Methodology: Artificial Intelligence, Data Science, and Agent-based Modeling

What is an agent-based model (ABM)? An agent-based model is a computerized, digital model that aims to simulate a number of decision-makers (agents) and institutions, which change status and/or interact with one another over time according to a set of prescribed rules. By and large, all agents are often embedded in a hierarchical structure, dwell on and interact with an (often) dynamic landscape, and learn and adapt in response to changes in other agents and the environment. Building on a fundamental philosophy of methodological individualism, agent-based modelers advocate a focus on the uniqueness of individual agents, local environment, and multiple dynamical agent-agent or agent-environment interactions, which may account for many complexity features. Therefore, agent-based modelers warn against aggregating individual decisions or local-level characteristics so as to avoid potential misleading results.

Agent-based modeling is a major bottom-up tool powerful to understand complexity in many theoretical and empirical systems, to perform space-time analysis, and to represent or envision many complex landscape processes or human-environment interactions. Currently, the CHES group is pushing toward 4-D (x, y, z, and time) ABM (see this example), which involves stakeholders in an iterative process of model development for information sharing, collective learning and exchange of ideas on a given concrete issue among researchers and other stakeholders.

However, ABM suffers from a major challenge in relation to in-depth understanding of mechanistic processes: to reconcile the generative power of truly bottom-up, mechanistic modeling frameworks and pattern-informed, empirical top-down approaches that reproduce macro-level patterns yet lack adequate explanatory power. Other ABM challenges include developing integrated human-environment ABMs, modeling human behavior, module reusability and transparency, model verification and validation , high-performance computing, and building spatially explicit ABMs (particularly considering the “telecoupling” effects; Liu et al. 2013). All these challenges arise from ABM’s greater complexity (compared to traditional models), a price that must be paid for ABM’s high flexibility and capacity to capture complex systems dynamics.

Traditional artificial intelligence (AI) uses machines to simulate human intelligence, implying a solution for most of the above ABM challenges (An et al. in review). Yet AI’s capability to nourish complex systems science hinges upon the availability of data in high volumes. Advances in data availability (new forms of data such as big data, ethnography input, and social media texts) and data science have yielded many methods, programming tools, and appropriate data infrastructures. This advantage boosts AI’s power to understand human intelligence and simulate how agents perceive, act, and react to other agents and/or changes in the environment(s) around them. Through “training”, machine learning (one very promising AI method) can help derive model structures that verify or rebut the underlying mechanisms, forces, and/or processes behind macro-patterns. Recently, machine learning has advanced so dramatically to not only offer “black-box” predictions, but also to recover mechanisms behind observed data. In a successful instance, a graph neural network model has been trained to derive the closed-form, symbolic expression of Newton’s law of motion based on the mass, charge, geographic positioning information, and so on of all particles in the experiment (Cranmer et al. 2020). Simply put, the data-mining approach was able to ultimately produce a learned mathematical function that exactly “recovers” Newton’s formula F=G(m_1 m_2)/r^2 without any previous clue or assumption regarding its form (See the figure below). This suggests AI’s major potential to recover laws or mechanisms in many domains.

Figure 1. Derivation of the Newtonian law of gravitational force based on data on particles (represented as circles of different colors) over time using a machine learning approach. F, G, m1, m2, and r represent the force between Particles 1 and 2, gravitational constant, the mass of Particle 1, the mass of Particle 2, and the distance between the two particles. The double arrows represent forces between particles. GNN represents graph neural network (Cranmer et al. 2020).

Below is a list of references and exemplar articles that help illustrate the uniqueness and power of AI, DS, and ABM.

Readings and References:

An, L., V. Grimm, A. Sullivan, B.L. Turner II., N. Malleson, A. Heppenstall, C. Vincenot, D. Robinson, X. Ye, J. Liu, E. Lindvist, and W. Tang (2023). Modeling agent decision and behavior in the light of data science and artificial intelligence. Environmental Modeling & Software, 105713, https://doi.org/10.1016/j.envsoft.2023.105713.

An, L., V. Grimm, A. Sullivan*, B.L. Turner II., N. Malleson, A. Heppenstall, C. Vincenot, D. Robinson, X. Ye, J. Liu, E. Lindvist, and W. Tang (2021). Challenges, tasks, and opportunities in modeling agent-based complex systems. Ecological Modeling 457: 109685; doi: 10.1016/j.ecolmodel.2021.109685.

An, L. V. Grimm, A. Sullivan, B. L. Turner II, Z. Wang, N. Malleson, R. Huang, A. Heppenstall, C. Vincenot, D. Robinson, X. Ye, J. Liu, E. Lindvist, and W. Tang (2021). Agent-based complex systems and agent-based modeling (working paper).

An, L., A. Zvoleff, J. Liu, and W. Axinn (2014). Agent based modeling in coupled human and natural systems (CHANS): Lessons from a comparative analysis. Annals of Association of American Geographers 104(4):723-745.

An, L. (2012). Modeling human decisions in coupled human and natural systems: review of agent-based models. Ecological Modelling 229(24):25-36.

An, L., M. Linderman, J. Qi, A. Shortridge, and J. Liu (2005). Exploring complexity in a human-environment system: An agent-based spatial model for multidisciplinary and multi-scale integration. Annals of Association of American Geographers 95(1):54-79.

Chin, A., L. An, J. Florsheim, L. Laurencio, R. Marston, A. Parker, G. Simon, and E. Wohl (2016). Feedbacks in human-landscape systems: lessons from the Waldo Canyon Fire of Colorado, USA. Geomorphology 252(2016): 40-50.

Cranmer, M., A. Sanchez-Gonzalez, P. Battaglia, R. Xu, K. Cranmer, D. Spergel, and S. Ho (2020). Discovering symbolic models from deep learning with inductive biases. arXiv:2006.11287 [cs.LG]. https://arxiv.org/abs/2006.11287.

Epstein J.M., and R. Axtell (1996). Growing Artificial Societies: Social Science From the Bottom Up. Washington: Brookings Institute.

Epstein J.M. (2006). Generative Social Science: Studies in Agent-Based Computational Modeling. Princeton, New Jersey: Princeton University Press.

Liu, J., V. Hull, M. Batistella, R. DeFries, T. Dietz, F. Fu, T. W. Hertel, R. C. Izaurralde, E. F. Lambin, S. Li, L. A. Martinelli, W. J. McConnell, E. F. Moran, R. Naylor, Z. Ouyang, K. R. Polenske, A. Reenberg, G. de Miranda Rocha, C. S. Simmons, P. H. Verburg, P. M. Vitousek, F. Zhang, and C. Zhu. 2013. Framing sustainability in a telecoupled world. Ecology and Society 18(2): 26.

Parker, D.C., S.M. Manson, M.A. Janssen, M.J. Hoffmann, and P. Deadman (2003). Multi-agent systems for the simulation of land-use and land-cover change: A review. Annals of the Association of American Geographers 93: 314–37.

Sullivan, A., A.M. York, and L. An (2018). Which perspective of institutional change best fits empirical data? An agent-based model comparison of rational choice and cultural diffusion in invasive plant management. Journal of Artificial Societies and Social Simulation 21(1):5.

Examples, Models, and/or Documents:

CHANS-ABM Submodels

Pseudo-code for CHANS ABM

Chitwan ABM

Introduction to Chitwan ABM- Presentation Sep. 27, 2013

River Geomorphology ABMs

GitHub - A Step-by-Step Guide