Simulating centurions (part 2)

Yesterday, I discussed an article written by Xavier Rubio and his colleagues about a simulation that tested the role of centurions in the Roman army. I also sent the authors some questions and they’ve spent a few days providing what I think are thoughtful and insightful answers. I reproduce the Q&A session below, highlighting some things that I think are particularly interesting or important. My thanks to Xavier Rubio, Pau Valdés Matías, and Eduard Ble for taking the time to answer my queries. Things that I think are particularly noteworthy I’ve highlighted below in bold.

Questions and answers

Josho Brouwers: First of all, thanks for taking the time to answer some questions. As you can see from my earlier discussion, I have some issues with the model that you adopted for your computer simulation. Let’s deal first with individual experience. We’ll ignore the problem regarding sources for reconstructing battle experience. Given the scale at which the simulation is run, what kind of individual behaviour was coded into the different agents?

Xavier Rubio: We really appreciate the opportunity to discuss this work. The idea behind this research is to show that computer simulation can be a useful tool to explore the past. Almost all scientific disciplines use some sort of formal models to test hypotheses, including historical disciplines such as archaeology or biology, so why would history be different?

The common approach for historians is to suggest descriptive models of Roman tactics based on the interpretation of written sources. This is a very important point: they are written in natural language, but they are a model anyway. In the paper we discuss the benefits of using a formal model instead, including analysis, transparency, and especially non-ambiguity. For example, in this case we explore the links between individual behaviour and large-scale dynamics. The agents’ behaviour is based on previous works, where psychological constraints are supposedly more relevant than physical stress.

In addition, we need to clarify that this a theory-building exercise: we don’t want to test the results against evidence because it would be rather difficult. Our aim is to tackle one concept and explore it using simulation as a sort of virtual laboratory. The results of this study allow us to identify new explanations and hypotheses that (hopefully) can be tested against data in other studies.

JB: One of the things I found particularly interesting was the way that you tried to simulate stress and fatigue in individual agents (pp. 253–255). Stress increases as agents move closer to the enemy and as their side sustains casualties. However, I’m not quite sure what you used as a basis for this information. Do we have any idea how stressful it was for Roman soldiers to fight? What do the numbers represent specifically?

XR: This is based on several published works, but at the same time it is common sense than any human will get nervous as he or she gets closer to the place where can be killed. Wouldn’t you get nervous if you see people getting killed around you? 

The common approach to model this situation is to establish a threshold where the agent loses the will to fight and it just wants to be as far as possible from the danger. This is based on psychological studies conducted on current extreme situations such as the emergence of panic during evacuations. In our case, the stress model is based on transmission. It means that the stress of each soldier is averaged by the stress of the companions he can see (in front of him, on his sides). If he is surrounded by soldiers that want to flee from the battlefield, he will probably follow them.

Pau Valdés Matías: We don’t know for sure how Roman soldiers experienced combat. Even today, it is an ongoing debate and there are many studies concerning the will to fight in recent conflicts. Yet, we have many works that refer how important his surroundings were for a soldier. Many authors have highlighted how the small units or the contubernium had a very important role in the army. We can refer to the work done by MacMullen (1984), Culham (1989), or Daly (2006).

Also, we have many examples from classical authors how important the behaviour of nearby comrades was for soldiers. We have many examples how soldiers follow their comrades when they flee. Also, the death or even injury of important figures has a very important effect on the morale of the soldiers near them. For example, Hannibal was injured in the Siege of Arse and many soldier abandoned the siege of the city (Liv. 21.7.10), the Iberian soldiers of the Carthaginian army retreated when the first soldiers died at Illipa (Liv. XXVIII, 15, 9), the defeat of Philip V at the Battle of Cynoscephalae in 197 BC follows the same scheme where multiples parts of the army start retreating and then the line collapses. We see the same during the Histrian Campaign (Liv. XLI, 18, 11-12), when soldiers hide the dead body of the consul to avoid loss of morale.

