Agent-Based Modeling in Art History: Simulating an Insane Asylum

This short paper reports on the development of a system that incorporates agent-based modeling (ABM) in art/architecture historical research and scholarship. ABM is a computational process simulating agents and their behaviors; the relationships between agents; and the interaction between agents and their environments. ABM has been used in a range of fields including predicting the spread of epidemics, behavior in economic systems, movement within the built environment, egress modeling (e.g. stadiums, submarines) and many more. (Macal, 2005; Lynne, 2015; Simeone, 2012 ) The development of this ABM system is supported by the Getty Foundation as part of the Advanced Topics in Digital Art History: 3D and (Geo)spatial Networks Summer Institute, Venice, 2018-2019.
In art/architectural history, ABM enables simulating inhabitants in addition to space itself and its formal qualities. Agents, programmed with basic rules or data to move autonomously within space, are modeled to recognize and “sense” their environment. This approach can expand both art historical questions and narratives by observing emergent movement in space and interactions between inhabitants. As a highly iterative computational process ABM allows for experimentation and new outcomes emerging from slight variations.
In our prototype of the Istanbul Toptasi Insane Asylum (functioned 1876-1924), we model medical and daily routines of the asylum inhabitants. (Ozludil, Wendell, 2016) This setting presents both a computational and philosophical challenge as ABM agents are typically assumed to be active, autonomous individuals with decision-making capability in a non-restricted environment. In contrast, the asylum is a highly-regulated environment with unpredictable (and arguably “irrational”) agents. Based on scattered evidence on life in the asylum and the scripted strict schedule, we model interactions between various agent types, as well as exposure of patients to natural light, air, and ventilation, all of which are measurable with the autonomous sensing of agents. The asylum presents a productive case study of applying ABM to art/architectural history as the movement of agents provides insight into the functioning of a nineteenth century imperial medical facility.

Objectives
The specific objective of the ABM in this case study is analyzing whether or not the scripted rules of daily life in the asylum could be implemented given the number and types of admitted patients in the year 1911. We iterate the process by adjusting the attributes and detailing the rules (where they are ambiguous) to produce various potential scenarios. While the objective here is limited to the simulation of the routines of patients, we do believe that the experiment provides some insight into the daily experience of patients, in the absence of textual evidence produced by them, such as letters, diaries. Our claim is not that this will be an all-encompassing understanding of the patient experience, but that it has the potential to open a window into these now-lost lives and, at a larger scale, to allow us to elaborate theoretical implications in question.

Method
Agents in the Toptasi Asylum simulation adhere to a strict set of rules. These rules allow the agents to behave autonomously over a long simulation time using only their initial programmed instructions. A central simulation clock is referenced for choreographing daily routines, instructing agents to find the shortest path to their next itinerary location using staircases and walkable surfaces. Below are three tables showing the attributes, rules, and the model timing that govern one agent type, the patient, chosen for this paper. (We used the framework presented in Axtell, Robert L., et al. 2002.)

Figure: Screen captures from the 3D model in SpatioScholar and historical photographs showing a male patient ward and the female patient courtyard. Left bottom image shows the ABM system within Unity3D. Sources: Right top: Ergin, Müessesat-ı Hayriye-yi Sıhhiye Müdiriyeti, 1911; right bottom: Mazhar Osman, Sıhhat Almanakı, 1933.

Toptasi Patient (Agent) Attributes

Each agent represents an asylum inhabitant that belongs to one
type
(In this case:
Patient. Others are Medical Officer, Medical Attendant, Administrative Officer, Administrative Attendant)

Each agent belongs to one
ward in the asylum
(Classification of the Toptasi Asylum into separate wards based on disease type)

Each agent belongs to one
sex that ties into the block and ward that they are admitted
(Female/Male)

Each agent is in one
state
(Classification of patients based on their recorded diagnosis)

Toptasi Patient (Agent) Simulation Rules

An agent starts the day, cleans self and makes their bed at 07:00
An agent proceeds to have breakfast in a dining hall at 08:00
An agent needs to be in their ward between 09:00-11:00 for daily medical visits
An agent proceeds to have lunch in a dining hall at 11:00
An agent has time dedicated to a mixture of airing, exercise, socializing, education etc. between 12:00-16:00
An agent proceeds to have dinner in a dining hall at 16:00
An agent proceeds to their ward for down time and eventually sleep at 17:00

Toptasi Agent Simulation Model Timing
A single clock increments each minute of simulated time. This clock broadcasts the current time to all agents for their internal itinerary movement. The simulation may run in either real time, where a minute within the simulation equates to a minute outside the simulation, or in a compressed timeline where the simulation clock calculates at a higher rate.

The Platform
The ABM system is built within the Unity3D application as a series of prefabricated objects and C# scripts. Agents are scripted objects that use the existing Unity3D codebase to path find and navigate within any imported 3D historical model. A natural language text file defines the attributes and daily itinerary of each agent cohort, allowing alternative and multiple agent setups to be quickly loaded from outside the simulation. An example itinerary line from this file, "1200": "Target DiningHall" defines a change in location by time and target object. he simulation agents continuously test a number of conditions using iterative and highly accurate raycasting methods: the proximity and visual connection to other agents, exposure to natural light and intervisibility to specific architectural and programmatic features. The outcome of these conditions are stored as data within each agent and are downloaded into a master external CSV file. The output CSV file makes the agents’ data available for data processing and visualization.

Bibliography
Axtell, Robert L., et al. "Population growth and collapse in a multiagent model of the Kayenta Anasazi in Long House Valley." Proceedings of the National Academy of Sciences 99.suppl 3 (2002): 7275-7279.
Crooks, Andrew, Christian Castle, and Michael Batty. "Key challenges in agent-based modelling for geo-spatial simulation." Computers, Environment and Urban Systems 32.6 (2008): 417-430
Hamill, Lynne, and Nigel Gilbert. Agent-based modelling in economics. John Wiley & Sons, 2015.
Indraprastha, Aswin, and Michihiko Shinozaki. "Constructing virtual urban environment using game technology." 26th eCAADe Conference: Architecture in Computro Antwerpen. 2008.
Lincoln, M. (2017, July 27). “Predicting the past: digital art history, modeling, and machine learning.” Retrieved June 11, 2018,
http://blogs.getty.edu/iris/predicting-the-past-digital-art-history-modeling-and-machine-learning/

Macal, Charles M., and Michael J. North. "Tutorial on agent-based modeling and simulation." Simulation conference, 2005 proceedings of the winter. IEEE, 2005.
Ozludil, Burcak, Augustus Wendell, and Ulysee Thompson, “Prototyping a Temporospatial Simulation Framework: Case of an Ottoman Insane Asylum,”
Complexity & Simplicity: Proceedings of eCAADe 2016, Vol. 2, 2016, pp. 485-491.

Simeone, Davide, et al. "An Event-Based Model to simulate human behaviour in built environments." Digital physicality: proceedings of the 30th eCAADe Conference. 2012.
Wurzer, Gabriel, Kerstin Kowarik, and Hans Reschreiter, eds. Agent-based modeling and simulation in archaeology. Vienna, Austria: Springer, 2015.