure to change the ‘heard-it’ variable from ‘no’ to ‘yes’. ‘Go’ is a procedure that each agent runs continuously, once the simulation is started by the user.
Each agent in the simulation moves along the paths suggested by the Antonine Itineraries. When an agent encounters another agent, they check to see if the other agent has heard the news. If they haven’t, they change their internal state to ‘yes, I’ve now heard the news’ and toddle off on their way. It’s a very simple model, but when you compare the time it takes for the message to spread to every agent in the simulation, it becomes apparent that the key variable is the shape of the network that they are interacting on: hence the simulation suggests that the way that space is conceptualized in the Itineraries in the various provinces has an impact on the process of Romanizaton there.
Simulations of course can become much more complex. In another simulation (2), I look at a collection of digital Romans as they play the games of patronage with one another, in different economic environments. Under what circumstances could violence emerge? (By asking that question, I’m implying a necessary connection between Roman patronage systems and civil violence. Don’t like the way I envision that relationship? Change the code!)
When I run the simulation, I end up mapping out all the landscape of all possible outcomes, given that understanding of the past. In which case, I then look at what did happen… and see where it fits on that landscape.
The Benefits of Simulation
Benefit #1: Crystallization of my own thoughts on history. If I can’t express it, I can’t code it.
Benefit #2: To critique the simulation, one has to perform a close-reading of the code, a prime historical skill.
Benefit #3: Seeing where what did happen fits into the realm of then-possible outcomes; a way of assessing the impact of Caesar’s decision not to cross the Rubicon as it were. (A kind of Schrodinger’s cat approach to history, I suppose!)
(Jeremiah has more thoughts on the benefits of simulation for historians and educators here. Read this now!)
What Happens Next
There are three tutorials that come packaged with Netlogo. In the next few weeks, I will be taking my first year Digital History students through these tutorials. Then we’ll craft a simple model of some historical process (I’m leaning towards something documented in Wikipedia: they’ll take what the ‘consensus’ view is, as crowdsourced, and explode it using the model). I will document here on Play the Past my lessons, what works, what didn’t work, and thoughts on making it all better. Stay tuned!
*Well, maybe that’s the wrong word. After all, we’re not raising the dead to divine the future, more like creating artificial life to divine our thoughts the past… cybermancy? Frankenmancy? Pensieve-mancy?
(1) 2006 S. Graham ‘Networks, Agent-Based Modeling, and the Antonine Itineraries’. In The Journal of Mediterranean Archaeology 19.1: 45-64
(2)2009 S. Graham “Behaviour Space: Simulating Roman Social Life and Civil Violence.” Digital Studies / Le Champ NuméRique, 1(2). Retrieved January 6, 2011, from http://www.digitalstudies.org/ojs/index.php/digital_studies/article/view/172/214
Leave a Reply
Here’s what I want to do; I use Netlogo.
I’d like to build a model that focusses attention on the ways resource extraction in peripheral areas could drive development (however we define that) in those areas. I’m looking at timber exploitation in the Roman world, for which there is very little archaeological or literary evidence.
So the problem is one of building a model that bridges that whole lack-of-evidence gap. I’ve got my network models of brick exploitation, which could provide a starting point. Instead of one all-encompassing model for all of this, I’m thinking of using small models that can be easily explained/queried… I’m thinking using a basic Wealth Distribution/Sugarscape model overlaid on two different substrates: my brick social networks, and then on GIS data related to the general sources of timber exploited in the Roman world. The timber one would first use the Travellersim model James Steiner and I developed to generate a social network onto which we could then run the WD/Sscape model. Ideally what would then result could be mapped back against what archaeological/historical evidence there is, thus providing new insights etc etc.
That’s one approach.
Alternatively, I can use the GIS Gradient Example plus the Cooperation model from the models library as a starting point. I’ve got an ascii digital elevation model showing a roughly 200 x 200 km square of central Italy. While I don’t know the locations of actual production sites for logging, etc, I think I could work the following:
- patches at a certain elevation range have a probability of coppiced woodland
-patches at another elevation range prob of wild woodland
- these grow at different rates
-Romans harvest these using different strategies (cooperation, greedy strategies from the models library cooperation model; other economic strategies?)
-uphill travel increases metabolism, downhill travel doesn’t increase metabolism
-use the output graphs from the wealth distribution model
So I’d then want to explore the model to see balance of settings are stable given different environments (mostly coppiced; mostly wild woodland; greedy/cooperative)
A second phase of the model might be to add some sort of transportation from the harvest site to the nearest river, a pickup/putdown routine that effects metabolism, so that we then get an idea of how geography intersects with the different environments/strategies.
In 2005, I was the first Post-doctoral Research Fellow in Roman Archaeology at the University of Manitoba. In that research, I was exploring the use of Agent Based Modeling methods for understanding various questions of Roman Archaeology. I’ve since moved to Carleton University. My models are now housed on some space I’ve got, and can be accessed below:
The Antonine Itineraries & Information Diffusion Published in the Journal of Mediterranean Archaeology 19.1
PatronWorld Published in Digital Studies
Travellersim Published in the Proceedings of the 2006 Computer Applications in Archaeology conference
City Life – a toy, still under development, for exploring flow in ancient cities
You should be able to download the relevant nlogo files, too – but nb the U Manitoba email address no longer applies.