JB: You programmed the individual soldiers to cast javelins (pila) at their enemies when they were within their predefined kill zones (p. 255). Maybe I missed this, but did the model limit the number of javelins that a soldier could throw?

XR: Yes, each agent has one pilum that can be thrown at any moment it is within the kill zone.

JB: Something that isn’t clear to me regards the way that casualties are treated in your model. On page 255, you and your colleagues write that if the damage sustained by an agent is less than lethality, the agent is regarded as a casualty. I assume the agent is then removed from the battle? Am I also correct in assuming the operational state of an agent is essentially binary (active/no longer active)?

XR: It does not work exactly this way. Every time an agent is in combat it has a small chance of getting disabled. We chose to simplify this because it was not relevant to our research questions. This is a crucial point: in science any model is designed to answer a particular research question. You cannot create a model to explain everything, because then it would be too complicated to perform a proper analysis. In this case, our questions were related to the role of the centurion, so the relevant concept here is how stress and casualties decrease the combat capabilities of a formation. 

JB: Am I right in believing that the simulation does not take into account what it was like to fight at the extreme left or right sides of the formation (pp. 255–256)? I imagine that battles would quickly devolve to some kind of general melee on those extreme edges, where there is more room to manoeuvre and soldiers from further down the ranks might feel compelled to move forward and support their comrades. The model does not simulate this? Is it possible to simulate something like that?

XR: Our model uses a technique called Agent-Based Model, where you define the individual behaviour and then see what type of macro-scale patterns can emerge from the interaction of the individuals. This is a nice case of emergence; we did not create any particular behaviour for the soldiers in the extreme sides, but we saw how the patterns you mention developed there. The agents in this zone perceived a lesser level of threat (as they ‘see’ less enemies), thus pushing forward more often than the rest of the line.

JB: One of the interesting observations from the model is that experienced centurions (values 1.8 to 2.0) are able not only to maintain the formation of their troops but also to degrade that of their opponents (pp. 257–258). Their position also doesn’t seem to matter as long as they’re located in the first rank, but effectiveness decreases if they are placed somewhere else at random. Maybe I missed it, but how did you envision the centurion’s effectiveness? Did he have a voice range to influence nearby soldiers, or were commands passed down along the line?

XR: We felt that providing the centurion with ‘superpowers’ would be cheating, as it would force a circular reasoning (if you model him as more influencing than the rest of soldiers, then his role will obviously be stronger than the rest). The only difference between a centurion and another soldier is that they have slightly higher experience value, meaning that they will stay in the battlefield under stress for a longer period of time. If they are in the first rank then other soldiers will be influenced by them (as they are influenced by the rest of the first rank).

The interesting result here is that this simple change has an enormous impact on the efficiency of the whole formation, even if centurions were low in numbers. This is a nice example of non-linear dynamics: small changes in a complex system can have a large impact its large-scale behaviour. If this impact happens even without special ‘powers’, it is obvious that the impact of centurions would only increase with additional signals to increase their influence (commands, different helmet, reputation, etc.).

JB: One of the points that I raised in my earlier blog post was that a lot of these models and simulations seem to be based on circular reasoning. In your model, you test the effects that centurions had on the flow of battle, basing your ideas largely on ancient sources that claim, indeed, that centurions were effective. One of the conclusions you reach is that the sources are correct: centurions did have an important effect on the outcome of battles (p. 260–261). How would you parry such criticism?

XR: We already knew that centurions were important, as all the written sources and authors say that. Our real research question is: why? Is it because they controlled the formation? Or did they guide by example? Did this change over different periods? 

Our questions have been tackled before by several scholars (Connolly, Sabin, Goldsworthy, etc.), but this is the first time a formal model was created to explore this question. We avoid circular reasoning by modelling only the individual behaviour, and then exploring what large-scale dynamics emerge from the interaction between individuals. In the end the method is exactly the same as in previous work with an important difference: the fact that we use a formal model allows us to see how the different factors combine. For example, we can actually find under what conditions it is more relevant to have more ranks or to have more experienced soldiers.

