Simulating centurions (part 1)

My blog post on simulating ancient battles has received quite a lot of constructive feedback. Clearly, it is a topic of some interest among readers of Ancient Warfare magazine. Among others, Xavier Rubio-Campillo commented on the post and felt that I perhaps gave the work he and his colleagues – Pau Valdés Matías and Eduard Ble – had presented at the International Ancient Warfare Conference in 2014 short shrift. He sent me a copy of the article that the presentation was based on and he and his colleagues were happy to answer some of my questions, which I’ll post tomorrow.

Simulating centurions

You can read the abstract on the MIT Press Journals website or download the draft of the article as submitted to the journal on Academia. In the article, the authors note that centurions were ‘key figures in the Roman combat system’ over a long period of time (p. 245). Analysis of Roman warfare has generally been descriptive, based on ancient written texts and some archaeological data. These descriptive models typically can’t anwer why centurions were so important. Important questions, such as where were the centurions deployed or what kind of influence they had on the battle, cannot be answered based on descriptive models alone.

The solution, the authors argue, is to make use of a model (simulation) that regards the Roman army as a complex system, in which large-scale patterns can emerge from small interactions, or from the interactions between different individual agents. In other words, a model needs to be able to produce results that are more than the sum of its parts, and which is neither deterministic (with simple causal relationships between causes and effects) nor chaotic (i.e. entirely unpredictable).

This is not the place to delve into the details of complex systems theory, but a brief example might clarify the basic concepts. Take the pyramids. It’s impossible for a single human being to build a pyramid on his or her own. But put a lot of human beings together and they are able to achieve more than merely their sum total (what is referred to as emergent behaviour). Whereas a single human might be able to build a simple shack, a large number of humans are able, through cooperation (i.e. interaction) to erect massive monuments such as the pyramids. If human societies were nothing more than the sum of individual actions of human beings working in isolation, we’d never have built more than large numbers of fairly simple shacks.

A model based on complex systems theory focuses on the interaction of agents within the system (i.e. an agent-based model or ABM), bound by a set of rules, resulting in emergent behaviour. Simple forms of such models are used, for example, in predicting crowd behaviour (e.g. an architect might want to know how people would flow through a large public building in case of a fire, to make sure that the exits are in the right place and a large body of people can escape in time). In the model created by the authors, most agents are Roman legionaries, while some are defined as centurions. Each has a set of rules attached before they are let loose in a computer simulation.

But how are the rules determined in the first place? The authors explain how they arrived at their rules on pp. 248–250. The model used in the article is heavily based on Phil Sabin’s work on Roman battles (and as such, whether or not you agree with the basic assumptions of the simulation depends in large part on what you think of Sabin’s reconstruction; Sean Manning goes deeper into this point in his review of Lost Battles). The simulation created by the authors focused on three aspects in particular: individual experience, ranged combat, and the figure of the centurion.

Individual experience, in this case, is again informed by Keegan’s The Face of Battle (which I find dubious) and observations based on soldiers’ behaviour during World War II. They programmed the individual agents to have a certain tolerance for battle and death, and once a threshold was hit the agents would retreat. When the simulation was run, the result showed that battle consisted of a sequence of attacks (charges) and retreats, not evenly along the entire line, but distributed throughout. (Reading this, I was struck by its similarities to Homeric warfare.) Perhaps the simulation is correct, but in my opinion it’s a weakness of the model that it seems to rely on evidence that is distinctly not Roman, i.e. experiences of soldiers from World War II.

The authors side with a single author (Zhmodikov) in suggesting that, based on the written sources, the use of ranged weapons was crucial. They write, ‘His findings suggest that the majority of the casualties chronicled in the written sources were the result of pila or javelins hurled at a distance, not hand-to-hand combat’ (p. 249). This is interesting, but as it stands unconvincing. You can read Zhmodikov’s article online. If your Russian is a little rusty, you can do what I did and have Google translate it for you. As far as I know, the idea that most of the casualties in Roman battles were inflicted by ranged weapons is not widespread, but I’m happy to be corrected on this.

The third and final element is the figure of the centurion himself. Ancient sources differ as regards the nature and functions of the centurion on the battlefield. Here, I feel that the authors perhaps take the written sources too much at face value: yes, Caesar describes his centurions as capable leaders, but this fits – one needs only read J.E. Lendon’s Soldiers and Ghosts, cited also in the article – to realize that the Romans were keen to heroize their soldiers and especially their officers in a pseudo-Homeric manner. Caesar would have been unlikely to write that his centurions were okay at best, incompetent fools at worst, regardless of the truth.

In any event, the computer simulation based on Sabin’s model was fed into the computer as an ABM, in which a battle between two formations (Blue and Red) was simulated. Different aspects were tested: depth of formations (about which there is much disagreement), the positions of centurions on the battlefield (extreme right of the front rank, a random position in the front rank, or anywhere in a particular century), and the behaviour of the centurions (essentially, leadership skill). The results are certainly interesting, and I encourage you check out the original article.

Xavier Rubio explained to me that the simulation is particularly useful as an experiment to play around with different values and to try and see how something might have functioned. Like Phil Sabin (see his reaction to the previous blog post, near the bottom), he is not claiming that the model might prove something conclusively, which I think is an important aspect to remember. Similarly, the main difference between a proper simulation and a wargame is that in the former humans are only involved in the initial setup (after which the simulation basically runs itself based on pre-determined rules), whereas in the latter the human participants make active choices in how they deal with the situations that arise throughout the game.

Tomorrow, I will post the Q&A with Xavier Rubio and his two colleagues