JB: Your model relies very much of the work of Phil Sabin. Have you explored alternative interpretations of Roman battle? Would the results not be very different if, for example, your model doesn’t simulate battle as a sequence of repeated charges and retreats?

XR: As mentioned before, we only modelled the individual behaviour, following Goldsworthy’s hypotheses (which Sabin integrated in his model). The continued charge and retreat dynamics is generated by the interaction of all the soldiers, as we did not explicitly code this behaviour. We didn’t introduce any rule such as ‘the whole line will retreat after 5 minutes’, but something similar to ‘you increase your stress if someone dies in front of you’.

PV: For the paper we had to choose a model, so we preferred Sabin’s model. It is by far the most complex proposal so far, as it integrates both the psychological constraints defined by Goldsworthy and some intriguing ideas about the importance of throwing weapons. At the same time, it matches quite accurately with both written and archaeological evidence. It is no wonder it seems to be now the most accepted by scholars (Lendon and Quesada, among others, use it as basis for further studies). Anyway, this was our decision, but the interesting thing about formal models is that we could build a different proposal and then compare both of them.

JB: From what I remember from your presentation in Aberystwyth, your model does not take terrain into account, am I right? Considering this and the lack of other factors modelled (e.g. age and experience of individual soldiers, nature of the opposing armies), does the model not represent an ideal situation, in the sense that historical outcomes may have been completely different, or that factors not taken into account by your model might have been equally as important?

XR: Following Ockham’s razor, in science you need to find the simplest model able to answer the research question. It means that you discard everything that you feel is not relevant. You need to justify these decisions, and in the paper we discuss why the role of the centurions would not change in different types of terrain. If you don’t see the expected pattern then is the moment you need to reconsider your choice, because you are probably missing something relevant.

A quote often cited when doing models says that all models are wrong, but some are useful’. The good thing of a formal model compared to a classical descriptive model is that you can actually see where the model is wrong, discuss solutions, and improve it.

JB: Finally, what are you currently working on? Do you have any plans to continue work on simulating aspects of Roman warfare?

XR: We had to put our plans on hold because Pau and Edu were finishing their PhDs. Now we can continue discussing new ideas combining models, history, and archaeology. For example, at some point we would love to go to a large scale and explore the dynamics of diplomacy in the ancient world using game theory and similar techniques currently applied to contemporary conflicts.

On the other hand, we do other stuff beyond simulations. In particular, we are applying a quantitative framework to understand the evolution of military equipment based on archaeological evidence.

Thanks again for this Q&A. We would love to discuss these topics again on the next conference we meet!

About the authors

Xavier Rubio-Campillo is a postdoctoral researcher at the Barcelona Supercomputing Centre where he leads the Humanities group. The group applies new quantitative methods and theories such as Agent-Based Models, data science, or cultural evolution to improve understanding of the past. He often works in transdisciplinary environments involving archaeologists, historians, and other computer scientists, like the SimulPast project and EPNet, centred around the economic dynamics of the Roman Empire. One of his favourite research lines is focused on the evolution of warfare. You can check out his research on Academia.

Pau Valdés Matías is a PhD student at the University of Barcelona. His work is centred on the logistics of the Roman Republican army during the third and second centuries BC. His others lines of research include international relations in the ancient world and the roman army in the northeast of the Iberian peninsula. His research can be consulted on Academia.

Eduard Ble recently finished his PhD on Roman Republican weaponry and military equipment at the University of Barcelona. During the last few years, he has applied artefactual analysis to the study of Roman camps and sieges from an archaeological point of view, to some initiatives of experimental archaeology, or even, as on this occasion, the analysis of Roman combat tactics.

Leave a comment

Related Posts